Distributed multi-layer data processing method and system based on supply chain financial platform

Abstract

The present application relates to the technical field of financial services, in particular to a distributed multi-layer data processing method and system based on a supply chain financial platform, comprising the following steps: extracting accounts receivable vouchers, orders and logistics document data within a target time window at an enterprise local node. The present application extracts and field encryption splits accounts receivable vouchers, orders and logistics document data within a target time window at an enterprise local node, and combines the generated encrypted transaction factor shares across nodes to build a distributed credit factor key, which can ensure the safe sharing and fusion of credit factors between enterprises on the premise that the original data is not disclosed. The key is used to aggregate the encrypted share values between enterprise nodes, and the normalized and mapped operation is used to obtain a penetrating enterprise credit score, so that the enterprise credit status can be quantitatively presented in a cross-subject, distributed computing environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of financial services, and in particular to a distributed multi-layer data processing method and system based on a supply chain financial platform. BACKGROUND

[0002] The technical field of financial services belongs to the cross field of the integration of finance and information technology, focusing on improving the efficiency of financial services, risk control ability and business intelligence level through digital means.

[0003] In the prior art, in the process of credit audit in units of enterprises, the static financial statements and credit records submitted by the enterprises are relied on, so that the credit limit cannot reflect the current performance ability and supply stability. For example, even if a small and medium-sized enterprise has stable performance records, if its core materials rely on a single upstream enterprise, the prior art cannot perceive the supply chain vulnerability, which can easily lead to misplacement of funds and magnify the risk of chain breakage. Therefore, improvement is needed. SUMMARY

[0004] The purpose of the present application is to solve the problems existing in the prior art, and to provide a distributed multi-layer data processing method and system based on a supply chain financial platform.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme, a distributed multi-layer data processing method based on a supply chain financial platform, comprising the following steps:

[0006] In the local node of the enterprise, the accounts receivable vouchers, orders and logistics documents data in the target time window are extracted, the encrypted transaction factor shares are generated, the encrypted transaction factor shares of multiple enterprise nodes are combined, and the distributed credit factor key is established;

[0007] According to the distributed credit factor key, the encrypted share values of each node are aggregated to obtain an aggregated credit calculation intermediate value, and the aggregated credit calculation intermediate value is calculated to obtain a penetrating enterprise credit score;

[0008] The transaction orders and logistics documents between upstream and downstream enterprises in the supply chain are collected, the enterprise nodes and supply relationships are identified, the enterprise node set is constructed, the supply relationships are weighted according to the transaction amount or material flow between enterprises, the weighted transaction relationship pairs are generated, all enterprise nodes and the weighted transaction relationship pairs are connected, and the supply chain financial network topology graph is established by structuring the link centering on the transaction related to the core enterprise credit.

[0009] According to the supply chain financial network topology graph, the supply path redundancy of the key material is obtained, the penetrating enterprise credit score is combined, and a value for adjusting the credit limit is output to obtain a dynamic financing risk adjustment coefficient.

[0010] Preferably, the step of obtaining the distributed credit factor key is:

[0011] At the enterprise local node, the accounts receivable data, order data and logistics document data are screened according to a target time window, the transaction date, amount, due date and signing time are checked, the order number is grouped and counted and accumulated according to the time granularity, the transaction frequency, average account period and performance rate are calculated, the field sequence is segmented and encrypted, and the encrypted transaction factor share is generated;

[0012] According to the encrypted transaction factor share, the encrypted transaction factor shares in the same target time window are aggregated according to the enterprise local node identifier, the share label and timestamp continuity are verified, the shares are combined and repeated shares are removed according to the unified field order, and the cross-node encrypted transaction factor share combination result is formed;

[0013] According to the cross-node encrypted transaction factor share combination result, the transaction frequency, average account period and performance rate index corresponding to the share are reconstructed according to the enterprise local node identifier, the combination order and share integrity are verified, the key fragments and version identifier are connected according to the preset splicing rule, and the distributed credit factor key is generated.

[0014] Preferably, the step of obtaining the aggregated credit calculation intermediate value is:

[0015] In the plurality of enterprise nodes participating in the financing application, the encrypted share value label and timestamp are verified by calling the distributed credit factor key, the encrypted share values in the same target time window are combined according to the node order, the missing shares and repeated shares are detected and the data is filled, and the aggregated credit calculation intermediate value is obtained.

[0016] Preferably, the step of obtaining the penetrating enterprise credit score is:

[0017] According to the aggregated credit calculation intermediate value, the minimum value and maximum value of the historical aggregated credit calculation intermediate value in the target time window are retrieved, the proportional position of the current aggregated credit calculation intermediate value relative to the minimum value and maximum value is calculated, the dimension is eliminated according to the proportional position and the interval consistency is maintained, and the normalized aggregated credit calculation intermediate value is obtained;

[0018] According to the normalized aggregated credit calculation intermediate value, the penetrating enterprise credit score is calculated.

[0019] Preferably, the step of obtaining the weighted transaction relationship pair is:

[0020] The transaction order and logistics document between the enterprises on the supply chain are extracted according to a target time window, and the order number, delivery order number, enterprise unified social credit code, buyer and seller identifier, delivery and receiving party identifier, transaction amount, logistics weight and piece, delivery time and signing time are extracted, the order number and delivery order number are de-duplicated, and the consistency of the buyer and seller identifier and the delivery and receiving party identifier is checked, each record is mapped to an enterprise node and the supply relationship direction is determined, and an enterprise node set and a supply relationship identification result are generated;

[0021] According to the enterprise node set and the supply relationship identification result, the transaction amount and the logistics weight of the enterprise node pair in the target time window are aggregated, the currency unit and the measurement unit are unified, the missing indicators are filled in the latest valid record, the transaction frequency of the enterprise node pair is counted, the records confirmed repeatedly in the same working day are removed, the single strength index is determined according to the transaction amount or the logistics weight, and the weighted transaction relationship pair is generated.

[0022] Preferably, the acquisition step of the supply chain financial network topology map is:

[0023] According to the weighted transaction relationship pair, all enterprise nodes and weighted transaction relationship pairs are connected, the transaction identifier related to the core enterprise credit is extracted, and the weighted transaction relationship pair related to the core enterprise is screened, a one-hop adjacent layer and a two-hop adjacent layer are constructed according to the core enterprise as the center, the edge direction and the single strength index are kept consistent, and the supply chain financial network topology map is formed.

[0024] Preferably, the acquisition step of the dynamic financing risk adjustment coefficient is:

[0025] In the supply chain financial network topology map, the directed path from the core enterprise node to the core enterprise is retrieved layer by layer according to the key material identifier, the transaction order number and the logistics document number are checked piece by piece, and the repeated paths are removed, all valid paths satisfying that the enterprise nodes are not shared and the edges are unique are screened, and the number of enterprise node non-shared valid paths is obtained.

[0026] According to the number of enterprise node non-shared valid paths, the supply path redundancy is calculated and normalized.

[0027] Preferably, the acquisition step of the dynamic financing risk adjustment coefficient further includes: according to the normalized supply path redundancy and the normalized penetrating enterprise credit score, the dynamic financing risk adjustment coefficient is calculated.

[0028] The application also provides a data processing system, comprising:

[0029] A data extraction module is used for extracting accounts receivable vouchers, order and logistics document data in a target time window at an enterprise local node, generating an encrypted transaction factor share, combining encrypted transaction factor shares of multiple enterprise nodes, and establishing a distributed credit factor key.

[0030] a credit calculation module configured to aggregate the encrypted share values of each node according to the distributed credit factor key to obtain an aggregated credit calculation intermediate value, and to calculate the aggregated credit calculation intermediate value to obtain a penetrating enterprise credit score;

[0031] a network topology construction module configured to collect transaction orders and logistics documents between upstream and downstream enterprises in a supply chain, identify enterprise nodes and supply relationships, construct an enterprise node set, assign weights to the supply relationships according to transaction amounts or logistics volumes between enterprises, generate a weighted transaction relationship pair, connect all enterprise nodes and the weighted transaction relationship pair, and structurally link transactions associated with a core enterprise credit to establish a supply chain finance network topology graph;

[0032] a risk adjustment module configured to obtain a supply path redundancy of a key material according to the supply chain finance network topology graph, combine the penetrating enterprise credit score, output a value for adjusting a credit limit, and obtain a dynamic financing risk adjustment coefficient.

[0033] Compared with the prior art, the application has the following advantages and positive effects:

[0034] The application can ensure secure sharing and fusion of credit factors between enterprises on the premise that original data is not disclosed, can obtain a penetrating enterprise credit score by normalizing and mapping the aggregated encrypted share values between enterprise nodes based on the key, can enhance the graph expression capability of supply relationships by matching and checking transaction orders and logistics documents between upstream and downstream enterprises in a supply chain, identifying real supply relationships between enterprise nodes, and constructing a weighted transaction relationship pair and a structured network graph, can provide a structural basis for subsequent risk control analysis, and can further extract multi-path supply structure features of a key material and quantify redundancy, thereby realizing dynamic risk adjustment capability of joint modeling of enterprise risk and supply chain stability. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 The application is illustrated by the following steps. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0037] Referring to Figure 1 The present application provides a technical scheme, a distributed multi-layer data processing method based on a supply chain financial platform, comprising the following steps:

[0038] In the enterprise local node, the accounts receivable vouchers, order and logistics document data in the target time window are extracted, the encrypted transaction factor shares are generated, the encrypted transaction factor shares of multiple enterprise nodes are combined, and the distributed credit factor secret key is established;

[0039] According to the distributed credit factor secret key, the encrypted share values of each node are aggregated to obtain an aggregated credit calculation intermediate value, and the aggregated credit calculation intermediate value is calculated to obtain a penetrating enterprise credit score;

[0040] The transaction orders and logistics documents between upstream and downstream enterprises in the supply chain are collected, the enterprise nodes and supply relationships are identified, the enterprise node set is constructed, the supply relationships are weighted according to the transaction amount or material flow between enterprises, the weighted transaction relationship pairs are generated, all enterprise nodes and weighted transaction relationship pairs are connected, and the supply chain financial network topology graph is established by structuring the link centering on the transaction related to the core enterprise credit.

[0041] According to the supply chain financial network topology graph, the supply path redundancy of the key material is obtained, combined with the penetrating enterprise credit score, the value for adjusting the credit limit is output, and the dynamic financing risk adjustment coefficient is obtained.

[0042] The steps for obtaining the distributed credit factor secret key are as follows:

[0043] In the enterprise local node, the accounts receivable voucher data, order data and logistics document data are screened according to the target time window, the transaction date, amount, due date and signing time are checked, the order number is grouped and counted according to the time granularity, the transaction frequency, average account period and performance rate are calculated, the field sequence is segmented and encrypted, and the encrypted transaction factor shares are generated;

[0044] According to the encrypted transaction factor shares, the encrypted transaction factor shares in the same target time window are aggregated according to the enterprise local node identifier, the continuity of the share label and the timestamp is verified, the repeated shares are removed according to the unified field order, and the cross-node encrypted transaction factor share combination result is formed.

[0045] According to the cross-node encrypted transaction factor share combination result, the transaction frequency, the average account period and the performance rate index corresponding to the share are reconstructed according to the enterprise local node identifier, the merging order and the share integrity are verified, the key fragments and the version identifier are connected according to the preset splicing rule, and the distributed credit factor key is generated.

[0046] Specifically, in the enterprise local node, the accounts receivable voucher data, the order data and the logistics document data in the target time window, for example, in the last 90 days, are screened out, and data checking is performed first, specifically, the transaction amount in the order data is compared with the amount in the accounts receivable voucher, and an amount difference threshold is set, which is obtained by adding 1.5 times of the standard deviation to the average value of the consistency error generated by the normal fluctuation in the historical transaction data, for example, the threshold is calculated to be 0.05% of the transaction amount, if the absolute value of the difference exceeds the threshold, it is marked as abnormal data and excluded from the subsequent calculation, at the same time, the delivery date in the order and the delivery time in the logistics document, and the due date in the accounts receivable voucher and the expected signing time in the logistics document are verified to ensure logical consistency, then, the data checked is grouped according to the order number, the number of occurrences of each order number in the target time window is counted, and the daily transaction count is accumulated in day granularity, then three core indicators are calculated based on the accumulation result, the transaction frequency is calculated as the total number of transactions in the target time window divided by the number of days, for example, 180 transactions are completed in 90 days, then the transaction frequency is 2 transactions per day, the average account period is obtained by calculating the number of days from the “payment confirmation date” to the “accounts receivable generation date” of each transaction, and then taking the average value of all transactions, and the performance rate is calculated by counting the number of transactions in the logistics document whose actual signing time is earlier than or equal to the order agreed due date, and then dividing by the total number of transactions, for example, 97 out of 100 transactions are timely performed, then the performance rate is 97%, after the index calculation, the three values of transaction frequency, average account period and performance rate are spliced into a field sequence according to the predetermined order (for example, frequency, account period, performance rate), and the sequence is encrypted by using the SM4 block cipher algorithm, in order to further improve the data security, the encrypted sequence is divided into n shares by applying the (t, n) threshold secret sharing scheme (such as Shamir secret sharing scheme), n is set as the number of participating nodes, t is set as the minimum number of shares that can recover the secret, for example, (3, 5) scheme is set, the encrypted sequence is taken as the secret S, a t-1 degree polynomial is randomly generated, and n shares are distributed to different calculation tasks or nodes, each share contains share data, enterprise node identifier, generation timestamp and share serial number, and the encrypted transaction factor share is generated.

[0047] According to the encrypted transaction factor shares generated in the preceding steps, a data aggregation process is started, and the encrypted transaction factor shares are broadcast and collected among all participating enterprise local nodes by a designated coordination node or through a decentralized Gossip protocol. During the aggregation process, first, the shares from different nodes are classified according to the enterprise local node identifier carried in the shares, and it is limited within the same target time window, for example, the fourth quarter of 2023. Then, the share set submitted by each node is checked, and the checking content includes the continuity of the share label and the timestamp. The share label is a unique identifier, for example, composed of “node ID-time window-sequence number”. It is checked whether there is a repeated label. If there is a repeated label, only the first received one is retained. The timestamp continuity check is to check whether the timestamps of the shares submitted by the same node within a target time window appear continuously at a preset frequency (for example, once a day). A timestamp tolerance window is set, for example, 24 hours plus or minus 2 hours. If the interval between the timestamps of two consecutive shares exceeds this range, it is considered that there is a data loss. For a single missing share, linear interpolation is used to fill in the values according to the values of the two valid shares before and after it. If the number of consecutive missing shares exceeds the preset upper limit of filling, for example, 3 consecutive shares, the data of the node in this time window is marked as unavailable. After the checking and filling are completed, the data set is horizontally merged according to the unified field order, for example, “timestamp, node A share, node B share,...”, to form a matrix structure data set. Then, the merged data set is de-duplicated. The hash value (for example, using the SHA-256 algorithm) of each share content is calculated, and the share label is compared to accurately identify and remove the duplicate shares with the same content, and finally form the cross-node encrypted transaction factor share combination result.

[0048] According to the cross-node encrypted transaction factor share combination result formed in the last process, first, according to the enterprise local node identifier, an index is established for each encrypted share, which can quickly map back to the original unencrypted data corresponding to the transaction frequency, average account period and performance rate index type, but does not touch the specific value, for example, a mapping table is established, {'Share Fragment 1': 'Transaction Frequency', 'Share Fragment 2': 'Average Account Period'}, which is used for subsequent structured reconstruction, then the final integrity check is carried out, in terms of merging order, verify whether the timestamp in the entire combination result is monotonically increasing to prevent data disorder, in terms of share integrity, set an integrity benchmark value, which is calculated according to the stable activity of historical nodes participating in the same period, for example, historically, 98% of the nodes can provide valid data within the specified window, so the benchmark value is set to 98%, the share integrity of the current window is calculated as the actual number of valid nodes divided by the total number of expected participating nodes, if the calculation result is lower than the benchmark value, the combination result of this batch is considered invalid and is not used to generate the secret key, after passing the integrity check, the generation process of the secret key begins, which needs to execute the reconstruction algorithm of the secret sharing scheme, for example, for the (3, 5) Shamir scheme used before, at least 3 valid shares are extracted from the share combination of different nodes, and the Lagrange interpolation polynomial is used to calculate in the finite field to restore the original, encrypted field sequence on a single node, then, all the restored encrypted field sequences of the nodes are connected according to the preset splicing rule, which is defined as sorting by the alphabetical order of the enterprise node ID, and then splicing the restored sequences of each node in turn to form a longer byte string, finally, a version identifier is added at the head of the byte string, in the format of "YYYYMMDD-HHMM- version number", for example, "20231231-1800-v2.1", to distinguish the secret keys generated at different times, and the final long byte string with version identifier is the distributed credit factor secret key.

[0049] The acquisition step of the aggregated credit calculation intermediate value is:

[0050] In the multiple enterprise nodes participating in the financing application, the distributed credit factor secret key is called to check the encrypted share value label and timestamp, the encrypted share values in the same target time window are merged according to the node order, the missing shares and repeated shares are detected and the data is filled, and the aggregated credit calculation intermediate value is obtained.

[0051] Specifically, among multiple enterprise nodes collaboratively participating in the financing application, the distributed credit factor key generated in the aforementioned steps is invoked to verify the encrypted share values ​​collected from each node within the current target time window. This verification process utilizes the version identifier and timestamp information embedded in the key, matching them with the tags and timestamps attached to the encrypted share values. A timestamp synchronization fault tolerance window is set, its value based on historical network latency statistics, for example, taking the historical 95th percentile latency value and setting it to 500 milliseconds. If the timestamp difference between the two is within this window, and the tag format conforms to the preset rule of "node ID-time window-serial number," then authentication is successful. The authenticated encrypted share values ​​are merged according to a pre-set node order, which is determined based on the enterprise's hierarchy and importance in the supply chain. For example, core enterprise nodes are ranked first, followed by their first-tier suppliers, and so on, forming a time-series aligned merged dataset. Subsequently, the dataset undergoes integrity checks, firstly detecting duplicate shares. Each encrypted share value has its SHA-256 hash digest calculated. If shares with the same hash value are found, only the first received record is retained, and the rest are marked as duplicates and discarded. Next, missing shares are detected by checking the continuity of timestamps in the merged dataset. The preset share generation frequency is once a day. If the interval between two adjacent timestamps exceeds 25 hours (24 hours plus a 1-hour buffer), it is determined to be a missing data. For the found missing shares, a data completion strategy is adopted. Specifically, the two most recent valid share values ​​are taken and linearly interpolated to obtain a replacement value. However, if the number of consecutively missing shares exceeds a preset limit, such as 3 consecutive days, the interpolation is stopped, and the data of that node in this segment is marked as unreliable. After all completion and processing are completed, the share values ​​of all nodes are aggregated at each time point. The aggregation method is to calculate the arithmetic mean of the share values ​​of all nodes, and finally obtain a single time series numerical stream, which is the intermediate value of aggregated credit calculation.

[0052] The steps for obtaining a penetrating corporate credit score are as follows:

[0053] Based on the aggregated credit calculation median, retrieve the minimum and maximum values ​​of the historical aggregated credit calculation median values ​​within the target time window, calculate the proportional position of the current aggregated credit calculation median value relative to the minimum and maximum values, eliminate the dimensions according to the proportional position while maintaining interval consistency, and obtain the normalized aggregated credit calculation median value.

[0054] Based on the median value of the normalized aggregated credit calculation, the penetrating corporate credit score is calculated using the following formula:

[0055] ;

[0056] in, It is a penetrating corporate credit score. for a predetermined lower bound of a score interval, for a predetermined upper bound of a score interval, for a normalized aggregated credit calculation intermediate value, for a credit stratification sensitivity coefficient, for a credit score inflection point.

[0057] Specifically, according to the aggregated credit calculation intermediate value obtained in the previous process, in order to eliminate the dimensional influence caused by the inconsistent fluctuation range of data in different time windows, normalization processing is performed. First, the reference range for normalization needs to be determined, which is achieved by retrieving the aggregated credit calculation intermediate values stored in the historical database. The historical tracing period is set to the past 24 months. From the data of these 24 months, the global minimum value and the global maximum value of all aggregated credit calculation intermediate values are found. For example, it is found through retrieval that the historical minimum value is 125.6 and the maximum value is 845.2. These two values will be used as the boundaries for normalization. Then, the current aggregated credit calculation intermediate value is compared with these two boundary values to calculate its relative position. The calculation process is as follows: subtract the historical minimum value from the current aggregated credit calculation intermediate value to obtain a difference value, and then divide this difference value by the difference (i.e., the range) between the historical maximum value and the historical minimum value. The calculation result is a proportion value between 0 and 1, which represents the position of the current credit level in the historical fluctuation range. For example, if the current aggregated credit calculation intermediate value is 650.0, the calculation process is (650.0 - 125.6) divided by (845.2 - 125.6), resulting in approximately 0.7287. This process converts the original value with a specific business dimension into a dimensionless value with consistent interval (0 to 1) by calculating the proportional position. The final proportion value obtained is the normalized aggregated credit calculation intermediate value.

[0058] Formula: The normalized credit intermediate value is mapped to a specified credit score interval. The S-shaped curve characteristic makes the score more sensitive to changes in input values near the inflection point, and tends to be flat at both ends, i.e., the risk change caused by changes in credit status of enterprises with medium credit levels is the largest, while the risk levels of enterprises with excellent or poor credit are relatively stable. The parameters and in the formula define the range of the score, set the central reference of the credit level, and control the differentiation of credit stratification.

[0059] For the lower limit of the predetermined score interval, the setting of this parameter refers to the score range of the domestic mainstream enterprise credit rating system, and combines the risk management requirements of financial institutions, and is set to a fixed 400 points. This score represents the minimum credit level that requires strict risk review.

[0060] For the upper limit of the predetermined score interval, the setting logic is consistent with the lower limit, and the score upper limit is set to a fixed 900 points by referring to industry standards. This score represents the level of enterprises with excellent credit conditions and extremely low risk.

[0061] For the normalized aggregate credit calculation intermediate value, this parameter is calculated by the aforementioned steps and is an input variable of the formula. Its value range is [0, 1], representing the relative position of the current credit of the enterprise in the historical data range. In this example, the example value calculated by the aforementioned steps is used, i.e. .

[0062] For the credit layering sensitivity coefficient, this coefficient determines the steepness of the credit score curve, i.e. the sensitivity of the score to changes. The value of this coefficient is set based on statistical analysis of the relationship between the default rate of enterprises in historical data and the normalized aggregate credit calculation intermediate value. First, collect samples of enterprises that defaulted within the past 36 months and their corresponding values, as well as samples of enterprises that performed normally, construct a logistic regression model to predict the relationship between and : After training the model, the absolute value of the coefficient reflects the actual impact of credit conditions on default risk, the value of which is adjusted and set based on this coefficient to ensure that the score differentiation matches the actual risk distribution, for example, after regression analysis, the is -9.8, considering the need for more detailed division of enterprises with medium credit in business, the is set to 10.

[0063] For the credit score inflection point, it represents the center point of the credit score curve, corresponding to the value of the credit rating "medium". The value of this parameter is set based on the statistics of the normalized aggregate credit calculation intermediate values of all enterprises in the past 24 months. The arithmetic mean of these values is calculated as the inflection point. This way, the scoring model can dynamically adapt to changes in the overall credit level of the entire supply chain ecosystem, for example, by statistically analyzing the data from the past 24 months, the average value of all is calculated to be 0.52.

[0064] According to the parameters, the calculation is as follows:

[0065] The set values of the parameters are substituted into the formula: , , , , .

[0066] The index part is calculated:

[0067] ;

[0068] The denominator part is calculated:

[0069] ;

[0070] The fractional part is calculated:

[0071] ;

[0072] The final penetration enterprise credit score is calculated:

[0073] ;

[0074] The result shows that the calculated penetration enterprise credit score is 844.79, which is the specific credit score of the enterprise in the [400, 900] score interval. According to the preset score level division standard, for example, 800 points and above are defined as "credit excellent", 700-799 points are defined as "credit good", 600-699 points are defined as "credit general", and less than 600 points are defined as "credit poor", then the evaluation result of 844.79 points belongs to the category of "credit excellent".

[0075] The acquisition steps of the weighted transaction relationship pair are:

[0076] The transaction orders and logistics documents between upstream and downstream enterprises in the supply chain are extracted according to the target time window, including order number, delivery order number, enterprise unified social credit code, buyer and seller identifier, sender and receiver identifier, transaction amount, logistics weight and piece count, delivery time and signing time. The order number and delivery order number are de-duplicated, and the consistency of the buyer and seller identifier and the sender and receiver identifier is checked. Each record is mapped to an enterprise node and the supply relationship direction is determined. The enterprise node set and supply relationship identification result are generated;

[0077] According to the enterprise node set and the supply relationship identification result, the transaction amount and logistics weight of the enterprise node pair in the target time window are aggregated, the currency unit and measurement unit are unified, and the missing indicators are filled in with the latest valid record. The transaction frequency of the enterprise node pair is counted, and the records confirmed repeatedly in the same working day are removed. The single strength index is determined according to the transaction amount or logistics weight, and the weighted transaction relationship pair is generated.

[0078] Specifically, in the transaction order and logistics document data source between upstream and downstream enterprises in the supply chain, the structured data fields including order number, delivery order number, enterprise unified social credit code, buyer and seller identification, sender and receiver identification, transaction amount, logistics weight and piece, delivery time and signing time are automatically extracted through API interface within the target time window of the last 180 days. First, the order number and delivery order number are combined into a composite primary key, and the hash table is used for traversal and deduplication to remove all duplicate records of the composite primary key, and keep the first occurrence. Then, consistency check is performed on each record after deduplication. Specifically, the buyer's enterprise unified social credit code in the order data is compared with the receiver's enterprise unified social credit code in the logistics document, and the seller code is compared with the delivery code. A Jaro-Winkler similarity threshold of enterprise name string matching is set. The threshold is obtained by statistical analysis of a large number of samples in historical data that are not completely matched due to input errors but are actually the same enterprise, and the 10th percentile of the similarity distribution is taken, for example, 0.92. Only when the codes are completely consistent or the name similarity is higher than the threshold, the record is considered valid. After the check, the enterprise unified social credit code in the record is mapped to the node of the graph, and the directionality of the supply relationship is determined according to the buyer and seller identification, that is, a directed edge is established from the seller node to the buyer node. Finally, all unique enterprise unified social credit codes constitute the enterprise node set, and all verified directional transaction relationships constitute the supply relationship identification result.

[0079] According to the enterprise node set and the supply relationship identification result generated according to the foregoing steps, for each identified enterprise node pair (for example, supplier A and purchaser B), all related transaction data within a set target time window is aggregated, specifically, first, the transaction amount and the logistics weight of all transactions are accumulated, and unit uniformization processing is performed before aggregation, all transaction amounts in non-renminbi are uniformly converted into renminbi according to the middle price exchange rate published by the China Foreign Exchange Trading Center on the day of the transaction, all logistics weight units such as tons and pounds are uniformly converted into kilograms, for missing indicators found in the aggregation process, for example, a transaction has only a transaction amount but no logistics weight, a forward filling method is used to find the latest transaction between the enterprise node pair with complete records, calculate the "weight / amount" ratio, and apply the ratio to the current transaction amount to estimate the missing logistics weight, but this method is only used when the coefficient of variation of the historical transaction ratio is less than a preset stability threshold (for example, 0.2) to avoid false filling due to large changes in transaction categories, then, the transaction frequency of the enterprise node pair within the time window is counted, to avoid false high frequency caused by system retransmission or manual repeated confirmation, multiple records with the same transaction amount and logistics weight occurring within the same working day (for example, 08:00 to 18:00 Beijing time) are considered as one transaction and are combined and removed, finally, one of the transaction amount and the logistics weight is determined as a single indicator for measuring relationship strength, the selection rule is: calculate the data completeness rate of the two indicators in all transactions, preferentially select the indicator with a higher completeness rate, if the completeness rates are the same, default to selecting the transaction amount, and the selected indicator value is used as the weight to generate a weighted transaction relationship pair.

[0080] The acquisition step of the supply chain financial network topology map is:

[0081] According to the weighted transaction relationship pair, all enterprise nodes and weighted transaction relationship pairs are connected, the transaction identifier associated with the core enterprise credit is extracted, and the weighted transaction relationship pair associated with the core enterprise is screened, a one-hop adjacent layer and a two-hop adjacent layer are constructed with the core enterprise as the center and the edge direction and the single strength indicator are kept consistent, and a supply chain financial network topology map is formed.

[0082] Specifically, according to the weighted transaction relationship pair generated according to the foregoing steps, a directed weighted graph is constructed using a graph database or a memory graph computing framework, wherein each enterprise in the enterprise node set is a vertex of the graph, each relationship in the weighted transaction relationship pair is a directed edge connecting the vertices, and the weight of the edge is the numerical value of the single strength indicator. Then, the identification (unified social credit code of enterprises) of the core enterprise is extracted from a pre-defined list of core enterprises, all edges in the graph are traversed, and edges whose starting node or terminal node is the core enterprise are screened out. These edges and the non-core enterprise nodes connected by the edges jointly constitute a subgraph centered on the core enterprise. Then, based on the subgraph, a breadth-first search (BFS) algorithm is executed to construct an adjacency layer. From the core enterprise node, all nodes that can be directly accessed through an edge (i.e., direct suppliers and direct customers of the core enterprise) are defined as a one-hop adjacency layer. Subsequently, from the nodes in the one-hop adjacency layer, a one-step breadth-first search is again performed, and new nodes that have not been marked as core enterprises or one-hop adjacency layers are accessed and defined as a two-hop adjacency layer (i.e., suppliers of suppliers or customers of customers). In constructing the two-layer network structure, the direction of the original edge (from the seller to the buyer) and the weight value of the single strength indicator on the edge are completely retained. Finally, the core enterprise, the one-hop adjacency layer nodes, the two-hop adjacency layer nodes, and all the weighted directed edges connecting them are integrated to form a structured local network view centered on the core enterprise, i.e., a supply chain finance network topology graph.

[0083] The step of obtaining the dynamic financing risk adjustment coefficient is:

[0084] In the supply chain finance network topology graph, the directed paths from the core enterprise node to the core enterprise are retrieved layer by layer according to the key material identification, the transaction order numbers and logistics document numbers are checked one by one and duplicate paths are removed, all valid paths that satisfy the conditions of enterprise node non-sharing and edge uniqueness are screened out, and the number of enterprise node non-sharing valid paths is obtained.

[0085] According to the number of enterprise node non-sharing valid paths, the supply path redundancy is calculated and normalized, and the calculation formula is:

[0086] ;

[0087] wherein, is the normalized supply path redundancy, is the number of enterprise node non-sharing valid paths, is a redundancy decay coefficient;

[0088] According to the normalized supply path redundancy and the normalized penetrating enterprise credit score, the dynamic financing risk adjustment coefficient is calculated, and the calculation formula is:

[0089] ;

[0090] wherein, is a dynamic financing risk adjustment coefficient, is a normalized penetration enterprise credit score, is a normalized supply path redundancy, is a business model dependency coefficient.

[0091] In particular, in the supply chain finance network topology graph constructed in the previous step, according to the key material identification specified by the core enterprise in the financing application, such as a specific semiconductor chip model "X-2048", a path retrieval process is started, which uses a depth-first search algorithm to start from the core enterprise node and trace upstream against the directed edges in the graph representing the supply direction (i.e. from buyer to seller), traversing all supply paths that can eventually link to the core enterprise. Whenever the algorithm finds a complete path, such as from "raw material supplier C" to "component manufacturer B" to "core enterprise A", it temporarily stores the path consisting of a sequence of nodes, and at the same time, in order to verify the real transaction background of the path, it needs to backtrack to the original transaction order and logistics document data to check whether each segment of the path (C->B, B->A) contains transaction records related to the key material identification "X-2048", and associates the order numbers and logistics document numbers of these records. After completing the preliminary discovery of all potential paths, path deduplication is performed, the node sequence of each path is converted into a standardized string (e.g. concatenated after sorting by node ID), and a set data structure is used to remove completely identical paths. Then, enter the key effective path screening stage, the goal is to find a set of independent backup supply routes, the screening standard is "enterprise nodes do not share", and a greedy algorithm is executed: first sort all found paths in descending order of length (number of nodes passed), select the longest path as the first effective path, then remove all paths that share any intermediate node (non-core enterprise node) with the selected path from the remaining path list, and select the next longest path from the updated path list, repeat the process until the path list is empty, finally count the total number of selected effective paths to get the number of enterprise node non-sharing effective paths.

[0092] Formula: The above formula is to convert the discrete, unbounded effective path number ( ) into a continuous, normalized to [0, 1) interval redundancy measure ( ), which adopts the form of a negative exponential function, which conforms to the law of diminishing marginal effect of supply chain redundancy, i.e. from no backup path ( , ) to one backup path ( The risk reduction effect is most obvious, and the additional risk reduction effect brought by each additional redundant path will gradually weaken, and the parameter controls the speed of this attenuation, so that the model can adjust the sensitivity to the number of redundant paths according to the characteristics of different industries.

[0093] The number of shared effective paths for the enterprise node is not shared, which is an integer greater than or equal to 1, and represents the number of independent upstream supplier links that supply key materials for the core enterprise. In this example, through the aforementioned graph retrieval and screening process, there are 4 independent supply paths for the key materials of the core enterprise, so .

[0094] The redundancy attenuation coefficient reflects the attenuation speed of the redundancy value brought by each additional supply path, and its value range is (0, +∞), The setting of is based on the analysis of the supply chain characteristics of a specific industry. For industries with strong supplier substitutability and low switching costs (such as standard part manufacturing), the value is small; for industries with highly specific suppliers and long certification cycles (such as aerospace and precision medical equipment), the value is large, and its specific value is determined through statistical analysis of historical supply chain disruption event data: collect cases of production delays caused by supplier disruptions within the industry, establish a relationship model between the probability of disruption event occurrence and the number of enterprise node non-shared effective paths , and fit the decay rate of the model as a reference value for For example, for the electronics manufacturing industry, analysis shows that when the number of independent suppliers increases from 1 to 3, the supply disruption risk decreases by about 80%. Based on this empirical data, the value of is set so that when , is close to 0.8, and the value of .

[0095] According to the parameter, the calculation is as follows:

[0096] Substitute the set value of each parameter into the formula: , .

[0097] Calculate the power of the index:

[0098] ;

[0099] Calculate the value of the exponential function:

[0100] ;

[0101] The final normalized supply path redundancy is calculated:

[0102] ;

[0103] The result shows that the calculated normalized supply path redundancy is 0.9093, which reflects the risk resistance ability of the supply chain at the structural level. A value close to 1 means that the core enterprise has a very diversified and independent supply source for this critical material, and even if one or more paths are problematic, the risk of overall supply disruption is lower.

[0104] Formula: The above formula is to combine the enterprise's credit level ( ) and the structural stability of the supply chain ( ) by weighted geometric mean, to generate a dynamic financing risk adjustment coefficient ( ). This Cobb-Douglas function form allows a certain complementarity between the two factors, i.e. the weaker side can be partially compensated by the stronger side, but any extreme low (close to 0) of either side will cause the final adjustment coefficient to drop sharply. The parameter as a weight quantifies the relative dependence of different business models on "endogenous enterprise credit" and "external supply chain stability", making the risk assessment model more targeted and adaptable.

[0105] is the normalized penetration enterprise credit score, which is derived from the calculation results of the early credit scoring step. It quantifies the enterprise's overall transaction performance as a value between 0 and 1, reflecting the enterprise's performance and credit status. In this example, the value calculated in the previous step is directly referenced, .

[0106] is the normalized supply path redundancy, which is calculated by the previous step and quantifies the structural robustness of the supply chain. In this example, the calculated result of the previous step is, .

[0107] is the business model dependence coefficient, which ranges from 0 to 1, and is used to adjust the weight of enterprise credit and supply chain redundancy in the comprehensive risk assessment. Its setting depends on the business model of the core enterprise and is determined by a quantitative assessment model. This model includes three assessment dimensions: critical material cost proportion ( ), supplier replacement complexity ( ), and product market concentration ( ), each dimension is scored by industry experts (1-10), and then normalized (divided by 10), and finally The calculation formula is: , where the weight is determined by historical data regression analysis, for example For a mobile phone manufacturer that is highly dependent on specific imported chips, the key material cost proportion score is 9, the replacement complexity is 10, but the product market is scattered, and the concentration is 4, so after normalization , the calculation shows that the dependence on the supply chain is Therefore .

[0108] According to the parameters, the calculation is as follows:

[0109] Substitute the parameter values into the formula: , , .

[0110] Calculate :

[0111] ;

[0112] Calculate the power of each part:

[0113] ;

[0114] ;

[0115] Calculate the final dynamic financing risk adjustment coefficient:

[0116] ;

[0117] The result shows that the calculated dynamic financing risk adjustment coefficient is 0.8852, which is a comprehensive coefficient that combines the enterprise's good credit ( close to 0.73) and very stable supply chain structure ( close to 0.91). Due to the high dependence of the business model on the supply chain ( small value), the weight of the supply chain is greater, and the final adjustment coefficient is higher, close to 0.9, which means that the enterprise's overall financing risk is low. When granting credit, the benchmark credit limit can be multiplied by the coefficient 0.8852 to obtain a relatively relaxed and risk-adjusted credit limit.

[0118] The above merely describes the preferred embodiments of the present application, and is not intended to limit the present application in other forms. Any skilled person in the art can modify or change the disclosed technical content into equivalent embodiments with equivalent changes, and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application, without departing from the technical solution content of the present application, still falls within the protection scope of the present application.

Claims

1. A distributed multi-layer data processing method based on a supply chain finance platform, characterized in that, The method comprises the following steps: At the enterprise local node, the accounts receivable data, order data and logistics document data in the target time window are extracted, the encrypted transaction factor shares are generated, the encrypted transaction factor shares of multiple enterprise nodes are combined, and the distributed credit factor key is established; According to the distributed credit factor key, the encrypted share values of each node are aggregated to obtain an aggregated credit calculation intermediate value, and the aggregated credit calculation intermediate value is calculated to obtain a penetrating enterprise credit score; The transaction orders and logistics documents between upstream and downstream enterprises in the supply chain are collected, the enterprise nodes and supply relationships are identified, the enterprise node set is constructed, the supply relationships are weighted according to the transaction amount or material flow between enterprises, the weighted transaction relationship pairs are generated, all enterprise nodes and the weighted transaction relationship pairs are connected, and the supply chain financial network topology graph is established by structuring the links centered on the transaction related to the core enterprise credit; According to the supply chain financial network topology graph, the supply path redundancy of the key material is obtained, the penetrating enterprise credit score is combined, and a value for adjusting the credit limit is output to obtain a dynamic financing risk adjustment coefficient; The distributed credit factor key is obtained by: At the enterprise local node, the accounts receivable data, order data and logistics document data in the target time window are extracted, the encrypted transaction factor shares are generated, the encrypted transaction factor shares of multiple enterprise nodes are combined, and the distributed credit factor key is established; According to the encrypted transaction factor shares, the encrypted transaction factor shares in the same target time window are aggregated according to the enterprise local node identifier, the continuity of the share label and the timestamp is verified, the shares are combined in a unified field order and repeated shares are removed to form a cross-node encrypted transaction factor share combination result; According to the cross-node encrypted transaction factor share combination result, the indexes of the transaction frequency, the average account period and the performance rate corresponding to the shares are reconstructed according to the enterprise local node identifier, the combination order and the share integrity are verified, the key fragments and the version identifier are connected according to the preset splicing rule, and the distributed credit factor key is generated; The aggregated credit calculation intermediate value is obtained by: At multiple enterprise nodes participating in the financing application, the encrypted share value label and the timestamp are verified by calling the distributed credit factor key, the encrypted share values in the same target time window are combined in the node order, the missing shares and the repeated shares are detected and the data is filled to obtain the aggregated credit calculation intermediate value; The penetrating enterprise credit score is obtained by: According to the aggregated credit calculation intermediate value, the minimum value and the maximum value of the historical aggregated credit calculation intermediate value in the target time window are retrieved, the proportional position of the current aggregated credit calculation intermediate value relative to the minimum value and the maximum value is calculated, the dimension is eliminated according to the proportional position and the interval consistency is maintained to obtain the normalized aggregated credit calculation intermediate value; According to the normalized aggregated credit calculation intermediate value, the penetrating enterprise credit score is calculated, and the calculation formula is: ； wherein, is a through-the-line business credit score, L is a predetermined lower score interval, U is a predetermined upper score interval, is a normalized aggregate credit score intermediate value, k is a credit score layer sensitivity coefficient, is a credit score inflection point; The dynamic financing risk adjustment coefficient is obtained by: In the supply chain finance network topology map, the directed path from the core enterprise node to the core enterprise is retrieved layer by layer according to the key material identification, the transaction order number and the logistics document number are checked piece by piece, and the repeated paths are removed, all the valid paths satisfying the enterprise node non-sharing and the edge uniqueness are screened, and the number of enterprise node non-sharing valid paths is obtained; According to the number of enterprise node non-sharing valid paths, the supply path redundancy is calculated and normalized, and the calculation formula is: ； wherein, N is the number of non-shared valid paths for the enterprise node, is the redundancy decay coefficient.

2. The distributed multi-tier data processing method based on the supply chain finance platform according to claim 1, characterized in that, The acquisition step of the weighted transaction relationship pair is: According to the enterprise node set and the supply relationship identification result, the transaction amount and the logistics weight of the enterprise node pair in the target time window are aggregated, the currency unit and the measurement unit are unified, the missing indicators are filled in the latest valid record, the transaction frequency of the enterprise node pair is counted, the records confirmed repeatedly in the same working day are removed, the single strength index is determined according to the transaction amount or the logistics weight, and the weighted transaction relationship pair is generated. The acquisition step of the supply chain finance network topology map is:

3. The distributed multi-tier data processing method based on the supply chain finance platform according to claim 1, characterized in that, According to the weighted transaction relationship pair, all enterprise nodes and weighted transaction relationship pairs are connected, the transaction identification related to the core enterprise credit is extracted, and the weighted transaction relationship pairs related to the associated core enterprise are screened, a one-hop adjacent layer and a two-hop adjacent layer are constructed according to the core enterprise as the center, the edge direction and the single strength index are kept consistent, and the supply chain finance network topology map is formed. The acquisition step of the dynamic financing risk adjustment coefficient further includes: according to the normalized supply path redundancy and the normalized penetrating enterprise credit score, the dynamic financing risk adjustment coefficient is calculated.

4. The distributed multi-tier data processing method based on the supply chain finance platform according to claim 1, characterized in that, It includes:

5. The data processing system of the distributed multi-tier data processing method based on the supply chain finance platform according to any one of claims 1-4, characterized in that, A data extraction module is used to extract accounts receivable vouchers, order and logistics documents in a target time window in an enterprise local node, generate encrypted transaction factor shares, combine encrypted transaction factor shares of multiple enterprise nodes, and establish a distributed credit factor key; A credit calculation module is used to gather encrypted share values of each node according to the distributed credit factor key to obtain an aggregated credit calculation intermediate value, and calculate the aggregated credit calculation intermediate value to obtain a penetrating enterprise credit score; A network topology construction module is used to collect transaction orders and logistics documents between upstream and downstream enterprises on a supply chain, identify enterprise nodes and supply relationships, construct an enterprise node set, assign weights to supply relationships according to transaction amounts or logistics quantities between enterprises, generate a weighted transaction relationship pair, connect all enterprise nodes and the weighted transaction relationship pair, and link the structure centering on the transaction related to the core enterprise credit to establish a supply chain finance network topology map; ​ A risk adjustment module is configured to obtain a supply path redundancy of the key material according to the supply chain finance network topology map, combine the penetrating enterprise credit score, output a value for adjusting the credit limit, and obtain a dynamic financing risk adjustment coefficient.

Images