An industry chain intelligent business management system based on finance and tax data

Abstract

The application relates to the field of supply chain management and discloses an industry chain intelligent business management system based on financial and tax data, which comprises the following steps: a data preprocessing module receives invoice metadata flow, removes abnormal data based on billing time sequence dispersion, and generates a composite material quality identifier by fusing multi-dimensional features; a topological modeling module constructs a time sequence signal based on the identifier, calculates a correlation coefficient, and establishes a proxy bill of materials mapping matrix; a feature calculation module extracts a baseline invoice circulation phase difference, and calculates a coherence accumulation index based on a red word offset time sequence signal when a dynamic circulation phase difference anomaly occurs; and a strategy execution module generates a horizontal scheduling instruction or a vertical blocking instruction according to the comparison result of the coherence accumulation index and a safety threshold. The application can accurately separate capacity constraints and hidden quality defects without the underlying production data, and realizes adaptive scheduling and risk isolation control of the cross-node supply chain.

Description

Technical Field

[0001] This invention relates to the field of supply chain management technology, specifically to an intelligent business management system for the industrial chain based on financial and tax data. Background Technology

[0002] With the development of collaborative manufacturing across the industrial chain, the demand for supply chain business management across enterprise nodes is increasing. Traditional business management systems typically rely on internal manufacturing execution systems and production scheduling plans to track and schedule material flow. However, in actual cross-enterprise supply chain networks, due to considerations of protecting trade secrets, strict data silos usually exist between enterprises at each node, making it difficult for the management system to directly obtain the underlying real bill of materials and equipment-level operating data. This objective environment prevents the system from transparently constructing the processing and transformation mapping relationship of upstream and downstream materials from a global perspective.

[0003] Without underlying production data support, when order lead times are extended or deliveries are delayed across the entire supply chain network, existing management methods often only reach the superficial stage of lead time monitoring and overdue warnings. Further analysis reveals that these delays typically have different physical causes. They could be due to simple equipment failures and order backlogs at upstream nodes causing capacity blockages, or they could be caused by hidden quality defects in raw materials, leading to returns and rework at downstream production lines, resulting in systemic delays. Due to a lack of deep feature mining and correlation analysis methods for cross-node common data (such as financial and tax invoice data), existing systems cannot decouple and separate the combined impact of limited capacity and the spread of quality defects at the data level. This results in the management system being unable to promptly and accurately locate the source node causing the anomaly when facing business fluctuations. It can only rely on manual experience for lengthy post-event investigations and offline coordination, unable to automatically output cross-node horizontal capacity allocation instructions or vertical risk blocking instructions based on objective data, making it difficult to achieve adaptive closed-loop control of business risks in the supply chain. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent business management system for the industrial chain based on financial and tax data. This system solves the problem that existing systems, when disconnected from the underlying production data of node enterprises, cannot effectively distinguish whether delivery delays in cross-enterprise supply chain networks are caused by limited production capacity or hidden quality defects, thus failing to achieve adaptive scheduling and risk isolation control of the entire supply chain.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a supply chain intelligent business management system based on financial and tax data, comprising:

[0006] The data preprocessing module is used to receive invoice metadata streams from external invoice data sources without accessing the node enterprise's MES, BOM list, production schedule, or equipment operation data. The invoice metadata streams include at least the invoice date, buyer identifier, seller identifier, tax classification code, specifications, transaction quantity, transaction amount, invoice type, and red ink reason code or negative sales reason field. Based on the node enterprise's invoice records, the module calculates the invoice time series dispersion and generates a composite material identifier by combining the specifications text and tax classification code.

[0007] The topology modeling module is used to construct a time series signal based on the composite material identifier and calculate the correlation coefficient to establish a proxy bill of materials mapping matrix;

[0008] The feature calculation module is used to perform time delay calculation on the time series signal associated with the agent bill of materials mapping matrix to extract the baseline invoice flow phase difference; calculate the dynamic flow phase difference in real time, and when the dynamic flow phase difference is greater than the sum of the baseline invoice flow phase difference and the preset tolerance threshold, confirm that the dynamic flow phase difference is abnormal. After confirming the abnormality, calculate the coherence cumulative index based on the red ink reversal time series signal constructed from quality-related red ink records after being filtered by the red ink reason code or negative sales reason field.

[0009] The strategy execution module is used to generate horizontal scheduling instructions or vertical blocking instructions based on the proxy bill of materials mapping matrix, according to the comparison result between the coherence accumulation index and the pre-loaded security threshold.

[0010] The above processing enables the system to establish a digital material conversion mapping by utilizing the statistical fluctuation patterns of external financial and tax data over time, without needing to obtain the enterprise's internal production schedule. It also enables the separation of characteristics of capacity constraints and quality defects by using external invoice time series signals, transforms macro business data into executable link reconstruction instructions, and achieves closed-loop control of the supply chain network status.

[0011] Furthermore, after calculating the invoicing time-series dispersion based on the invoicing records of the node enterprises, the data preprocessing module is also used for:

[0012] The total number of invoices issued daily by enterprises at statistical nodes within a set historical observation time window is used to generate a quantity sequence.

[0013] The ratio of the standard deviation to the expected value of the quantity sequence is calculated as the invoice issuance time series dispersion.

[0014] When the invoicing time-series dispersion is greater than the preset dispersion threshold, it is determined that the corresponding node enterprise has the characteristics of centralized invoicing or supplementary invoicing.

[0015] If the logistics receipt date, warehousing acceptance date, or contract fulfillment date of the corresponding node enterprise can be obtained, then the time series can be reconstructed using the logistics receipt date, warehousing acceptance date, or contract fulfillment date.

[0016] If an available auxiliary date cannot be obtained, the corresponding node enterprise will be added to the filter list, and the invoice date of the corresponding node enterprise will not be used in the calculation of the proxy bill of materials mapping matrix and the base invoice flow phase difference within the current calculation period.

[0017] The advantage of this design lies in the fact that in actual corporate financial operations, there is often a habit of rushing to issue invoices at the end of the month. This kind of digital pulse caused by non-real production activities can seriously pollute the data. This solution, by introducing a coefficient of variation model, can accurately identify and filter out such distorted fluctuations caused by financial settlement cycles, purifying time-series data at its source and eliminating interference for building a stable business baseline in the future.

[0018] Furthermore, the data preprocessing module combines the specification model text and tax classification code to generate a composite material identifier, specifically including:

[0019] Extract the specification and model field from the invoice metadata stream, clean and segment the specification and model field by deleting units of measurement, common words and invalid symbols, and convert the processed specification and model field into a high-dimensional feature vector;

[0020] The high-dimensional feature vectors are clustered using a density-based spatial clustering algorithm to generate micro-category labels. The clustering confidence is calculated based on the distance from the sample to the cluster center, the average similarity within the cluster, the number of samples under the same tax classification code, and the outlier ratio.

[0021] When the cluster confidence level meets the preset text confidence threshold, the tax classification code and micro-category identifier are combined to generate a formal composite material identifier; when the cluster confidence level does not meet the text confidence threshold, a temporary composite material identifier is generated, and its weight in subsequent mapping calculations is reduced or it is transferred to the manual confirmation queue.

[0022] This approach, which integrates macro-level coding with micro-level text, effectively compensates for the overly coarse granularity of a single tax classification code. In real-world supply chain mapping, tax codes alone cannot distinguish between materials of the same type but different parameters. Through the aforementioned clustering transformation, unstructured, free-form specification text is converted into machine-readable structured tags, enabling fine-grained material comparison across enterprise networks.

[0023] Furthermore, the topology modeling module constructs a time-series signal based on the composite material identifier and calculates the correlation coefficient to establish a proxy bill of materials mapping matrix, specifically including:

[0024] Based on the composite material identification, the input and output materials are extracted, and the invoice transaction quantities are summed and accumulated by day to construct the corresponding input time series signal and output time series signal respectively.

[0025] The correlation coefficient between the input time series signal and the output time series signal is obtained by using the Pearson correlation coefficient to calculate the correlation.

[0026] When the Pearson correlation coefficient is greater than the preset correlation threshold, and the sample size, composite material identification confidence level and time correction method of the corresponding material pair meet the proxy mapping entry conditions, the corresponding material pair will be written into the proxy material list mapping matrix.

[0027] The proxy bill of materials mapping matrix records at least the input material identifier, output material identifier, correlation coefficient, observation window, sample size, confidence level of composite material identifier, and time correction method.

[0028] Through this scheme, the system uses the degree of coordinated fluctuation of time-series signals at the external data level to reverse-engineer the physical material processing and transformation relationship of the target node, and determines the range boundary of the types of purchased raw materials and finished products involved in the node at the data level.

[0029] Furthermore, the feature calculation module performs time delay calculation on the time series signal associated with the agent bill of materials mapping matrix to extract the baseline invoice flow phase difference, specifically including:

[0030] Set a time delay step, calculate the cumulative sum of the dot product of the input time series signal and the output time series signal after adding the corresponding time delay step, and construct a cross-correlation sequence;

[0031] Within the set upper limit of summation, the time delay step is incrementally increased to perform an extreme value search, and the lag step that makes the cross-correlation sequence reach the global maximum value is determined as the reference invoice circulation phase difference.

[0032] Through the above technical solution, the system transforms the statistical lag of discrete financial and tax vouchers on the time axis into an equivalent physical circulation cycle of the target node for a specific material, and persistently stores it as a standard operating parameter for the monitoring process.

[0033] Furthermore, when the dynamic flow phase difference is greater than the sum of the baseline invoice flow phase difference and a preset tolerance threshold, the feature calculation module confirms that the dynamic flow phase difference is abnormal, specifically including:

[0034] Establish a real-time sliding time window and reuse the cross-correlation function to obtain the dynamic flow phase difference of the current business state within the window;

[0035] When the dynamic flow phase difference is greater than the sum of the base invoice flow phase difference and the system preset tolerance threshold, it is determined that the delivery extension condition is met, and the dynamic flow phase difference is confirmed to be abnormal.

[0036] Through the above technical solution, the system extracts dynamic status based on the real-time acquisition window, effectively filters reasonable small time deviations in daily industrial production activities using tolerance thresholds, accurately locates abnormal lag phenomena in the physical flow process, and triggers decoupling procedures.

[0037] Furthermore, the feature calculation module calculates the coherence accumulation index based on the red-ink cancellation time series signal constructed from quality-related red-ink records after filtering by the red-ink reason code or negative sales reason field, specifically including:

[0038] Extract records where the invoice type indicator is red ink reversal or negative sales, and filter the reasons based on the red ink reason code, the negative sales reason field, and the available return order number, warehouse acceptance status, or contract performance record;

[0039] Records with reason fields pointing to price adjustments, sales allowances, invoicing errors, tax rate adjustments, contract changes, or accounting reversals will be removed or have their weight reduced. Records with reason fields pointing to returns, rejections, quality disputes, non-compliance with specifications, unqualified warehousing and acceptance, or those associated with return orders will be treated as quality-related red-ink records.

[0040] The input terminal, constructed from quality-related red-word records, and the output terminal, constructed from quality-related red-word records, are reconstructed according to the time step of the timestamp.

[0041] The phase difference of the reference invoice circulation is set as the time shift amount, and the input red ink cancellation timing signal is shifted backward to align the input red ink cancellation timing signal with the output red ink cancellation timing signal on the time axis.

[0042] The coherence accumulation index is obtained by multiplying the output red-word cancellation timing signal with the shifted input red-word cancellation timing signal.

[0043] In this way, the system mathematically superimposes the current finished product return event with the raw material return event from a baseline production cycle back, separating the interference caused by pure capacity constraints and accurately characterizing the degree of correlation of quality abnormalities along the supply chain.

[0044] Furthermore, the policy execution module is also used to calculate the pre-loaded security threshold, specifically:

[0045] Offline statistics were performed on the historical coherence cumulative index sequence of the target node enterprise within a set historical observation period in which there were no obvious quality anomalies.

[0046] Before offline statistics, exclude records with concentrated supplementary invoice issuance cycles, red-ink records where the reason field points to price adjustment or invoicing error, time windows with insufficient sample size, and material pairs with cluster confidence scores below the text confidence threshold.

[0047] Calculate the expected value and standard deviation of the historical coherence cumulative index sequence;

[0048] The mathematical expectation is added to the product of the confidence coefficient and the standard deviation to obtain and load the security threshold within the target node enterprise.

[0049] Based on statistical parameters of historical sample sets, the system defines the critical point at the computational level that distinguishes between fluctuations in normal business backgrounds and the spread of systemic quality defects, providing a quantitative basis for generating subsequent scheduling or blocking actions.

[0050] Furthermore, the strategy execution module generates horizontal scheduling instructions based on the proxy bill of materials mapping matrix, specifically including:

[0051] When the coherence cumulative index is less than or equal to the safety threshold, and the red text reason field does not point to centralized quality return, rejection, quality dispute, specification non-compliance or unqualified warehousing acceptance, the output material identifier that currently has phase anomaly is used as the retrieval condition to match parallel backup node enterprises that supply the same output material identifier in the agent material list mapping matrix.

[0052] Obtain the average daily number of invoices issued by each backup node enterprise within the historical window, and calculate and derive the corresponding purchase order allocation coefficient;

[0053] The horizontal scheduling instruction is generated based on the purchase order allocation coefficient to adjust the purchase allocation of the standby node.

[0054] Therefore, when the system confirms that the material delay is caused by pure capacity blockage, it can optimize multi-path resources based on the agent material network and reallocate purchase orders according to historical capacity weights, thereby eliminating the single-point transmission bottleneck of the current supply chain network.

[0055] Furthermore, the strategy execution module generates a vertical blocking instruction based on the proxy bill of materials mapping matrix, specifically including:

[0056] When the coherence accumulation index is greater than the security threshold, extract the set of timestamps whose amplitude is greater than the set peak threshold from the red text cancellation timing signal at the input end;

[0057] Reverse search the red-letter invoice records in the timestamp set, extract the seller identification number, composite material identifier, red-letter reason code, return order number and abnormal date, and mark the corresponding seller as a suspected quality source node;

[0058] The vertical blocking instruction is generated based on the suspected quality source node, the corresponding composite material identifier, the red reason code, the coherence cumulative index, and the auxiliary verification field.

[0059] The vertical blocking instruction triggers the freezing of procurement authority, stricter warehousing verification, quality inspection review or manual review based on the risk level, and the controlled object is limited to the corresponding composite material labeled materials supplied by the suspected quality source node.

[0060] When the red-ink cancellation signal shows a clear upstream and downstream correlation, the system further traces back the abnormal input materials and their supply nodes, and generates control instructions to suspend procurement or strengthen warehousing verification.

[0061] This invention provides an intelligent business management system for the industrial chain based on financial and tax data. It has the following beneficial effects:

[0062] 1. This invention extracts specification text and tax classification codes from invoice metadata streams, generates composite material identifiers using density clustering algorithms, and establishes a proxy material list mapping matrix by combining correlation calculations of time series signals. This design, even under conditions where internal production scheduling drawings and production execution system data of node enterprises are unavailable, can reverse-engineer the processing and transformation relationships of upstream and downstream materials solely based on the characteristics of external financial and tax documents, thus solving the technical deficiency of the difficulty in transparently extracting the underlying supply chain topology in cross-enterprise networks.

[0063] 2. This invention extracts the baseline invoice flow phase difference by constructing a cross-correlation sequence, and when the dynamic flow phase difference triggers anomaly detection, it performs translational alignment and dot product superposition calculations on the red-ink cancellation timing signals at the input and output ends. This technical feature transforms discrete invoice events such as returns into a continuous signal coherence measurement process, separating the mixed impact of simple capacity blockage and hidden quality defects on order delivery delays at the data model level, thus improving the accuracy of attributing supply chain anomalies.

[0064] 3. This invention establishes a safety threshold mechanism to perform conditional branching judgments on the coherence accumulation index. When it is confirmed that there is no quality defect propagation, parallel backup nodes are retrieved to generate horizontal scheduling instructions. When it is confirmed that the quality defect has propagated vertically, the timestamp is extracted in reverse to generate vertical blocking instructions. This mechanism directly converts the calculation results of multi-dimensional time-series signals into procurement adjustment actions that can be read by external enterprise resource planning systems, realizing adaptive capacity allocation and automatic isolation of risk sources across the supply chain in the physical flow process. Attached Figure Description

[0065] Figure 1 This is a system module architecture diagram of the present invention;

[0066] Figure 2This is a flowchart of the system operation method of the present invention;

[0067] Figure 3 This is a cross-correlation curve of the reference invoice circulation phase difference in this invention;

[0068] Figure 4 This is a histogram comparing the effects of the control scheme of the present invention. Detailed Implementation

[0069] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Please see the appendix Figure 1 This invention provides an intelligent business management system for the industrial chain based on financial and tax data, including a data preprocessing module, a topology modeling module, a feature calculation module, and a strategy execution module. The system can be deployed on a chain leader enterprise server, an industry collaboration platform server, or a distributed computing cluster.

[0071] The data preprocessing module receives invoice metadata from a specified industry chain range via an application programming interface (API). This metadata includes at least the invoice date, buyer, seller, tax classification code, specifications, quantity, invoice type, and a reason code for red-ink invoices or negative sales figures. The system does not require access to the node enterprise's MES, ERP production scheduling data, BOM list, equipment operation data, or work-in-process details. Where certain business systems are accessible, the system can also receive return order numbers, warehousing and acceptance status, contract fulfillment milestones, or logistics receipt dates as auxiliary verification fields; however, these fields are not necessary conditions for establishing agent material mapping relationships.

[0072] The strategy execution module communicates with external enterprise resource planning systems, procurement management systems, or supplier collaboration platforms to output business control instructions such as adjusting procurement allocation, tightening warehousing verification, freezing suspected problematic materials, and requiring manual supplier review. These control instructions can be configured to execute automatically, require manual confirmation, or only generate warning prompts, depending on the risk level.

[0073] Please see the appendix Figure 2 The specific operating flow of the modules included in the above system includes the following steps:

[0074] S10, Receive invoice metadata stream and generate composite material identifier:

[0075] The data preprocessing module receives invoice metadata streams and extracts fields such as invoice date, buyer / seller identifier, tax classification code, specifications, quantity, amount, invoice type, and red-ink reason code or negative sales reason field. In cases where return orders, warehousing acceptance status, contract fulfillment milestones, or logistics receipt dates exist, these are loaded as auxiliary verification fields. The data preprocessing module calculates the invoicing time-series dispersion of node enterprises, removes node enterprise data with significantly distorted time distributions, cleans, clusters, and performs confidence level assessments on the specifications text, and combines the tax classification code with the micro-category identifier to generate a composite material identifier.

[0076] S20, Establish the proxy bill of materials mapping matrix:

[0077] The topology modeling module extracts the input and output material sets of the target node enterprise based on the composite material identifier, constructs a time series signal according to the invoice issuance time, calculates the correlation coefficient between the input and output time series signals, and filters material pairs with correlation coefficients greater than the correlation threshold to establish a proxy material list mapping matrix.

[0078] S30, Extract the phase difference in the circulation of the base invoice:

[0079] The feature calculation module applies the cross-correlation function to perform time delay calculation on the normal purchase and sales time series signals associated with the agent bill of materials mapping matrix, and finds the lag step that makes the cross-correlation sequence reach its maximum value, which is then determined as the baseline invoice circulation phase difference.

[0080] S40, Confirm the anomaly and calculate the coherence cumulative index:

[0081] The feature calculation module calculates the dynamic flow phase difference in real time. When the dynamic flow phase difference is greater than the sum of the baseline invoice flow phase difference and the tolerance threshold, the dynamic flow phase difference is confirmed to be abnormal. After the dynamic flow phase difference is abnormal, the feature calculation module first filters the red ink reversal records based on the red ink reason code, the negative sales item reason field, and the available return orders, warehousing acceptance status, or contract performance records. Non-quality-related red ink records such as price adjustments, discounts, invoicing errors, and contract changes are removed or have their weight reduced. Only records related to material abnormalities, such as returns, rejections, quality disputes, and specification discrepancies, are used to construct the input and output red ink reversal timing signals. Subsequently, the baseline invoice flow phase difference is set as a time shift amount to shift the input red ink reversal timing signal, and the coherence cumulative index between the output red ink reversal timing signal and the shifted input red ink reversal timing signal is calculated.

[0082] S50 generates horizontal scheduling instructions or vertical blocking instructions:

[0083] The strategy execution module receives the dynamic flow phase difference and coherence accumulation index. When the coherence accumulation index is less than or equal to the safety threshold, it generates a horizontal scheduling instruction to adjust the purchase order allocation coefficient of the standby node. When the coherence accumulation index is greater than the safety threshold, the strategy execution module combines the red text reason field, the abnormal date, the invoice record number and the auxiliary verification field to check the suspected abnormal source, locate the suspected source node, and generate vertical blocking, stricter warehouse entry verification or manual review instructions according to the risk level.

[0084] The technical details of each step of the above method will be explained in detail below, taking into account specific business scenarios.

[0085] The data preprocessing steps provided by this invention specifically include the following processes:

[0086] S101, the data preprocessing module receives invoice metadata within the target industry chain and organizes each record into a unified data structure. Before data entry, the system checks for missing key fields. If the invoice date, buyer / seller identifier, tax classification code, specification text, quantity, or invoice type is empty, the record will not participate in subsequent calculations. This tuple data includes timestamp, buyer identifier, seller identifier, tax classification code, specification text, transaction quantity, transaction amount, invoice type field, red ink reason code, or negative sales reason field. For return order numbers, warehousing acceptance status, contract fulfillment nodes, and logistics receipt dates that can be obtained from external business systems, electronic voucher platforms, or supplier collaboration platforms, the data preprocessing module records them as auxiliary fields, but does not use these auxiliary fields as necessary inputs for system operation.

[0087] This embodiment uses days as the smallest time granularity; in scenarios requiring higher frequency monitoring, hours or shifts can also be used as the time granularity.

[0088] In the above plural data, the buyer's identifier and the seller's identifier correspond to the unique tax identification number registered by the enterprise in the system. The tax classification code is a standard administrative classification code, used to determine the category of goods at a macro level. The specification text is the specification and model string recorded in the original invoice record, used to characterize the micro-engineering parameters of the goods. The transaction quantity record corresponds to the physical transfer value. The invoice type bit is a discrete state variable; when the transaction is executed normally, this state variable is set to 0; when a quality return or rejection occurs, because the tax system will issue a corresponding red-ink reversal voucher, this state variable is set to 1.

[0089] Since red-ink reversals can be caused by a variety of reasons such as returns, rejections, quality disputes, price adjustments, sales allowances, invoicing errors, or contract changes, the system does not directly identify records with the invoice type position of 1 as quality return records. Instead, it further classifies the reasons by combining the red-ink reason code, the negative sales item reason field, and auxiliary verification fields.

[0090] S102, The data preprocessing module extracts the total number of days for each node enterprise within the historical observation time window. Within the invoice records, the invoice issuance timeline dispersion is calculated. Since the invoice issuance time does not necessarily coincide with the logistics occurrence time, it is necessary to first identify obvious concentrated invoice issuance samples. Without processing these samples, subsequent phase difference calculations may reflect invoicing habits rather than the material flow rhythm. The coefficient of variation can quantify and identify such non-productive time-point distortions. The data preprocessing module counts the total number of invoice records issued daily by the node enterprise within the time window, generating a discrete quantity sequence arranged by day. The data preprocessing module calculates the coefficient of variation of the sequence as the invoicing time series dispersion. Its mathematical calculation model is as follows: ;

[0091] In the formula, Representative quantity sequence Total number of days in the historical observation time window within the standard deviation, This represents the mathematical expectation of the quantity sequence. To ensure the completeness of the algorithm's logic execution, the data preprocessing module performs a pre-judgment of the mathematical expectation. If the expectation is zero, it indicates that no business has occurred at the node within the current observation window, and the data preprocessing module directly adds the node to the silent list to stop subsequent calculations. If the expectation is zero, the coefficient of variation formula is calculated normally.

[0092] The data preprocessing module compares the calculated invoice time series dispersion with the system's preset dispersion threshold. In one embodiment, the dispersion threshold is set to a range of 0.8 to 1.5. In actual use, it can be calibrated according to the invoice distribution of the industry sample, the size of the enterprise, and the length of the observation window.

[0093] At that time, it was determined that the node enterprise had the characteristics of centralized invoicing or supplementary invoicing. If the node enterprise also has logistics receipt date, warehousing acceptance date or contract performance date, the system can use the above auxiliary dates to reconstruct the time series; if no auxiliary dates are available, the node enterprise is added to the filter list, and its invoice dates are not used in the proxy mapping and phase difference calculation in the current calculation period.

[0094] S103, the data preprocessing module performs text feature extraction on the remaining invoice metadata stream filtered by temporal discreteness. The data preprocessing module extracts the same tax classification code. All specification texts below Fields. The data preprocessing module processes the specification text. After cleaning and removing units of measurement, generic words, and invalid symbols, word segmentation is performed. In this embodiment, TF-IDF can be used to convert the specification text into vectors, giving comparable weights to terms such as model number, packaging, size, and material in similarity calculations.

[0095] Subsequently, the data preprocessing module employs a density-based spatial clustering algorithm to cluster the high-dimensional feature vectors. During this clustering process, the system sets the cluster neighborhood radius parameter. The minimum number of points contained (MinPts) is used, and cosine similarity is employed as the distance metric between feature vectors. When the cosine distance between two high-dimensional feature vectors is less than a set neighborhood radius parameter... When these parameters are similar, they are grouped into the same density-reachable region. Clustering processes group specification texts describing similar physical parameters into the same cluster, generating corresponding micro-category identifiers. The system simultaneously records cluster confidence scores, which are determined by the distance from the sample to the cluster center, the average similarity within the cluster, the number of samples under the same tax classification code, and the outlier ratio. In practical applications, the system normalizes each clustering indicator to obtain a comprehensive score between 0 and 1, with the text confidence threshold typically set between 0.75 and 0.85. For specification texts with a comprehensive clustering score below this text confidence threshold, the system does not directly generate composite material labels but instead marks the sample as a pending confirmation sample. Pending confirmation samples may not participate in proxy mapping or may enter the manual confirmation queue or the enterprise material dictionary verification process. This step effectively converts unstructured free text into structured category labels that the system can further perform logical operations on.

[0096] S104, the data preprocessing module encodes and fuses macroscopic and microscopic attributes to generate a composite material identifier. The data preprocessing module also incorporates tax classification codes from the original invoice records. Micro-category identifiers generated by clustering Combined with text clustering confidence, a composite material identifier is established. For samples whose confidence level meets the threshold, the system combines the tax classification code with the micro-category identifier into a formal composite material identifier; for samples whose confidence level does not meet the threshold, the system generates a temporary composite material identifier and reduces its weight or suspends its use in subsequent mapping calculations. Its expression is: ;

[0097] In the formula, This represents a logical string concatenation operation; For micro-level category identification; Generate composite material identification codes from the original invoice records, using the tax classification codes. Subsequently, the data preprocessing module moves the statistical dimensions pointing to the physical material flow in the original invoice records down from the macro-level classification to the granularity of the composite material identifier.

[0098] Subsequent modeling will not directly use the company's internal material codes, but will instead use composite material identifiers. As a statistical granularity, this identifier is used to approximate the distinction between different materials and does not replace the company's official material code. Composite material identifiers are also used. It can effectively compensate for the data mapping ambiguity caused by the excessively wide granularity of a single tax classification at the data processing level, and meet the computational needs of different supply chain network nodes in fine-grained production feature comparison.

[0099] The steps for establishing the proxy bill of materials mapping matrix provided by this invention specifically include the following processes:

[0100] S201, the topology modeling module acquires the preprocessed invoice metadata stream and selects the target node enterprise within the network as the analysis object. When the buyer identification number in the invoice record matches the target node enterprise, that record corresponds to the target node's procurement input; the system aggregates the composite material identifiers in such records to form an input material set. .

[0101] When the seller identification number in the invoice record matches the target node enterprise, that record corresponds to the sales output of the target node; the system aggregates the composite material identifiers in such records to form an output material set. Through the above division, the system obtains the scope of purchased materials and the scope of sold materials observable by the target node enterprise under the invoice data caliber. This scope is not equivalent to the enterprise's internal BOM or formal material master data, but rather a proxy material observation set formed by the buyer and seller, tax classification codes, specification texts, and transaction quantities.

[0102] S202, Topology Modeling Module for Input Material Set Specific input materials and output material set Specific output materials The system filters and extracts data records where the invoice type indicator shows a normal circulation status to eliminate interference from abnormal return transactions on the regular production baseline modeling. Within a set historical observation time window, the topology modeling module sums the number of invoice transactions belonging to the same day, using days as the discrete time step.

[0103] For dates without corresponding transaction records, the system pads the quantity for that day with zeros to ensure alignment of different material sequences on the same timeline. The topology modeling module arranges the accumulated daily transaction totals in chronological order and constructs the input time series signals corresponding to specific input materials. and the output time series signal corresponding to a specific output material. .

[0104] S203, the topology modeling module uses statistical measures to calculate the degree of co-variance between the input and output time series signals. For materials with stable production relationships, procurement inputs and sales outputs are usually not completely synchronized, but they will show a certain correlation over a longer observation window.

[0105] Therefore, the system records highly correlated input-output combinations as candidate mapping relationships for subsequent phase difference calculations. The topology modeling module uses the Pearson correlation coefficient to calculate the co-oscillation relationship between the input and output sequences. For material pairs with low confidence levels in composite material identification, large corrections to invoice dates using auxiliary fields, or insufficient sample sizes, the system reduces their correlation score weight or excludes them from the proxy mapping table. Let the Pearson correlation coefficient for a specific material pair be... The mathematical model is:

[0106] In the formula, Represents the total number of days in the historical observation time window. This represents the specific number of days on the timeline. This represents the arithmetic mean of the input time series signal within the historical observation time window. This represents the arithmetic mean of the output time series signal within the historical observation time window; For a specific input material, the input time series signal is used. This refers to the output time series signal corresponding to a specific output material.

[0107] To ensure the closed-loop operation of the algorithm and prevent division-by-zero errors, the topology modeling module verifies the variance of the time series signal before calculating the Pearson correlation coefficient. If the number of transactions in the input or output time series signal remains constant within the time window, resulting in a variance of zero, the topology modeling module skips the correlation calculation for that specific material pair and records the correlation coefficient value as zero.

[0108] S204, the topology modeling module will calculate the obtained Pearson correlation coefficient. The correlation coefficient is compared numerically with a preset correlation threshold. This threshold can be determined based on historical samples of known stable supply relationships. In one embodiment, this threshold is set between 0.6 and 0.85; when the industry's materials are highly substitutable, this threshold can be appropriately increased to reduce mismatches. When the Pearson correlation coefficient... When the correlation threshold is greater than the threshold and the corresponding material pair meets the preset proxy mapping input conditions, the topology modeling module determines that there is a highly correlated material conversion correspondence between the specific input material and the specific output material.

[0109] The conditions for entering proxy mapping include: the number of valid transaction sample days within the time window is greater than the set sample size threshold (e.g., the number of valid transaction days is not less than 30 days), and the confidence level of the composite material identifier meets the aforementioned text confidence threshold requirement. For material pairs that exceed the threshold and meet the entry conditions, the system adds a new record to the proxy mapping table of the target node. The record includes the input material identifier, output material identifier, correlation coefficient, observation window, sample size, confidence level of the composite material identifier, time correction method, and whether auxiliary verification fields are used. The proxy mapping table is used for subsequent anomaly analysis and is not used as a replacement for the company's official BOM document. This technique of constructing a proxy bill of materials mapping matrix based on the time-series signal's coordinated fluctuation characteristics enables the system to form a set of calculable proxy material relationships based on the invoice sequence, even if it is not connected to the target company's MES or drawing database. This relationship is used for anomaly analysis and can be updated as new invoice data is entered.

[0110] The benchmark invoice circulation phase difference extraction step provided by this invention specifically includes the following processing steps:

[0111] S301, the feature calculation module obtains the proxy bill of materials mapping matrix established by the topology modeling module. The feature calculation module traverses the input and output material pairs with established strong correlations in this matrix. For each pair of input materials with a mapping relationship, the module identifies... Output material identification The feature calculation module extracts the input time series signal under the corresponding normal business conditions. With output time series signal .

[0112] For nodes with stable production, there is usually a certain time interval between purchase invoices and sales invoices. This interval is affected by processing cycles, inventory turnover, and invoicing habits.

[0113] In invoice data, the aforementioned circulation interval manifests as a phase misalignment between the input and output time series signals on the time axis. This phase difference reflects the observable circulation rhythm under the invoice data caliber, potentially encompassing factors such as processing cycles, inventory turnover, settlement habits, and invoicing delays. Therefore, the system uses it as an anomaly comparison benchmark, rather than directly equating it to the actual internal production rhythm of the enterprise.

[0114] S302, the feature calculation module applies a discrete-time cross-correlation function to calculate the time delay of the input and output time series signals, in order to quantitatively evaluate the similarity between the two signals under different time misalignments. The feature calculation module sets the time delay step size. And construct cross-correlation sequences Its mathematical calculation model is as follows: ;In the formula, The variable representing the specific number of days on the timeline. This represents the total number of days in the historical observation time window; For the constructed cross-correlation sequence; This is the output time series signal after adding a time delay step; Set the time delay step for feature calculation module 30; The input time series signal corresponding to a specific input material.

[0115] To prevent array out-of-bounds read errors at time series boundaries, the feature calculation module uses time variables... The summation limit is strictly limited to 1. . Not suitable for approaching Otherwise, there would be too few overlapping samples available for calculation. In this embodiment, No more than One-third. The feature calculation module increments the time delay step by step within this value range. And calculate the corresponding dot product summation value to generate a complete cross-correlation sequence. .

[0116] S303, The feature calculation module calculates the cross-correlation sequence generated by the cross-correlation module. The extreme value search operation is performed. The feature calculation module compares the cross-correlation sequence values ​​generated under different time delay steps and finds the specific lag step that makes the cross-correlation sequence reach the global maximum. The mathematical expression for this extreme value solution process is: ;

[0117] In the formula, This represents the value of the independent variable that maximizes the cross-correlation function; For the feature calculation module in the generated cross-correlation sequence; This is the time delay step.

[0118] If multiple delay steps correspond to the same maximum cross-correlation value, the system selects the smaller delay step as the maximum cross-correlation value. This is to avoid misjudging normal short cycles as long cycles.

[0119] Furthermore, if the calculated global maximum value is zero, or the maximum cross-correlation value is lower than the minimum phase confidence threshold, or the difference in cross-correlation values ​​corresponding to multiple delay steps is lower than the peak discrimination threshold, it indicates that the specific material pair does not have a stable and distinguishable cooperative hysteresis relationship on the time axis. The feature calculation module marks the material pair as unreliable in phase and does not use it in real-time anomaly monitoring. The baseline invoice flow phase difference for this material conversion chain is determined.

[0120] The technique described in the manual for extracting the phase difference of benchmark invoice circulation essentially transforms the statistical lag of discrete financial and tax documents on the time axis into an equivalent physical production circulation cycle representing the target node enterprise for a specific material. The feature calculation module stores this benchmark invoice circulation phase difference along with the corresponding observation window, cross-correlation peak value, peak discrimination, sample size, and time correction method. These parameters serve as benchmark parameters for subsequent real-time anomaly monitoring and delivery date widening comparisons. If the sample structure changes significantly within a subsequent window, the system can recalculate and update these benchmark parameters.

[0121] The real-time anomaly decoupling and coherence analysis steps provided by this invention specifically include the following processing steps:

[0122] S401, the feature calculation module has a length of [length missing]. A real-time sliding time window is used, which slides forward in days as the system calendar time increases. To ensure sufficient statistical samples are included within the window, the time window length is... The standard setting is the phase difference of the base invoice circulation. Two to three times the number of days; if this value results in insufficient sample size, then the minimum number of sample days threshold shall prevail.

[0123] The feature calculation module extracts normal-state purchase and sales data within a real-time sliding time window. Before performing cross-correlation calculations, the feature calculation module counts the number of valid transaction days within the window. If the number of valid transaction days is lower than the set minimum calculation sample size threshold, the calculation results from the previous period are maintained, meaning the system does not update. If the threshold is met, the dynamic flow phase difference is recalculated. .

[0124] The feature calculation module will dynamically transfer the phase difference. Phase difference with pre-persistently stored baseline invoices A comparison calculation is performed to determine whether the current material conversion chain meets the delivery time extension conditions. The trigger conditions for delivery time extension are set as follows: ;

[0125] In the formula, The dynamic transition phase difference of the target node enterprise's current business status; The basis for the invoice circulation phase difference; This represents the tolerance threshold. This tolerance threshold... This is used to absorb minor fluctuations such as normal invoicing delays, holiday shutdowns, and logistics delays. In specific implementation scenarios, the tolerance threshold... The phase difference in the flow of the baseline invoice is set based on different material attributes. Ten percent to twenty percent of the value.

[0126] when When this occurs, the system marks the material pair as having a phase anomaly and initiates subsequent return signal analysis. This marking only indicates that the current invoice processing speed has slowed down and does not directly identify the cause of the anomaly.

[0127] S402, after triggering the return signal analysis process, the feature calculation module extracts a data set spanning backwards from the current time point, and extracts invoice records with the invoice type field indicating red-ink reversal or negative sales from this data set. The system does not directly treat all red-ink reversal records as abnormal return records; instead, it first filters the reasons based on the red-ink reason code, the negative sales reason field, the return order number, the warehouse receipt status, and the contract performance record.

[0128] For red-ink records with reason fields indicating price adjustments, sales discounts, invoicing errors, tax rate adjustments, contract changes, or accounting reversals, the system removes them as non-quality-related red-ink records or includes them in the statistics with a preset low weight. For red-ink records with reason fields indicating returns, rejections, quality disputes, specification discrepancies, unqualified receiving inspections, or those associated with return orders, the system includes them as quality-related red-ink records in the construction of the input and output red-ink reversal timing signals. After completing the reason filtering, the feature calculation module arranges quality-related red-ink records by date and accumulates the daily quantity or transaction amount.

[0129] The system counts the number of red-word cancellations at the input and output ends daily, generating input-end red-word cancellation timing signals representing the intensity of upstream input anomalies. And the output terminal red-letter cancellation timing signal representing the abnormal intensity of downstream output. .

[0130] To ensure the continuity of time-domain operations, for dates within a set time span where no returns occurred, the feature calculation module uniformly assigns the corresponding signal amplitude to zero. This transforms scattered red-ink invoice records into two sequences on the same time axis, facilitating translation and dot product calculations.

[0131] S403, the feature calculation module calculates the phase difference of the base invoice flow. The input red-letter cancellation timing signal is shifted backward by a time shift parameter, aligning it with the output red-letter cancellation timing signal on the time axis. The feature calculation module then performs a dot product and superposition of the output red-letter cancellation timing signal and the shifted input red-letter cancellation timing signal to calculate the coherence accumulation index. Its mathematical calculation model is as follows:

[0132] ;In the formula, It is the coherence cumulative index; For the specific number of days on the time axis; This represents the current end time of the real-time sliding time window; The length of the real-time sliding time window; This is the timing signal for canceling the red text at the output terminal; This is the input timing signal for the red-letter cancellation after shifting the phase difference of the base invoice; The basis is the phase difference in the circulation of invoices.

[0133] The technique described in the instruction manual for extracting red-ink invoices and calculating a coherence cumulative index is used to separate defective returns from delivery delays caused by limited net production capacity at the external data level. This calculation focuses on the current defective return events of finished products. Going back one baseline production cycle on the timeline Raw material return incident Perform dot product aggregation.

[0134] If a return peak also appears at the input end near the reversed date, it indicates a temporal correspondence between output-end returns and input-end anomalies. (Coherence Cumulative Index) Used to indicate the degree of overlap between the input quality-related red peak and the output quality-related red peak under the reference phase difference constraint.

[0135] The higher the index, the greater the likelihood that a quality-related red-letter event will occur at the time corresponding to the mapping relationship of the proxy material. This index is not used as a standalone conclusion for quality incidents, but is used together with the red-letter cause, auxiliary verification fields, and safety thresholds as the basis for generating horizontal scheduling or vertical blocking instructions.

[0136] The control command generation and execution steps provided by this invention specifically include the following processes:

[0137] S501, the strategy execution module receives the delivery time extension determination result and coherence accumulation index output by the feature calculation module. The policy enforcement module preloads security thresholds into the system. It is used to define the boundary between normal business background fluctuations and the spread of systemic quality defects at the numerical calculation level.

[0138] This security threshold The safety threshold is obtained by offline statistical analysis of the historical coherence cumulative index sequence by the target node enterprise within a set historical observation period without significant quality anomalies. When analyzing the historical sequence, the system excludes records with concentrated supplementary invoice issuance periods, red-ink records with price adjustments or invoicing errors as the cause field, insufficient sample size windows, and material pairs with cluster confidence scores below the text confidence threshold, thereby reducing the impact of non-quality factors on the safety threshold. Safety Threshold The calculation formula is set as follows: ;

[0139] In the formula, Confidence coefficient; The security thresholds are pre-loaded into the system for the policy execution module; The mathematical expectation of the historical coherence cumulative index sequence; is the standard deviation of the historical coherence cumulative index series.

[0140] Confidence coefficient The standard calibration value range is set between 2.5 and 3.5. For quality-sensitive materials, A smaller value can be taken; for low-risk materials, A larger value can be chosen to reduce the probability of false triggering. Comparison of strategy execution modules. and Based on the comparison results, the system selects either horizontal scheduling or vertical blocking of the process.

[0141] S502, when At that time, the system did not find any obvious upstream and downstream return correlation, so it prioritized processing the abnormal phase as insufficient capacity, order queue or invoicing delay.

[0142] The strategy execution module initiates a horizontal scheduling mechanism. Using the abnormal output material identifier as a search criterion, the strategy execution module searches the global agent mapping table for other node enterprises capable of supplying the same type of output material.

[0143] S503, provided at least one parallel backup node enterprise is matched, the strategy execution module calculates the procurement allocation weight based on the historical invoicing frequency data of each backup node enterprise. Let the total number of parallel backup node enterprises matched by the system be... , No. The historical average daily number of invoices issued by each parallel backup node enterprise is: Then the corresponding purchase order allocation coefficient The calculation model is as follows: ;

[0144] In the formula, To allocate resources to the first during network reconstruction Percentage of purchase orders (allocation coefficient) of each parallel standby node enterprise. For the first The average daily number of invoices issued by each parallel backup node enterprise within the historical window; This represents the total number of parallel backup node enterprises matched by the system. For the first The average daily number of invoices issued by each parallel backup node enterprise within the historical window.

[0145] Before calculation, the system first determines Is it zero? If the denominator is zero. A value of zero indicates that all matched backup nodes have no historical transaction data within the observation window. The strategy execution module will then assign the allocation coefficients of each backup node enterprise to the relevant data. Unified assignment The strategy of proportional and equal distribution is implemented.

[0146] The strategy execution module uses the calculated allocation coefficients. A horizontal scheduling instruction message is generated and sent to an external enterprise resource planning system or procurement management system via an application programming interface. The external system can generate procurement adjustment suggestions, pending purchase orders, or automatic purchase orders according to the allocation coefficient. For scenarios with high risk levels or high material value, the horizontal scheduling instruction is executed after entering a manual confirmation process.

[0147] To ensure logical completeness, if the strategy execution module does not find a matching parallel backup node enterprise in the global proxy bill of materials mapping matrix, it generates a single-point capacity blockage early warning message and pushes it to the front-end business console, transferring it to the manual evaluation process to prevent the system from waiting continuously.

[0148] S504, when the coherence cumulative index Greater than the safety threshold Furthermore, when the red-ink records involved in the calculation primarily correspond to reasons such as returns, rejections, quality disputes, non-compliance with specifications, or failure to pass warehousing acceptance, the strategy execution module determines that the current delivery delay poses a risk of hidden quality defect transmission and initiates a vertical blocking process. This judgment indicates that the quality risk has a high temporal coherence in the agency material chain and is not directly equivalent to the final quality incident determination.

[0149] The strategy execution module parses and generates highly coherent input red-word cancellation timing signals. Extract signals from the sequence whose amplitude exceeds a set peak threshold. The corresponding set of timestamps. This peak threshold. This is used to filter out sporadic, occasional return noise, low-amount discounts, and red-ink records with unclear reasons. Its value can be determined based on the unit value of the material, historical return frequency, regular defect rate, single-batch transaction quantity, and transaction amount, and is stored in the system configuration table. In one specific embodiment, the system uses the average daily normal return amount of the enterprise at that node over the past year as a baseline, and dynamically sets the peak threshold to 3-5 times this average daily baseline; or directly sets it to 5% of the total transaction quantity of a single specific high-value material. For red-ink records with missing reason fields or inconsistent auxiliary verification fields, the system reduces their peak contribution weight.

[0150] The strategy execution module extracts the invoice metadata stream and reverses the search for red-ink invoice records within the timestamp set. The module then extracts the seller identification number, composite material identifier, red-ink reason code, return order number, and exception date from the corresponding red-ink invoice record, marking the corresponding seller as a suspected quality source node. If the auxiliary fields indicate that the red-ink record pertains to a price adjustment, discount, or invoicing error, the seller is either not included in the suspected source set or only enters the manual review team.

[0151] If the same abnormal date corresponds to multiple sellers, the system will include all sellers in the suspected source set and retain their respective composite material identifiers and invoice record numbers in the blocking instruction.

[0152] S505, the strategy execution module extracts the source node enterprise identifier and the specific composite material identifier it provides from the above-mentioned location, and encapsulates it according to the external data interface protocol to generate a vertical blocking instruction.

[0153] Instructions can be sent to external systems via HTTP interface, message queue, or remote procedure call. The communication protocol itself can be implemented using existing technologies, and this embodiment does not limit it.

[0154] The strategy execution module sends the vertical blocking instruction to the external enterprise resource planning (ERP) system for execution. Upon receiving the instruction, the ERP system can perform corresponding actions based on the risk level: freeze the procurement authority for high-risk materials corresponding to the composite material identifiers of suspected source nodes; increase the warehousing verification level or add quality inspection review for medium-risk materials; and generate a manual supplier review task for materials with incomplete evidence fields. These control actions only apply to composite material identifiers supplied by suspected source nodes and do not directly freeze all business of that supplier.

[0155] The above process enables the system to generate anomaly judgment results based on invoice time series and red ink reversal records without reading equipment-level production data, and to convert the results into business actions such as procurement allocation, warehousing verification or supply freezing.

[0156] Specific application examples:

[0157] Once an electronics assembly company is selected as the target node company, the system will execute the following specific steps:

[0158] Data preprocessing and label generation:

[0159] The data preprocessing module sets the historical observation time window. The timeframe is 30 days. The daily average invoice issuance sequence of the target node enterprise is extracted, and the expected value of this sequence over 30 days is calculated. 120 images, standard deviation There are 18 invoices. Substitute them into the formula to calculate the invoice issuance time-series dispersion. The system's preset dispersion threshold The calibration is 1.0, because The system determines that the target node's business is stable and the data is valid.

[0160] Furthermore, the data preprocessing module maps the specification text, such as the main control chip model A, into micro-category identifiers using density clustering. And it is combined with the tax classification code to generate a composite material label. (as input material identifier) Similarly, the composite material identification for the finished product "multi-functional control motherboard" is generated. (as an output material identifier) ).

[0161] Modeling and reference phase difference extraction:

[0162] The topology modeling module extracts the total quantity of normal trading days to construct a time series signal, and substitutes it into the formula to obtain the Pearson correlation coefficient. If the value is greater than the preset threshold of 0.65, the corresponding material conversion relationship is established and written into the agent material list mapping matrix.

[0163] The feature calculation module sets the maximum delay boundary to 10 days and increments the time delay step by step. Calculated using the discrete-time cross-correlation function, when Timing, cross-correlation sequence Obtain the global extreme peak value. Based on this, the feature calculation module calculates the phase difference of the benchmark invoice flow. The lockout period is 3 days, which is physically equivalent to a standard production cycle time of 3 days for the control motherboard.

[0164] Real-time exception decoupling and scheduled execution:

[0165] Feature calculation module sets real-time sliding time window The duration is 9 days. At a certain system runtime point, the current dynamic flow phase difference is calculated. Extend to 6 days. Set tolerance threshold. for 15%, or 0.45 days, because Once the broadening trigger condition is met, the mass disturbance decoupling procedure is initiated.

[0166] Assuming on the 9th day of the window, the control motherboard experiences a return, and the output terminal cancels the timing signal in red. ;according to Shifting backwards, the invoice search reveals that a return invoice for the main control chip was issued to the supplier on day 6 (9-3=6), with the timing signal being reversed in red at the input. Coherence Cumulative Index .

[0167] The policy execution module loads the security threshold parameters: Confidence coefficient To obtain the safety threshold .because The system confirmed that the delivery delay was caused by the spread of latent quality defects across nodes. The strategy execution module then extracted the seller identification number from the red-ink invoice on the 6th day, located a certain semiconductor supplier B, and generated a vertical blocking instruction to freeze the subsequent purchasing rights of that node.

[0168] Application testing:

[0169] To verify the reliability of the technical solution provided by this invention in a complex industrial data environment, an experimental verification was conducted using a real historical invoice metadata database from a regional electronics manufacturing industry cluster spanning nearly 12 months. This database contains 542 node enterprises, with an average daily circulation of over 150,000 digital invoices.

[0170] Experimental Environment and Group Setup: A control group and an experimental group were established. The control group system adopted the traditional enterprise resource planning system monitoring logic, relying solely on the number of overdue days for a single order and the absolute threshold of the local return rate for alarms, without importing discrete-time cross-correlation functions for phase alignment and coherence calculations; the experimental group deployed the complete modules of this invention (data preprocessing, topology modeling, feature calculation, and strategy execution modules).

[0171] Test execution standard: Both systems were input with the same 12-month historical test dataset in parallel. To ensure the objectivity of the evaluation, all scheduling and early warning instructions output by the system were manually calibrated and verified back-to-back by three senior supply chain audit specialists, in conjunction with the corresponding company's offline physical material flow ledgers, actual return orders, and quality inspection reports.

[0172] Summary of results:

[0173] like Figure 3 As shown, cross-correlation sequences The amplitude varies with the time delay step The increase exhibits a typical statistical distribution characteristic of first rising and then falling, and in A sharp and unique global peak is formed at this point. This curve distribution visually verifies the scientific validity of the discrete-time cross-correlation function used in this invention for time-domain analysis, proving that the system can accurately map the originally disordered and noisy external invoice time series to the physical production rhythm (i.e., flow phase difference) determined within the target enterprise through mathematical transformation. This provides an absolutely reliable computational benchmark for the subsequent accurate separation of pure capacity blockage and quality creep.

[0174] Combination Figure 4 The characteristics of the experimental data, as can be seen from the intuitive presentation, are as follows:

[0175] On the one hand, the accuracy of defect tracing has improved significantly. For example... Figure 4 As shown in the data distribution area on the left, the control group system tends to treat small but highly correlated quality disturbances as normal fluctuations, with a quality defect localization accuracy of only 73.5%. In contrast, the experimental group's column height is significantly higher, thanks to the coherence accumulation index multiplicatively amplifying and convolutionally matching the correlated return events on the time tick, resulting in an accuracy jump to 96.2%.

[0176] On the other hand, a precipitous reduction in the scheduling false alarm rate was achieved. For example... Figure 4 As shown in the data distribution area on the right, when a surge in orders leads to limited net production capacity, the false alarm rate of the control group is as high as 19.8%, which is very likely to trigger secondary supply disruption risks; while the experimental group shows a cliff-like drop in the bar chart, and the false alarm rate of horizontal scheduling is effectively reduced to only 2.1%.

[0177] Conclusion: Based on the above appendix Figure 3 -Appendix Figure 4 The quantitative verification results fully demonstrate that the multidimensional decoupling and dynamic control scheduling mechanism described in this invention can deeply explore the data structure of financial and tax invoices, scientifically separate the temporal characteristics of production capacity constraints and quality defects, and achieve precise, flexible, and adaptive cross-organizational supply chain management under the stringent condition that it does not require deep integration with the factory's underlying industrial data.

Claims

1. A supply chain intelligent business management system based on financial and tax data, characterized in that, include: The data preprocessing module is used to receive invoice metadata streams from external invoice data sources without accessing the node enterprise's MES, BOM list, production schedule, or equipment operation data. The invoice metadata streams include at least the invoice date, buyer identifier, seller identifier, tax classification code, specifications, transaction quantity, transaction amount, invoice type, and red ink reason code or negative sales reason field. Based on the node enterprise's invoice records, the module calculates the invoice time series dispersion and generates a composite material identifier by combining the specifications text and tax classification code. The topology modeling module is used to construct a time series signal based on the composite material identifier and calculate the correlation coefficient to establish a proxy bill of materials mapping matrix; The feature calculation module is used to perform time delay calculation on the time series signal associated with the agent bill of materials mapping matrix to extract the reference invoice flow phase difference; The dynamic flow phase difference is calculated in real time, and when the dynamic flow phase difference is greater than the sum of the baseline invoice flow phase difference and the preset tolerance threshold, the dynamic flow phase difference is confirmed to be abnormal. After the abnormality is confirmed, the coherence cumulative index is calculated based on the red ink reversal timing signal constructed from quality-related red ink records after being filtered by the red ink reason code or the negative sales item reason field. The strategy execution module is used to generate horizontal scheduling instructions or vertical blocking instructions based on the proxy bill of materials mapping matrix, according to the comparison result between the coherence accumulation index and the pre-loaded security threshold.

2. The intelligent business management system for the industrial chain based on financial and tax data according to claim 1, characterized in that, After calculating the invoicing time-series dispersion based on the invoicing records of the node enterprises, the data preprocessing module is also used for: The total number of invoices issued daily by enterprises at statistical nodes within a set historical observation time window is used to generate a quantity sequence. The ratio of the standard deviation to the expected value of the quantity sequence is calculated as the invoice issuance time series dispersion. When the invoicing time-series dispersion is greater than the preset dispersion threshold, it is determined that the corresponding node enterprise has the characteristics of centralized invoicing or supplementary invoicing. If the logistics receipt date, warehousing acceptance date, or contract fulfillment date of the corresponding node enterprise can be obtained, then the time series can be reconstructed using the logistics receipt date, warehousing acceptance date, or contract fulfillment date. If an available auxiliary date cannot be obtained, the corresponding node enterprise will be added to the filter list, and the invoice date of the corresponding node enterprise will not be used in the calculation of the proxy bill of materials mapping matrix and the base invoice flow phase difference within the current calculation period.

3. The intelligent business management system for the industrial chain based on financial and tax data according to claim 1, characterized in that, The data preprocessing module combines specification and model text with tax classification codes to generate composite material identifiers, specifically including: Extract the specification and model field from the invoice metadata stream, clean and segment the specification and model field by deleting units of measurement, common words and invalid symbols, and convert the processed specification and model field into a high-dimensional feature vector; The high-dimensional feature vectors are clustered using a density-based spatial clustering algorithm to generate micro-category labels. The clustering confidence is calculated based on the distance from the sample to the cluster center, the average similarity within the cluster, the number of samples under the same tax classification code, and the outlier ratio. When the cluster confidence level meets the preset text confidence threshold, the tax classification code and micro-category identifier are combined to generate a formal composite material identifier; when the cluster confidence level does not meet the text confidence threshold, a temporary composite material identifier is generated, and its weight in subsequent mapping calculations is reduced or it is transferred to the manual confirmation queue.

4. The intelligent business management system for the industrial chain based on financial and tax data according to claim 1, characterized in that, The topology modeling module constructs a time-series signal based on the composite material identifier and calculates the correlation coefficient to establish a proxy bill of materials mapping matrix, specifically including: Based on the composite material identification, the input and output materials are extracted, and the invoice transaction quantities are summed and accumulated by day to construct the corresponding input time series signal and output time series signal respectively. The correlation coefficient between the input time series signal and the output time series signal is obtained by using the Pearson correlation coefficient to calculate the correlation. When the Pearson correlation coefficient is greater than the preset correlation threshold, and the sample size, composite material identification confidence level and time correction method of the corresponding material pair meet the proxy mapping entry conditions, the corresponding material pair will be written into the proxy material list mapping matrix. The proxy bill of materials mapping matrix records at least the input material identifier, output material identifier, correlation coefficient, observation window, sample size, confidence level of composite material identifier, and time correction method.

5. The intelligent business management system for the industrial chain based on financial and tax data according to claim 1, characterized in that, The feature calculation module performs time delay calculation on the time series signal associated with the agent bill of materials mapping matrix to extract the baseline invoice flow phase difference, specifically including: Set a time delay step, calculate the cumulative sum of the dot product of the input time series signal and the output time series signal after adding the corresponding time delay step, and construct a cross-correlation sequence; Within the set upper limit of summation, the time delay step is incrementally increased to perform an extreme value search, and the lag step that makes the cross-correlation sequence reach the global maximum value is determined as the reference invoice circulation phase difference.

6. The intelligent business management system for the industrial chain based on financial and tax data according to claim 1, characterized in that, When the feature calculation module determines that the dynamic flow phase difference is abnormal when it is greater than the sum of the baseline invoice flow phase difference and a preset tolerance threshold, the specific steps include: Establish a real-time sliding time window and reuse the cross-correlation function to obtain the dynamic flow phase difference of the current business state within the window; When the dynamic flow phase difference is greater than the sum of the base invoice flow phase difference and the system preset tolerance threshold, it is determined that the delivery extension condition is met, and the dynamic flow phase difference is confirmed to be abnormal.

7. The intelligent business management system for the industrial chain based on financial and tax data according to claim 1, characterized in that, The feature calculation module calculates the coherence accumulation index based on the red-ink cancellation time series signal constructed from quality-related red-ink records after filtering by the red-ink reason code or negative sales reason field. Specifically, it includes: Extract records where the invoice type indicator is red ink reversal or negative sales, and filter the reasons based on the red ink reason code, the negative sales reason field, and the available return order number, warehouse acceptance status, or contract performance record; Records with reason fields pointing to price adjustments, sales allowances, invoicing errors, tax rate adjustments, contract changes, or accounting reversals will be removed or have their weight reduced. Records with reason fields pointing to returns, rejections, quality disputes, non-compliance with specifications, unqualified warehousing and acceptance, or those associated with return orders will be treated as quality-related red-ink records. The input terminal, constructed from quality-related red-word records, and the output terminal, constructed from quality-related red-word records, are reconstructed according to the time step of the timestamp. The phase difference of the reference invoice circulation is set as the time shift amount, and the input red ink cancellation timing signal is shifted backward to align the input red ink cancellation timing signal with the output red ink cancellation timing signal on the time axis. The coherence accumulation index is obtained by multiplying the output red-word cancellation timing signal with the shifted input red-word cancellation timing signal.

8. The intelligent business management system for the industrial chain based on financial and tax data according to claim 1, characterized in that, The policy execution module is also used to calculate pre-loaded security thresholds, specifically: Offline statistics were performed on the historical coherence cumulative index sequence of the target node enterprise within a set historical observation period in which there were no obvious quality anomalies. Before offline statistics, exclude records with concentrated supplementary invoice issuance cycles, red-ink records where the reason field points to price adjustment or invoicing error, time windows with insufficient sample size, and material pairs with cluster confidence scores below the text confidence threshold. Calculate the expected value and standard deviation of the historical coherence cumulative index sequence; The mathematical expectation is added to the product of the confidence coefficient and the standard deviation to obtain and load the security threshold within the target node enterprise.

9. The intelligent business management system for the industrial chain based on financial and tax data according to claim 1, characterized in that, The strategy execution module generates horizontal scheduling instructions based on the proxy bill of materials mapping matrix, specifically including: When the coherence cumulative index is less than or equal to the safety threshold, and the red text reason field does not point to centralized quality return, rejection, quality dispute, specification non-compliance or unqualified warehousing acceptance, the output material identifier that currently has phase anomaly is used as the retrieval condition to match parallel backup node enterprises that supply the same output material identifier in the agent material list mapping matrix. Obtain the average daily number of invoices issued by each backup node enterprise within the historical window, and calculate and derive the corresponding purchase order allocation coefficient; The horizontal scheduling instruction is generated based on the purchase order allocation coefficient to adjust the purchase allocation of the standby node.

10. The intelligent business management system for the industrial chain based on financial and tax data according to claim 1, characterized in that, The strategy execution module generates vertical blocking instructions based on the proxy bill of materials mapping matrix, specifically including: When the coherence accumulation index is greater than the security threshold, extract the set of timestamps whose amplitude is greater than the set peak threshold from the red text cancellation timing signal at the input end; Reverse search the red-letter invoice records in the timestamp set, extract the seller identification number, composite material identifier, red-letter reason code, return order number and abnormal date, and mark the corresponding seller as a suspected quality source node; The vertical blocking instruction is generated based on the suspected quality source node, the corresponding composite material identifier, the red reason code, the coherence cumulative index, and the auxiliary verification field. The vertical blocking instruction triggers the freezing of procurement authority, stricter warehousing verification, quality inspection review or manual review based on the risk level, and the controlled object is limited to the corresponding composite material labeled materials supplied by the suspected quality source node.

Images