Supply chain data-oriented enterprise upstream and downstream collaborative risk control prediction method and system

Abstract

The application discloses an enterprise upstream and downstream cooperative risk control prediction method and system facing supply chain data, relates to the technical field of supply chain financial risk management, and comprises the following steps: extracting a payment time deviation value between enterprises to construct a payment behavior sequence, calculating supply chain cooperative credit rating based on fluctuation amplitude and historical default records, and identifying a risk source enterprise; a recursive partition time window is adopted to extract a periodic index and a mutation index of the payment behavior, and a payment behavior portrait is constructed; correlation enterprises are identified through phase overlap of the payment behavior track, and a risk transmission sequence is formed; based on the evolution law of the payment behavior portrait and the risk transmission sequence, a risk trigger node is predicted, and a risk prevention and control scheme containing a credit limit control instruction and a behavior monitoring strategy is generated. The application can effectively identify a supply chain risk source, predict a risk transmission path, and improve the supply chain financial risk control capability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of supply chain financial risk management, in particular to an enterprise upstream and downstream collaborative risk control prediction method and system for supply chain data. BACKGROUND

[0002] With the rapid development of supply chain finance, the upstream and downstream transaction relationship between enterprises is becoming increasingly complex, and the credit risk between enterprises also presents the characteristics of transmission and chain. The existing enterprise credit risk assessment method mainly analyzes based on static financial indicators and single enterprise dimension, which is difficult to effectively identify and early warn the risk transmission path in the supply chain network.

[0003] The traditional enterprise credit evaluation model often ignores the dynamic characteristics of the payment behavior between enterprises, and cannot timely capture the real-time changes of the enterprise operating conditions; the existing method does not fully consider the synergy effect between the upstream and downstream enterprises in the supply chain, and it is difficult to accurately identify the risk source enterprise and its associated enterprises; lacking in-depth analysis of the evolution law of enterprise payment behavior, it is difficult to effectively predict the risk trigger time point, and the existing technology usually uses a fixed time window for analysis, which is difficult to adapt to the difference characteristics of different enterprise payment behaviors. In the identification of risk transmission path, it is mainly dependent on the simple correlation calculation, and the rich time sequence information contained in the enterprise payment behavior track is not fully utilized, and in the risk prevention and control scheme formulation link, a unified risk control strategy is often used, lacking differentiation processing for different enterprise payment behavior characteristics, resulting in unsatisfactory risk control effect. There is a lack of systematic and forward-looking consideration in credit limit regulation and behavior monitoring. SUMMARY

[0004] The purpose of the present application is to provide an enterprise upstream and downstream collaborative risk control prediction method and system for supply chain data, which analyzes the evolution law and synergy characteristics of the payment behavior between enterprises, realizes the accurate identification and early warning of the risk transmission path in the supply chain network, and provides more effective decision support for the risk prevention and control of supply chain finance.

[0005] The enterprise upstream and downstream collaborative risk control prediction method for supply chain data provided in the embodiment of the present application comprises the following steps:

[0006] Obtain the transaction records between the upstream and downstream enterprises in the supply chain, extract the payment time deviation value between the enterprises, and construct the payment behavior sequence;

[0007] Calculate the fluctuation amplitude based on the payment behavior sequence, calculate the supply chain collaborative credit rating based on the fluctuation amplitude and historical default records, and mark the corresponding enterprise as a risk source when the supply chain collaborative credit rating continuously decreases;

[0008] The payment behavior sequence of the risk source enterprise is mapped to a multi-dimensional feature space by using a recursive partitioning time window, and the periodic index and mutation index of the payment behavior are extracted to construct a payment behavior portrait.

[0009] An enterprise payment behavior trajectory is constructed in the multi-dimensional feature space, and associated enterprises are identified by the phase overlap degree of the payment behavior trajectory, and a risk transmission sequence is formed based on the synchronous change intensity.

[0010] The evolution law of the payment behavior portrait in the multi-dimensional feature space is used to predict the risk trigger node in combination with the risk transmission sequence, and a risk prevention and control scheme including credit limit control instructions and behavior monitoring strategies is generated.

[0011] Further, the transaction records between the upstream and downstream enterprises in the supply chain are obtained, and the payment time deviation value between the enterprises is extracted to construct a payment behavior sequence including:

[0012] The transaction record information between the upstream and downstream enterprises in the supply chain is obtained, and the transaction record information includes transaction amount and payment time;

[0013] The time deviation value between the actual payment time and the scheduled payment time in the transaction record information is calculated and standardized;

[0014] The correlation coefficient of the standardized time deviation value and the historical default record of the enterprise is calculated to generate a payment feature value;

[0015] Based on the payment feature value and the transaction correlation degree of the enterprise in the supply chain, a payment behavior matrix is constructed, the payment feature values between adjacent enterprises in the payment behavior matrix are weighted calculated to obtain a payment behavior fluctuation value, and the payment behavior sequence is constructed according to the payment behavior fluctuation value in time sequence.

[0016] Further, the fluctuation amplitude is calculated based on the payment behavior sequence, the supply chain collaborative credit rating is calculated based on the fluctuation amplitude and the historical default record, and when the supply chain collaborative credit rating continuously decreases, the corresponding enterprise is marked as a risk source including:

[0017] The payment behavior sequence of the supply chain enterprise is obtained, and the sliding variance test is used to identify the mutation point of the time interval sequence;

[0018] The payment behavior sequence is divided into multiple time periods with the mutation point as the boundary, and the payment behavior fluctuation amplitude in each time period is calculated;

[0019] The abnormal payment time point is identified according to the deviation degree of the payment behavior fluctuation amplitude, and the historical default record of the transaction enterprise corresponding to the abnormal payment time point is obtained;

[0020] Calculate the fluctuation range of payment behavior and the change trend and correlation coefficient of historical default records respectively, and take the correlation coefficient as the supply chain collaborative credit rating;

[0021] Monitor the change trend of the supply chain collaborative credit rating, and mark the corresponding enterprise as a risk source when the supply chain collaborative credit rating continuously decreases.

[0022] Further, the payment behavior sequence of the risk source enterprise is mapped to a multi-dimensional feature space by using a recursive partitioning time window, and the periodic index and mutation index of the payment behavior are extracted to construct a payment behavior portrait, including:

[0023] Extract the change features of adjacent time points in the payment behavior sequence of the risk source enterprise to construct a feature matrix, calculate the fluctuation cumulative value of the trend inflection point in the fluctuation curve based on the feature matrix, and generate the fluctuation curve;

[0024] Calculate the fluctuation cumulative value of the trend inflection point in the fluctuation curve, compare the fluctuation cumulative value with a historical fluctuation threshold to determine a candidate partition position;

[0025] Construct a time window at the candidate partition position, and determine the window boundary based on the aggregation degree distribution characteristics of the payment behavior in the time window;

[0026] Recursively divide the time window boundary to obtain time sequence segments, and merge adjacent segments according to the similarity of the change law of the payment behavior in the time sequence segments;

[0027] Construct a transition window at the boundary of the merged segments, calculate the fluctuation slope and dispersion of the payment behavior in the transition window, adjust the range of the transition window, and obtain the recursive partitioning time window;

[0028] Extract the time dimension feature, amount dimension feature and frequency dimension feature of the payment behavior in the recursive partitioning time window to construct a multi-dimensional feature space, extract the periodic index and mutation index of the payment behavior in the multi-dimensional feature space, and construct a payment behavior portrait.

[0029] Further, construct an enterprise payment behavior trajectory in the multi-dimensional feature space, identify associated enterprises through the phase overlap degree of the payment behavior trajectory, and form a risk transmission sequence based on the synchronous change intensity, including:

[0030] Extract the amount change and frequency change of adjacent time points in the enterprise payment behavior sequence to construct a change feature matrix, and map the change feature matrix to a multi-dimensional feature space to construct a payment behavior trajectory;

[0031] Construct an initial time window for the payment behavior trajectory, calculate the fluctuation range of the trajectory in the initial time window, and dynamically adjust the length of the initial time window according to the fluctuation range;

[0032] In the adjusted time window, the angle change sequence of different enterprise payment behavior trajectories is calculated, and the phase overlap degree is obtained based on the angle change sequence;

[0033] The fluctuation feature of the phase overlap degree is extracted, the fluctuation cumulative value is calculated, the fluctuation cumulative value is compared with the dynamic threshold value calculated based on historical data, the continuous overlap interval is identified, and the associated enterprise is identified based on the continuous overlap interval;

[0034] The change synchronization of the payment behavior trajectory of the associated enterprise in the continuous overlap interval is calculated, the synchronous change intensity is obtained, and the risk transmission sequence is formed based on the synchronous change intensity.

[0035] Further, the evolution rule of the payment behavior portrait in the multi-dimensional feature space is utilized, the risk trigger node is predicted in combination with the risk transmission sequence, and the risk prevention and control scheme including the credit limit control instruction and the behavior monitoring strategy is generated.

[0036] The time sequence feature of the payment behavior portrait in the multi-dimensional feature space is extracted, the change trend of the time sequence feature in adjacent time windows is calculated, and the behavior abnormal mode is identified;

[0037] The payment behavior feature of the associated enterprise is extracted based on the behavior abnormal mode, the divergence degree and the aggregation degree of the payment behavior feature are calculated, and when the divergence degree exceeds the preset degree threshold, the corresponding time point is marked as a risk trigger node;

[0038] The risk trigger node is backtracked, the evolution trajectory segment before triggering is extracted, the cumulative intensity of the feature change in the evolution trajectory segment is analyzed, and the risk outbreak time window is determined;

[0039] In combination with the position of the enterprise in the risk transmission sequence, the influence range in the risk outbreak time window is calculated, and the credit limit control instruction is set;

[0040] The behavior monitoring strategy is generated according to the credit limit control instruction, and the risk prevention and control scheme including the credit limit control instruction and the behavior monitoring strategy is output.

[0041] Further, in the adjusted time window, the angle change sequence of different enterprise payment behavior trajectories is calculated, and the phase overlap degree is obtained based on the angle change sequence includes:

[0042] The payment behavior trajectory vector sequence in the adjusted time window is obtained;

[0043] The payment behavior trajectory vector sequence is decoupled into instantaneous amplitude and instantaneous phase, and the dynamic boundary point is identified based on the mutation degree of the instantaneous amplitude and the instantaneous phase;

[0044] The payment behavior trajectory vector sequence is divided into multiple adaptive segments according to dynamic boundary points, and the angle change amount of adjacent payment behavior trajectory vectors is calculated in the adaptive segments to form an angle change sequence;

[0045] The angle change sequence is subjected to multi-scale decomposition to identify a mutation position, and the mutation position is combined with the angle change sequence to generate an enhanced angle change sequence;

[0046] The enhanced angle change sequences of any two enterprises are subjected to nonlinear dynamic normalization, and the phase overlap degree is calculated based on the timing matching degree of the mutation position in the normalization process.

[0047] In the embodiment of the application, an enterprise upstream and downstream collaborative risk control prediction system oriented to supply chain data is provided, and the system comprises:

[0048] A sequence construction module is configured to obtain transaction records between upstream and downstream enterprises in a supply chain, extract payment time deviation values between the enterprises, and construct payment behavior sequences;

[0049] A risk source identification module is configured to calculate fluctuation amplitudes based on the payment behavior sequences, calculate supply chain collaborative credit ratings based on the fluctuation amplitudes and historical default records, and mark corresponding enterprises as risk sources when the supply chain collaborative credit ratings continuously decrease;

[0050] A portrait construction module is configured to map payment behavior sequences of risk source enterprises to a multi-dimensional feature space by using recursive partitioning time windows, extract periodic indicators and mutation indicators of payment behaviors, and construct payment behavior portraits;

[0051] An associated enterprise identification module is configured to construct enterprise payment behavior trajectories in the multi-dimensional feature space, identify associated enterprises through phase overlap degrees of the payment behavior trajectories, and form risk transmission sequences based on synchronous change intensities;

[0052] A scheme generation module is configured to predict risk trigger nodes by using evolution rules of payment behavior portraits in the multi-dimensional feature space and combining risk transmission sequences, and generate risk prevention and control schemes containing credit limit control instructions and behavior monitoring strategies.

[0053] In the embodiment of the application, a technical solution is also provided, which is an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps in any of the preceding methods when executing the computer program.

[0054] In the embodiment of the application, a technical solution is also provided, which is a computer readable storage medium having computer program instructions stored thereon, wherein the computer program instructions are executed by a processor to implement the steps in any of the preceding methods.

[0055] This invention improves the accuracy of identifying risk-generating enterprises by constructing enterprise payment behavior sequences and calculating their fluctuation amplitudes, combined with historical default records, to achieve dynamic assessment of supply chain collaborative credit rating. It employs a recursive partitioned time window to map payment behavior sequences, enabling adaptive capture of payment behavior characteristics from different enterprises. By analyzing the phase overlap of payment behavior trajectories, it identifies related enterprises, enhancing the accuracy of risk transmission path identification. Based on the evolutionary patterns of payment behavior profiles, it predicts risk triggering nodes, achieving forward-looking risk warnings. Furthermore, by generating differentiated credit limit adjustment instructions and behavior monitoring strategies, it improves the targeting and effectiveness of risk prevention and control solutions. This invention enables early identification, accurate warning, and proactive prevention of risks in the supply chain network, significantly improving the efficiency of supply chain finance risk management. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 A flowchart of an enterprise upstream and downstream collaborative risk control prediction method based on supply chain data provided in an embodiment of the present invention;

[0058] Figure 2 This is a schematic diagram of the structure of a collaborative risk control and prediction system for upstream and downstream enterprises based on supply chain data, provided in an embodiment of the present invention. Detailed Implementation

[0059] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. In the following description relating to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements.

[0060] The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0061] like Figure 1 As shown, Figure 1A flowchart of an enterprise upstream and downstream collaborative risk control prediction method for supply chain data provided by the embodiment of the present application, the method comprising the following steps:

[0062] Obtaining transaction records between upstream and downstream enterprises in the supply chain, extracting payment time deviation values between enterprises, and constructing payment behavior sequences;

[0063] Calculating fluctuation amplitude based on payment behavior sequences, calculating supply chain collaborative credit rating based on fluctuation amplitude and historical default records, and marking corresponding enterprises as risk sources when supply chain collaborative credit rating continuously decreases;

[0064] Using recursive partitioning time window, mapping payment behavior sequences of risk source enterprises to multi-dimensional feature space, extracting periodic indicators and mutation indicators of payment behavior, and constructing payment behavior portraits;

[0065] Constructing enterprise payment behavior trajectories in multi-dimensional feature space, identifying associated enterprises through phase overlap degree of payment behavior trajectories, and forming risk transmission sequence based on synchronous change intensity;

[0066] Using evolution law of payment behavior portraits in multi-dimensional feature space, combining risk transmission sequence, predicting risk trigger nodes, and generating risk prevention and control scheme containing credit limit control instructions and behavior monitoring strategies.

[0067] Obtaining transaction records between upstream and downstream enterprises in the supply chain, extracting payment time deviation values between enterprises, and constructing payment behavior sequences include:

[0068] Obtaining transaction record information between upstream and downstream enterprises in the supply chain, the transaction record information containing transaction amount and payment time;

[0069] Calculating time deviation values of actual payment time and scheduled payment time in transaction record information, and performing standardization processing;

[0070] Calculating correlation coefficient of standardized time deviation values and historical default records of enterprises, and generating payment characteristic values;

[0071] Based on payment characteristic values and transaction association degree of enterprises in the supply chain, constructing payment behavior matrix, performing weighted calculation on payment characteristic values between adjacent enterprises in the payment behavior matrix, obtaining payment behavior fluctuation values, and constructing payment behavior sequences according to time sequence.

[0072] First, obtain transaction record information between upstream and downstream enterprises on the supply chain from the enterprise resource planning system or supply chain management platform. The transaction record information includes key data items such as transaction amount, agreed payment time, and actual payment time. During data acquisition, use data cleaning techniques to remove outliers and missing values to ensure data quality. For missing payment time data, the average payment period of the enterprise's historical transaction records can be used for reasonable filling; for abnormal transaction amount, compare it with the mean of the enterprise's recent transaction amount, and mark the data with a deviation exceeding the preset threshold as abnormal and correct it.

[0073] After obtaining the complete transaction record, calculate the time deviation value of the actual payment time and the agreed payment time. The time deviation value can be represented as the difference in days between the actual payment date and the agreed payment date. If the agreed payment date of a transaction is May 10, 2024, and the actual payment date is May 15, 2024, the time deviation value is positive 5 days, indicating delayed payment; if the actual payment date is May 5, 2024, the time deviation value is negative 5 days, indicating early payment.

[0074] To eliminate the impact of payment habits differences between different enterprises, standardize the time deviation value. Standardization combines centralization and normalization to make the time deviation values of each enterprise distributed on a comparable scale. For all transaction records of Enterprise A, calculate the average value and standard deviation of the time deviation value, then subtract the average value from each transaction's time deviation value and divide by the standard deviation to get the standardized time deviation value. After standardization, the payment behavior characteristics of enterprises can be compared horizontally.

[0075] Next, calculate the correlation coefficient between the standardized time deviation value and the enterprise's historical default records. Historical default records can be obtained from credit rating agencies or financial institutions, including overdue payments, and supplier arrears. Use the sliding window technique with a three-month window to calculate the correlation between the fluctuation of the time deviation value and the default event within the window. By using the weighted average method to integrate the correlation coefficients of multiple time windows, the final payment characteristic value is generated. The higher the payment characteristic value, the stronger the correlation between the enterprise's payment behavior and historical default risk.

[0076] Based on the payment characteristic value and the transaction association degree of enterprises in the supply chain, construct a payment behavior matrix. The payment behavior matrix is a two-dimensional matrix, with rows and columns representing enterprises in the supply chain, and matrix elements representing the payment behavior characteristics between the two enterprises. The transaction association degree can be measured by transaction frequency, transaction amount proportion, etc. If the annual transaction amount between Enterprise A and Enterprise B accounts for 30% of Enterprise A's total transaction amount, the transaction association degree between them is high; if it only accounts for 1%, the association degree is low. Multiply the transaction association degree by the payment characteristic value to fill the corresponding position of the payment behavior matrix.

[0077] The payment behavior fluctuation value is obtained by weighted calculation of the payment characteristic value between adjacent enterprises in the payment behavior matrix. The weighted calculation considers the transmission relationship between enterprises in the supply chain network, and a graph network analysis method is used to identify key node enterprises. For enterprises in a key position in the supply chain, the change of their payment behavior will have a greater impact on the entire supply chain, and therefore a higher weight is given. The payment behavior fluctuation value is also considered in terms of seasonal factors and industry periodicity, and the regular fluctuations are removed by time series decomposition technology, and the abnormal fluctuation component is retained.

[0078] Finally, the payment behavior sequence is constructed according to the payment behavior fluctuation value in time sequence. The payment behavior fluctuation values of the last six months are organized into a time sequence, and an adaptive threshold detection method is used to identify abnormal points and trends in the sequence. If the payment behavior fluctuation value shows an upward trend for three consecutive months and exceeds the industry average, it is determined to be a potential risk signal; if the payment behavior fluctuation value shows a sudden and large increase, it may indicate that the enterprise is facing urgent financial pressure. The payment behavior sequence is visualized as a trend curve, which intuitively reflects the change trajectory and risk accumulation process of the enterprise's payment behavior.

[0079] Through in-depth analysis of supply chain payment behavior data, early identification and early warning of enterprise risk are realized. By analyzing the structure of the supply chain network, the risk transmission path is identified to prevent single-point risk from evolving into systemic risk; breaking down information silos, achieving information sharing and collaborative risk control between upstream and downstream enterprises; reducing data dependency, even in the case of incomplete disclosure of enterprise financial data, effective risk assessment can be made based on transaction behavior; strong adaptability, can automatically adjust the risk judgment standard according to the industry characteristics and market environment, reduce the misjudgment rate, and improve the accuracy of risk control.

[0080] The fluctuation amplitude is calculated based on the payment behavior sequence, the supply chain collaborative credit rating is calculated based on the fluctuation amplitude and historical default records, and when the supply chain collaborative credit rating continuously decreases, the corresponding enterprise is marked as a risk source including:

[0081] Obtain the payment behavior sequence of the supply chain enterprise, and use sliding variance test to identify the mutation point of the time interval sequence;

[0082] Divide the payment behavior sequence into multiple time periods with the mutation point as the boundary, and calculate the payment behavior fluctuation amplitude in each time period;

[0083] Identify abnormal payment time points according to the deviation degree of the payment behavior fluctuation amplitude, and obtain the historical default records of the transaction enterprise corresponding to the abnormal payment time points;

[0084] Calculate the change trend and correlation coefficient of the payment behavior fluctuation amplitude and the historical default records respectively, and take the correlation coefficient as the supply chain collaborative credit rating;

[0085] Monitor the trend of supply chain collaborative credit rating changes, and mark the corresponding enterprise as a risk source when the supply chain collaborative credit rating continues to decline.

[0086] The payment behavior sequence is extracted from the historical transaction data of the enterprise, including the payment time, amount, payment object, and other key information of each transaction. These data can be obtained through enterprise resource planning systems or supply chain management platforms. For raw data, data cleaning and preprocessing are required, including removing outliers, filling missing values, and standardizing processing. The payment behavior sequence is usually plotted with time as the horizontal axis, recording the payment activities of the enterprise at each time point, forming time series data. In the payment behavior sequence of Enterprise A in the past half year, the payment information between the enterprise and its upstream and downstream enterprises is recorded, including payment date, amount, deviation from the agreed payment date, and other data items.

[0087] The sliding variance test is used to identify the mutation points in the time interval sequence. The time interval sequence refers to the sequence of time differences between adjacent payment behaviors. The sliding variance test calculates the local variance within a time window and slides the time window to detect abnormal fluctuations in the sequence. An appropriate size of the time window is selected, usually 30 days, and the sliding step is 1 day. The variance of the payment time interval within the window is calculated. When the sliding variance value at a certain time point is significantly higher or lower than the historical average level, the point is identified as a mutation point. To improve the detection accuracy, a double-threshold judgment mechanism is used, setting the upper threshold and the lower threshold. The upper threshold is usually twice the average value of the historical variance, and the lower threshold is 0.5 times the average value of the historical variance.

[0088] The payment behavior sequence is divided into multiple time periods based on the mutation points, and the payment behavior fluctuation amplitude in each time period is calculated. The fluctuation amplitude calculation is based on the statistical characteristics of the payment time deviation, including the mean, standard deviation, and extreme value range of the time deviation. The time deviation refers to the difference between the actual payment time and the agreed payment time. For each divided time period, the average value of the time deviation of all transactions in the period is calculated, and the standard deviation of the time deviation is calculated as a quantitative indicator of the fluctuation amplitude. A larger standard deviation indicates that the payment behavior is unstable and may have potential risks.

[0089] According to the deviation degree of the payment behavior fluctuation amplitude, an abnormal payment time point is identified, and the historical default record of the transaction enterprise corresponding to the abnormal payment time point is obtained. The deviation degree is calculated by Z-score, that is, the fluctuation amplitude of each time point is subtracted from the historical average fluctuation amplitude, and then divided by the standard deviation of the historical fluctuation amplitude. The absolute value greater than 2 is regarded as a significant deviation, and the corresponding time point is marked as an abnormal payment time point. For the identified abnormal payment time point, the transaction enterprise related to the time point is tracked, and the historical default record of these enterprises is obtained from the credit rating agency, financial institution or internal risk control system. The historical default record includes overdue payment, refusal to pay, delayed delivery and other credit events.

[0090] The change trend and correlation coefficient of the payment behavior fluctuation amplitude and the historical default record are calculated respectively, and the correlation coefficient is taken as the supply chain collaborative credit rating. The change trend is determined by time series analysis method, including moving average, trend decomposition and other techniques. The correlation coefficient is calculated by Pearson correlation coefficient, and the payment behavior fluctuation amplitude sequence and the historical default record sequence are paired for analysis. The correlation coefficient takes value in the range of-1 to 1, positive value represents positive correlation, negative value represents negative correlation, and absolute value represents correlation strength. High correlation coefficient indicates that the payment behavior fluctuation is highly correlated with credit risk, and the corresponding collaborative credit rating is low. In order to adapt to different industry characteristics, an industry adjustment factor is introduced, and the correlation coefficient is corrected according to the industry average payment period and default rate. The correlation coefficients of the payment behavior fluctuation amplitude and the historical default record of enterprise A and its main transaction partners enterprise B and enterprise C are 0.78 and 0.62 respectively, and the calculated supply chain collaborative credit rating is 0.7, which is at a medium risk level.

[0091] The change trend of the supply chain collaborative credit rating is monitored, and when the supply chain collaborative credit rating continuously decreases, the corresponding enterprise is marked as a risk source. The change trend monitoring adopts time series analysis method, including setting benchmark period, comparison period and warning threshold. The benchmark period is usually three months, and the comparison period is one month. When the collaborative credit rating in the comparison period decreases by more than the warning threshold compared with the benchmark period, the risk warning is triggered. The warning threshold is set to a decrease of 15%. Continuous decrease is defined as the collaborative credit rating showing a downward trend in at least two consecutive monitoring periods. In order to improve the accuracy of early warning, a multi-dimensional risk verification mechanism is introduced, which considers external factors such as market environment and industry cycle. For the enterprise marked as a risk source, a risk level classification is established, including mild risk, moderate risk and severe risk. Different risk levels correspond to different control measures, such as strengthening monitoring, adjusting credit limit, requiring guarantee, etc. In the monitoring of the collaborative credit rating of enterprise A, it is found that its rating decreases from 0.85 to 0.7, and then to 0.55 in three consecutive months, and the continuous decrease amplitude exceeds the warning threshold, so enterprise A is marked as a moderate risk source.

[0092] By analyzing the fluctuation characteristics of the payment behavior sequence, early signals of abnormal business operations of enterprises can be accurately identified, and the preposition of risk warning can be realized. Based on the sliding variance test mechanism, the identification accuracy of the payment behavior mutation point is improved, and the misjudgment rate is reduced. By establishing a correlation model between payment behavior and credit risk, quantitative assessment of potential risks is realized. Using the structure of the supply chain network, the risk transmission path can be tracked to prevent the spread of risks in the supply chain. Multi-dimensional risk verification is supported to reduce the influence of external environmental interference factors and improve the adaptability of risk judgment.

[0093] The payment behavior sequence of the risk source enterprise is mapped to a multi-dimensional feature space by using recursive partitioning time windows, and the periodic indicators and mutation indicators of the payment behavior are extracted to construct the payment behavior portrait, including:

[0094] The change characteristics of adjacent time points in the payment behavior sequence of the risk source enterprise are extracted to construct a feature matrix, and the fluctuation direction cumulative value is calculated based on the feature matrix to generate a fluctuation curve.

[0095] The fluctuation cumulative value of the trend inflection point in the fluctuation curve is calculated, and the fluctuation cumulative value is compared with the historical fluctuation threshold to determine the candidate partition position.

[0096] A time window is constructed at the candidate partition position, and the window boundary is determined based on the aggregation degree distribution characteristics of the payment behavior in the time window.

[0097] The time window boundary is recursively partitioned to obtain time sequence segments, and adjacent segments are merged according to the similarity of the payment behavior change law in the time sequence segments.

[0098] A transition window is constructed at the boundary of the merged segments, the fluctuation slope and dispersion of the payment behavior in the transition window are calculated, the range of the transition window is adjusted, and the recursive partitioning time window is obtained.

[0099] The time dimension features, amount dimension features and frequency dimension features of the payment behavior in the recursive partitioning time window are extracted to construct a multi-dimensional feature space, and the periodic indicators and mutation indicators of the payment behavior are extracted in the multi-dimensional feature space to construct the payment behavior portrait.

[0100] First, the change features of adjacent time points in the payment behavior sequence of the risk source enterprise are extracted to construct a feature matrix. The fluctuation direction cumulative value is calculated based on the feature matrix to generate a fluctuation curve. The payment behavior sequence of the risk source enterprise contains key attributes such as transaction time, transaction amount, and payment object. The change features are calculated by the difference of attribute values at adjacent time points, mainly including three dimensions of time interval change, amount change ratio, and transaction object stability. The time interval change represents the time difference between two consecutive transactions, the amount change ratio represents the relative change degree of adjacent transaction amounts, and the transaction object stability represents the change frequency of transaction partners. The feature matrix adopts a three-dimensional structure, with the horizontal axis as the time sequence and the vertical axis as the change feature type, and the matrix elements are the corresponding feature values. The fluctuation direction cumulative value calculation is to accumulate the change features at each time point according to the increase, decrease, or remain unchanged, with the increase recorded as 1, the decrease recorded as -1, and the unchanged recorded as 0. By continuously summing the cumulative values, a fluctuation curve is drawn, which intuitively shows the change trend of the payment behavior.

[0101] The fluctuation cumulative value of the trend inflection point in the fluctuation curve is calculated, and the fluctuation cumulative value is compared with the historical fluctuation threshold to determine the candidate partition position. The trend inflection point refers to the turning point where the fluctuation curve changes from rising to falling or from falling to rising, which is identified by calculating the first-order difference of the curve and finding the point where the difference value changes from positive to negative or from negative to positive. The fluctuation cumulative value is the cumulative direction value at the inflection point, reflecting the intensity of the change in payment behavior before the inflection point. The historical fluctuation threshold is calculated based on the payment behavior data of the enterprise in the past 12 months, and usually takes 2 times the standard deviation of the historical fluctuation cumulative value as the abnormal judgment standard. When the fluctuation cumulative value of a certain inflection point exceeds the historical fluctuation threshold, the inflection point is marked as a candidate partition position.

[0102] A time window is constructed at the candidate partition position, and the window boundary is determined based on the aggregation degree distribution characteristics of the payment behavior within the time window. The initial width of the time window is set to 15 days before and after the candidate partition position, totaling a time range of 30 days. The aggregation degree distribution characteristics of the payment behavior are measured by calculating the time density and attribute similarity of the transactions within the window. The time density refers to the number of transactions per unit time, and the attribute similarity refers to the relative stability of transaction attributes. The clustering algorithm is used to identify the aggregation area of the payment behavior within the window, and the density steep drop point at the edge of the aggregation area is considered as the potential window boundary. The window boundary determination process adopts an adaptive adjustment strategy, that is, the window range is automatically expanded or contracted according to the aggregation degree distribution characteristics until the optimal boundary is reached.

[0103] The time window boundary is recursively divided to obtain the time series segment, and adjacent segments are merged according to the similarity of the change rule of payment behavior in the time series segment. The recursive division adopts dichotomy, finds the point with the most significant change of payment behavior in each time window as the division point, divides the window into two sub-windows, and then repeats the process for the sub-windows until the size of the sub-window is less than the preset minimum threshold or the payment behavior in the sub-window tends to be consistent. The change significance is determined by variance analysis of indicators such as payment interval and amount fluctuation. The time window from the 105th day to the 135th day is recursively divided to obtain three time series segments: the 105th-115th day, the 116th-125th day, and the 126th-135th day. The adjacent segment merging is based on the similarity evaluation of the change rule of payment behavior, and the similarity calculation comprehensively considers factors such as payment frequency, amount distribution, and time rule. When the similarity of adjacent segments exceeds the preset threshold (usually 0.85), these segments are merged into a larger segment. The similarity calculation uses the cosine similarity method to compare the feature vectors of each segment. Through similarity analysis, the 105th-115th day and the 116th-125th day are merged into a new segment (the 105th-125th day), and finally two time series segments are obtained: the 105th-125th day and the 126th-135th day.

[0104] A transition window is constructed at the boundary of the merged segment, the fluctuation slope and dispersion of payment behavior in the transition window are calculated, the range of the transition window is adjusted, and the recursive partition time window is obtained. The initial range of the transition window is set to 5 days before and after the boundary of the merged segment, i.e. the 120th-130th day. The fluctuation slope represents the rate of change of payment behavior, and the trend slope of payment features in the transition window is calculated by the linear regression method. The dispersion represents the instability of payment behavior, which is measured by calculating the ratio of the standard deviation to the mean of the payment features in the transition window. According to the comprehensive score of the fluctuation slope and the dispersion, the range of the transition window is adaptively adjusted. If the fluctuation slope is large and the dispersion is high, it indicates that the change is violent and the transition window needs to be reduced; if the fluctuation slope is small and the dispersion is low, it indicates that the change is gentle and the transition window can be expanded. Through iterative optimization, the transition window is adjusted to the 123rd-127th day, forming the final recursive partition time window: the 105th-123rd day, the 123rd-127th day (transition window), and the 127th-135th day.

[0105] The time dimension feature, the amount dimension feature and the frequency dimension feature of the payment behavior in the recursive partition time window are extracted to construct a multi-dimensional feature space, the periodic index and the mutation index of the payment behavior are extracted in the multi-dimensional feature space, and the payment behavior portrait is constructed. The time dimension feature includes the average payment interval, the regularity of the payment time and the degree of deviation from the agreed time; the amount dimension feature includes the average transaction amount, the amount change amplitude and the amount distribution characteristics; the frequency dimension feature includes the number of transactions per unit time, the stability of the frequency and the seasonal variation. The payment behavior in each recursive partition time window is extracted to construct a multi-dimensional feature vector to form a feature space. In the feature space, the periodic characteristics of the payment behavior are identified by autocorrelation analysis, and the mutation characteristics are identified by variable point detection. The periodic index reflects the regularity of the enterprise payment behavior, including the main cycle length, the cycle stability and the cycle intensity; the mutation index reflects the abnormal change of the enterprise payment behavior, including the mutation frequency, the mutation amplitude and the mutation recovery time. For example, the analysis result of the electronic manufacturing industry supplier shows that in the time window of 105-123 days, the payment behavior presents a stable cycle of 30 days, while in the time window of 127-135 days, the payment cycle is shortened to 15 days, and the mutation frequency increases significantly, indicating that the enterprise payment strategy has changed significantly. The payment behavior portrait of the enterprise is constructed by integrating the features of the three dimensions of time, amount and frequency, including the normal payment mode, the abnormal payment trigger condition and the risk warning index.

[0106] By deeply analyzing the time series characteristics of the enterprise payment behavior, the payment behavior portrait is constructed, which provides effective technical support for the supply chain risk early warning. Through multi-dimensional feature extraction and recursive partition technology, the sensitive capture of the small changes of the payment behavior is realized; through the construction of the payment behavior portrait, the enterprise-specific risk identification model is formed, which improves the accuracy of risk judgment; the dynamic updating mechanism of the payment behavior portrait enables the risk control model to adapt to the changes of the enterprise behavior; through the correlation analysis of the payment behavior of the upstream and downstream enterprises, the collaborative warning of the overall risk of the supply chain is realized, and the spread and conduction of the risk in the supply chain are prevented.

[0107] The enterprise payment behavior trajectory is constructed in the multi-dimensional feature space, the associated enterprises are identified by the phase overlap degree of the payment behavior trajectory, and the risk conduction order is formed based on the synchronous change intensity, including:

[0108] The amount change and the frequency change of adjacent time points in the enterprise payment behavior sequence are extracted to construct a change feature matrix, and the change feature matrix is mapped to a multi-dimensional feature space to construct a payment behavior trajectory;

[0109] An initial time window is constructed for the payment behavior trajectory, the fluctuation amplitude of the trajectory in the initial time window is calculated, and the initial time window length is dynamically adjusted according to the fluctuation amplitude;

[0110] In the adjusted time window, the angle change sequence of different enterprise payment behavior trajectories is calculated, and the phase overlap degree is obtained based on the angle change sequence;

[0111] The fluctuation feature of the phase overlap degree is extracted, the fluctuation cumulative value is calculated, the fluctuation cumulative value is compared with the dynamic threshold value calculated based on historical data, the continuous overlap interval is identified, and the associated enterprise is identified based on the continuous overlap interval;

[0112] The change synchronization of the payment behavior trajectory of the associated enterprise in the continuous overlap interval is calculated, the synchronization change intensity is obtained, and the risk transmission sequence is formed based on the synchronization change intensity.

[0113] First, the amount change and frequency change of adjacent time points in the enterprise payment behavior sequence are extracted, a change feature matrix is constructed, and the payment behavior trajectory is constructed by mapping the change feature matrix to a multi-dimensional feature space. The enterprise payment behavior sequence refers to a data set formed by arranging all payment transactions of an enterprise in a certain period of time in chronological order, including transaction time, transaction amount, transaction object and other key attributes. The amount change refers to the difference between the transaction amounts of two adjacent time points, which can be represented as the ratio of the current transaction amount to the last transaction amount minus 1, a positive value indicating an increase and a negative value indicating a decrease. The frequency change refers to the change in the number of transactions per unit time, which is calculated by calculating the difference between the number of transactions in adjacent windows. For example, the payment behavior analysis of a certain enterprise shows that it has 180 transaction records in six months, the transaction amount is between 50,000 yuan and 500,000 yuan, and the average weekly transaction frequency is 7.5 times. The change feature matrix is constructed in a two-dimensional structure, with the horizontal axis representing the time sequence and the vertical axis representing the change feature type, and the matrix elements representing the corresponding feature values. When the change feature matrix is mapped to a multi-dimensional feature space, the feature vector of each time point is a point in the space, and the payment behavior trajectory is formed by connecting these points. The multi-dimensional feature space includes the amount change dimension, the frequency change dimension, the time interval dimension, etc., which is processed by principal component analysis method for dimension reduction, and the three principal components with the highest contribution rate are selected to construct a three-dimensional feature space. The payment behavior trajectory of this automobile parts enterprise presents a relatively stable spiral structure, reflecting the periodic characteristics of its payment behavior.

[0114] An initial time window is constructed for the payment behavior trajectory, the fluctuation amplitude of the trajectory within the initial time window is calculated, and the initial time window length is dynamically adjusted according to the fluctuation amplitude. The initial time window is set to 30 days, and the sliding step is 1 day. The fluctuation amplitude calculation is based on the displacement characteristics of the trajectory in the feature space, specifically including the ratio of the trajectory length to the straight-line distance, the curvature change of the trajectory, and the like. The trajectory length refers to the sum of the distances between all adjacent points on the trajectory, and the straight-line distance refers to the Euclidean distance between the starting point and the ending point of the trajectory. The greater the fluctuation amplitude, the more intense the payment behavior change. The dynamic adjustment of the time window length adopts an adaptive strategy, when the fluctuation amplitude exceeds the preset threshold, the window length is shortened to capture rapid changes; when the fluctuation amplitude is lower than the preset threshold, the window length is extended to obtain a more stable trend. The preset threshold is determined based on the fluctuation statistical characteristics of the historical payment behavior of the enterprise, and is usually the mean value of the historical fluctuation amplitude plus one standard deviation. For the payment behavior trajectory analysis of an enterprise, the fluctuation amplitude within the initial 30-day window is 1.85, which is higher than the preset threshold of 1.5, so the window length is adjusted to 20 days, and the fluctuation amplitude is recalculated to 1.42, which is lower than the preset threshold, and the final window length is determined to be 20 days.

[0115] Within the adjusted time window, the angle change sequence of different enterprise payment behavior trajectories is calculated, and the phase overlap degree is obtained based on the angle change sequence. Select the automobile parts enterprise and its upstream and downstream enterprises, including a main raw material supplier and two downstream vehicle manufacturers, a total of four enterprises constitute the analysis object. The angle change sequence calculation is to calculate the angle between the tangent vectors of two adjacent time points of the enterprise payment behavior trajectory in the feature space, reflecting the degree of change in the trajectory direction. The tangent vector is calculated by the central difference method of three adjacent time points. For each pair of enterprise combination, the angle change sequence of its payment behavior trajectory is calculated, a total of six pairs of combinations. The phase overlap degree is defined as the similarity of the angle change sequence, which is obtained by calculating the cross-correlation function of the angle change sequence. The phase overlap degree takes a value ranging from 0 to 1, and the larger the value, the more synchronous the payment behavior change of the two enterprises. The phase overlap degree of the automobile parts enterprise and its raw material supplier is 0.82, and the phase overlap degrees with the two downstream vehicle manufacturers are 0.65 and 0.73 respectively, indicating that the payment behavior change of the enterprise and its upstream supplier is more synchronous.

[0116] The fluctuation characteristics of the phase overlap degree are extracted, a fluctuation cumulative value is calculated, the fluctuation cumulative value is compared with a dynamic threshold value calculated based on historical data, a continuous overlap interval is identified, and associated enterprises are identified based on the continuous overlap interval. The fluctuation characteristics of the phase overlap degree are extracted by a time series analysis method, including moving average, volatility calculation and other techniques. The fluctuation cumulative value refers to the cumulative duration that the phase overlap degree is continuously higher than the average level, and reflects the duration of the synchronization of payment behavior. The dynamic threshold value is calculated based on the statistical distribution characteristics of the historical data, and an adaptive percentile method is used, usually taking the 90th percentile of the historical phase overlap degree distribution as the threshold reference. The continuous overlap interval refers to the time interval during which the phase overlap degree is continuously higher than the dynamic threshold value, and represents a period of high synchronization of payment behavior between enterprises. When the continuous overlap interval exceeds a preset minimum duration, it is considered that there is a significant association between the corresponding enterprises.

[0117] The change synchronization of payment behavior trajectories of associated enterprises in the continuous overlap interval is calculated to obtain a synchronization change intensity, and a risk transmission sequence is formed based on the synchronization change intensity. The change synchronization is evaluated by a Granger causality test method to analyze the lead-lag relationship of payment behavior changes between enterprises. A vector autoregressive model containing payment behavior characteristics of two enterprises is constructed, and the transmission direction of the change is determined by a significance test of the lag coefficient. The synchronization change intensity is defined as the statistical significance level of the Granger causality relationship, and reflects the influence degree of the payment behavior change. The identified associated enterprises are analyzed for change synchronization, for example, the Granger test result of a certain parts enterprise and a raw material supplier shows that the payment behavior change of the supplier leads the parts enterprise, the significance level is 0.01, and the synchronization change intensity is 0.95; and the test result with a downstream manufacturer shows that the payment behavior change of the parts enterprise leads the downstream manufacturer, the significance level is 0.05, and the synchronization change intensity is 0.87. Based on the synchronization change intensity and the transmission direction, a risk transmission sequence is formed: raw material supplier → parts enterprise → downstream manufacturer. The result shows that the payment behavior anomaly of the raw material supplier appears first, then is transmitted to the parts enterprise, and finally affects the vehicle manufacturer, forming a complete risk transmission link.

[0118] By in-depth analysis of the time sequence characteristics and spatial relationships of enterprise payment behavior trajectories, the identification of the supply chain risk transmission path is realized. Through multi-dimensional feature space mapping and payment behavior trajectory construction, the complex changes of enterprise payment behavior are intuitively expressed, and the perception ability of small change trends is improved; through phase overlap degree analysis and continuous overlap interval identification, the synchronization change characteristics of payment behavior between enterprises are accurately captured, and the misjudgment rate of associated enterprise identification is reduced; by using the Granger causality test technique, the transmission direction of the payment behavior change is scientifically determined, and a time sequence for risk early warning is provided.

[0119] The risk prevention and control scheme including the credit limit control instruction and the behavior monitoring strategy is generated by using the evolution law of the payment behavior portrait in the multi-dimensional feature space, combining the risk transmission sequence, predicting the risk trigger node, and including the credit limit control instruction and the behavior monitoring strategy.

[0120] The time sequence features of the payment behavior portrait are extracted in the multi-dimensional feature space, the change trend of the time sequence features in adjacent time windows is calculated, and the behavior abnormal pattern is identified.

[0121] The payment behavior features of the associated enterprises are extracted based on the behavior abnormal pattern, the divergence degree and the aggregation degree of the payment behavior features are calculated, and when the divergence degree exceeds the preset degree threshold, the corresponding time point is marked as a risk trigger node.

[0122] The risk trigger node is backtracked, the evolution trajectory segment before triggering is extracted, the cumulative intensity of the feature change in the evolution trajectory segment is analyzed, and the risk outbreak time window is determined.

[0123] The influence range in the risk outbreak time window is calculated in combination with the position of the enterprise in the risk transmission sequence, and the credit limit control instruction is set.

[0124] The behavior monitoring strategy is generated according to the credit limit control instruction, and the risk prevention and control scheme including the credit limit control instruction and the behavior monitoring strategy is output.

[0125] Firstly, the time sequence features of the payment behavior portrait are extracted in the multi-dimensional feature space, the change trend of the time sequence features in adjacent time windows is calculated, and the behavior abnormal pattern is identified. The payment behavior portrait is a feature set of the enterprise payment behavior, including time dimension features, amount dimension features and frequency dimension features. The time dimension features mainly include average payment interval, payment time regularity and deviation from the agreed time degree; the amount dimension features mainly include average transaction amount, amount fluctuation amplitude and amount distribution characteristics; the frequency dimension features mainly include the number of transactions per unit time, frequency stability and seasonal variation. The time sequence feature extraction adopts the sliding window method, and appropriate window length and sliding step are set. In each window, the statistics of each dimension feature are calculated, including mean, standard deviation, skewness, kurtosis, etc., to form a feature vector. The change trend calculation is to calculate the difference between the feature vectors of adjacent two time windows to obtain a trend vector. The direction and size of the trend vector reflect the direction and intensity of the payment behavior change. The behavior abnormal pattern recognition is based on the anomaly detection of the trend vector, and the local anomaly factor algorithm is adopted to calculate the relative density of each trend vector and its neighborhood trend vector to judge its abnormal degree. When the abnormal degree exceeds the preset threshold, it is identified as a behavior abnormal pattern.

[0126] Based on the behavior anomaly pattern, the payment behavior characteristics of associated enterprises are extracted, the divergence degree and the aggregation degree of the payment behavior characteristics are calculated, and when the divergence degree exceeds the preset degree threshold, the corresponding time point is marked as a risk trigger node. The associated enterprise refers to the upstream and downstream enterprises that have direct transaction relationship with the target enterprise in the supply chain, which is identified through transaction records and enterprise relationship graph. The payment behavior characteristics of the associated enterprise are extracted using the same feature extraction method as the target enterprise to obtain the feature vector of the same period. The divergence degree calculation measures the difference between the feature vectors of the payment behavior of the associated enterprises, and the standard deviation of the feature vector is used for calculation. The larger the standard deviation, the higher the divergence degree, and the more significant the difference in payment behavior of the associated enterprises. The aggregation degree calculation measures the concentration degree of the feature vector of the payment behavior of the associated enterprises, and the average distance of the feature vector is used for calculation. The smaller the average distance, the higher the aggregation degree, and the more similar the payment behavior of the associated enterprises. The preset degree threshold is determined based on the statistical distribution of historical data, and is usually taken as the higher percentile of the historical divergence degree. When the divergence degree of a certain time point exceeds the preset threshold, the time point is marked as a risk trigger node.

[0127] The risk trigger node is backtracked for features, the evolution trajectory segment before triggering is extracted, the cumulative intensity of feature change in the evolution trajectory segment is analyzed, and the risk outbreak time window is determined. Feature backtracking refers to tracking from the risk trigger node to the front to analyze the evolution process of the payment behavior characteristics. The backtracking time range is usually set to a period of time before the risk trigger node. The evolution trajectory segment refers to the trajectory formed by the payment behavior characteristic vector changing with time in the backtracking period. The cumulative intensity of feature change is obtained by calculating the cumulative sum of the distance between adjacent points on the trajectory, which reflects the overall intensity of payment behavior change. The inflection point analysis is performed on the cumulative intensity curve to identify the time point where the slope of the curve changes significantly, which is regarded as the key turning point of risk evolution. The risk outbreak time window is defined as the time interval from the key turning point to the risk trigger node. By analyzing the change characteristics of the cumulative intensity curve, the key stage of risk evolution can be accurately located to provide time guidance for risk prevention and control.

[0128] The influence range within the risk outbreak time window is calculated according to the position of the enterprise in the risk transmission sequence, and a credit quota control instruction is set. The risk transmission sequence refers to the path and order of risk transmission in the supply chain, which has been determined through previous analysis. The influence range calculation is based on a network transmission model, considering the connection strength between enterprises, the transaction amount proportion, and the historical collaborative volatility. The connection strength is calculated by the transaction frequency between enterprises, the transaction amount proportion refers to the proportion of the transaction amount of both parties in the total transaction amount of the enterprise, and the historical collaborative volatility is measured by the correlation of past payment behavior. The influence range is divided into three levels: directly affected enterprises, secondary affected enterprises, and indirectly affected enterprises. The credit quota control instruction is set according to the influence range and risk level, including the credit quota adjustment ratio, adjustment period, and adjustment conditions. Different credit quota adjustment strategies are set for directly affected enterprises, secondary affected enterprises, and indirectly affected enterprises, forming a gradient prevention and control system.

[0129] According to the credit quota control instruction, a behavior monitoring strategy is generated, and a risk prevention and control scheme containing the credit quota control instruction and the behavior monitoring strategy is output. The behavior monitoring strategy includes monitoring targets, monitoring frequency, monitoring indicators, and warning conditions. The monitoring target refers to the enterprise that needs to be focused on, including the risk source enterprise and its directly affected enterprises. The monitoring frequency is set according to the risk level, with high-risk enterprises having a high monitoring frequency and low-risk enterprises having a low monitoring frequency. The monitoring indicators include key features of payment behavior and abnormal behavior patterns, such as payment delay rate, large transaction proportion, transaction object change frequency, etc. The warning condition is the threshold setting that triggers the warning, which is usually determined based on the statistical distribution of historical data. The risk prevention and control scheme integrates the credit quota control instruction and the behavior monitoring strategy to form a complete risk response scheme. The scheme content includes five parts: risk overview, impact assessment, control measures, monitoring strategy, and emergency plan. The risk overview describes the risk type, risk level, and risk source; the impact assessment analyzes the risk influence range and degree; the control measures specify the credit quota adjustment scheme for each enterprise; the monitoring strategy details the monitoring content and warning mechanism; the emergency plan specifies the rapid response measures when the risk intensifies.

[0130] Through in-depth analysis of the time series characteristics of enterprise payment behavior portrait, accurate identification and effective prevention and control of supply chain risk are realized. Through multi-dimensional feature space construction and time series analysis, the sensitive capture of small changes in enterprise payment behavior is realized, greatly improving the early risk identification ability; through feature backtracking and evolution trajectory analysis, the risk accumulation process is revealed, providing a time window for risk intervention; combined with risk transmission sequence and influence range analysis, accurate risk tracing and transmission prediction are realized, preventing the risk from spreading in the supply chain; the dual protection mechanism of credit quota control instruction and behavior monitoring strategy; can automatically adjust the prevention and control strategy according to the risk evolution, adapt to the complex and variable supply chain environment.

[0131] In the adjusted time window, the angle change sequence of different enterprise payment behavior trajectories is calculated, and the phase overlap degree is obtained based on the angle change sequence, including:

[0132] Obtain the payment behavior trajectory vector sequence in the adjusted time window;

[0133] Decouple the payment behavior trajectory vector sequence into instantaneous amplitude and instantaneous phase, and identify the dynamic boundary point based on the mutation degree of the instantaneous amplitude and the instantaneous phase;

[0134] According to the dynamic boundary point, the payment behavior trajectory vector sequence is divided into multiple adaptive segments, and the angle change amount of adjacent payment behavior trajectory vectors is calculated in the adaptive segment to form an angle change sequence;

[0135] Multi-scale decomposition is performed on the angle change sequence to identify the mutation position, and the mutation position and the angle change sequence are combined to generate an enhanced angle change sequence;

[0136] Nonlinear dynamic normalization is performed on the enhanced angle change sequence of any two enterprises, and the phase overlap degree is calculated based on the time sequence matching degree of the mutation position in the normalization process.

[0137] Obtain the payment behavior trajectory vector sequence in the adjusted time window. The payment behavior trajectory vector sequence refers to the representation of the payment behavior characteristics of an enterprise in a specific time window in a multi-dimensional feature space. The multi-dimensional feature space includes but is not limited to payment amount, payment frequency, payment time interval, etc. The adjusted time window refers to the time range optimized through previous analysis, usually 30 to 90 days. The process of obtaining the payment behavior trajectory vector sequence includes three links of data collection, feature extraction and vector construction. In the data collection stage, the original transaction data of the enterprise is collected, including transaction time, transaction amount, transaction object and other key information. In the feature extraction stage, the feature values of each dimension are calculated based on the original transaction data, such as the average transaction amount, transaction frequency, standard deviation of transaction time interval, etc. In the vector construction stage, the feature values of each dimension are combined to form a feature vector, which is arranged in time sequence to form a trajectory vector sequence. For manufacturing enterprises in the middle of the supply chain, the payment behavior trajectory vector sequence reflects the dynamic changes of the transaction behavior of the enterprise with upstream and downstream enterprises.

[0138] The payment behavior trajectory vector sequence is decoupled into instantaneous amplitude and instantaneous phase, and a dynamic boundary point is identified based on the mutation degree of the instantaneous amplitude and the instantaneous phase. The decoupling process adopts the Hilbert-Huang transform method, and the payment behavior trajectory vector sequence is regarded as a complex signal and is decomposed into two components of instantaneous amplitude and instantaneous phase. The instantaneous amplitude represents the intensity of the change in payment behavior, and the instantaneous phase represents the state of the change in payment behavior. The trajectory vector sequence is normalized to eliminate the dimensional influence; an analytic signal is constructed, and the orthogonal component of the original signal is obtained through Hilbert transform; and the instantaneous amplitude and the instantaneous phase are calculated based on the original signal and the orthogonal component thereof. The mutation degree calculation adopts the sliding variance analysis method, and the variances of the instantaneous amplitude and the instantaneous phase are calculated in the sliding window. The greater the variance value is, the higher the mutation degree is. The dynamic boundary point identification is based on the mutation degree over-threshold method, and a proper threshold is set. When the mutation degree of the instantaneous amplitude or the instantaneous phase exceeds the threshold, the corresponding time point is marked as a dynamic boundary point. The threshold setting usually adopts an adaptive method, and is determined based on the statistical distribution of historical data, such as taking the 90th percentile of the historical mutation degree.

[0139] The payment behavior trajectory vector sequence is divided into multiple adaptive segments according to the dynamic boundary points, and the angle change amount of adjacent payment behavior trajectory vectors is calculated in the adaptive segments to form an angle change sequence. The adaptive segment refers to the division of the trajectory vector sequence into multiple continuous and non-overlapping subsequences according to the dynamic boundary points, and each subsequence is called a segment. The segmentation process considers the reliability of the boundary points, and adjacent boundary points are merged to avoid over-segmentation. The angle change amount calculation is to calculate the angle between adjacent trajectory vectors in each adaptive segment, which reflects the change degree of the payment behavior direction. The angle calculation adopts the vector inner product method, that is, the angle is obtained by taking the inverse cosine of the product of the inner product of two vectors and their module lengths. The angle change amount refers to the difference between the angles of two adjacent vectors, forming an angle change sequence. The angle change sequence directly reflects the fluctuation of the payment behavior in the direction, and the greater the change amount is, the more intense the change in payment behavior is. For enterprises in the supply chain, the angle change sequence can reveal the stability and regularity of the payment behavior, providing a basis for identifying abnormal patterns.

[0140] The mutation position is identified by multi-scale decomposition of the angle change sequence, and the enhanced angle change sequence is generated by combining the mutation position with the angle change sequence. The multi-scale decomposition uses wavelet transform method to decompose the angle change sequence into sub-sequences of different frequency scales. The wavelet transform selects Db4 wavelet basis, and the decomposition level is set to 4 layers to obtain an approximate component and four detail components. The mutation characteristics in the signal are identified by analyzing the energy distribution of the detail components. The mutation position identification is based on threshold detection of the detail component coefficients, and the energy threshold is set. When the detail component coefficient exceeds the threshold, the corresponding time point is marked as the mutation position. The threshold setting adopts an adaptive method based on the statistical distribution of the coefficients, such as taking the 95th percentile of the absolute value of the coefficients. The combination of the mutation position and the original angle change sequence adopts a weighted fusion method, which enhances the value of the original sequence at the mutation position to form the enhanced angle change sequence. In the enhancement process, the weight coefficient of the mutation position is usually set to 2 to 3 times, so that the mutation characteristics are more prominent in the sequence. The enhanced angle change sequence retains the change trend of the original sequence while strengthening the mutation characteristics, which is convenient for subsequent similarity analysis.

[0141] The enhanced angle change sequences of any two enterprises are subjected to nonlinear dynamic normalization, and the phase overlap degree is calculated based on the time sequence matching degree of the mutation positions in the normalization process. Nonlinear dynamic normalization is a sequence alignment technique used to calculate the similarity of two time sequences and find their corresponding relationship. The normalization process uses the classic dynamic time normalization algorithm to achieve sequence alignment by constructing a cumulative distance matrix and backtracking the optimal path. The distance metric uses Euclidean distance, and the path constraint uses Sakoe-Chiba bandwidth limitation with a bandwidth of 10% of the sequence length. The normalization result includes the normalization distance and the normalization path, and the normalization distance reflects the overall similarity of the two sequences, and the normalization path reflects the time correspondence of the two sequences. The time sequence matching degree of the mutation positions refers to the corresponding relationship of the mutation positions in the two sequences on the normalization path. The mapping points of the mutation positions in the two sequences on the normalization path are counted, and the time distance between the mapping points is calculated. The smaller the time distance, the closer the time sequence of the mutation. The phase overlap degree is defined as the closeness of the time sequence matching of the mutation positions, which is calculated by the normalized reciprocal of the mapping point time distance, with a value range of 0 to 1. The larger the value, the higher the phase overlap degree. For upstream and downstream enterprises in the supply chain, a higher phase overlap degree indicates that their payment behavior changes are highly synchronized, and there may be a risk transmission relationship.

[0142] By deeply analyzing the phase characteristics of the vector sequence of enterprise payment behavior trajectories, accurate identification of risk correlations among supply chain enterprises was achieved. Dynamic boundary point identification technology based on instantaneous amplitude and phase made time series segmentation more accurate, adapting to the differences in payment behavior among different enterprises; multi-scale decomposition and abrupt change location identification technology enhanced sensitivity to abnormal payment behavior patterns and reduced false alarm rates; nonlinear dynamic warping algorithms overcame the challenge of time series alignment and effectively handled the inconsistency in payment rhythms among different enterprises; the phase overlap index innovatively quantified the collaborative relationship of payment behavior among enterprises, providing a scientific basis for risk transmission analysis.

[0143] like Figure 2 As shown, Figure 2 This is a schematic diagram of the structure of an enterprise upstream and downstream collaborative risk control and prediction system for supply chain data provided in an embodiment of the present invention. The system includes:

[0144] The sequence construction module is used to obtain transaction records between upstream and downstream enterprises in the supply chain, extract payment time deviation values ​​between enterprises, and construct payment behavior sequences.

[0145] The risk source identification module is used to calculate the volatility based on the payment behavior sequence, calculate the supply chain collaborative credit rating based on the volatility and historical default records, and mark the corresponding enterprise as a risk source when the supply chain collaborative credit rating declines continuously.

[0146] The profile building module is used to map the payment behavior sequence of risk source enterprises to a multi-dimensional feature space by recursively partitioning time windows, extracting periodic indicators and mutation indicators of payment behavior, and building a profile of payment behavior.

[0147] The associated enterprise identification module is used to construct enterprise payment behavior trajectories in a multi-dimensional feature space, identify associated enterprises by the phase overlap of the payment behavior trajectories, and form a risk transmission order based on the intensity of synchronous changes.

[0148] The scheme generation module is used to utilize the evolution pattern of payment behavior profiles in a multi-dimensional feature space, combined with the risk transmission sequence, to predict risk triggering nodes and generate risk prevention and control schemes that include credit limit adjustment instructions and behavior monitoring strategies.

[0149] One technical solution provided in this embodiment of the invention is an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in any of the aforementioned methods.

[0150] The embodiment of the present application also provides a computer readable storage medium, and the computer readable storage medium stores a computer program.

[0151] The above-mentioned specific embodiments are the preferred embodiments of the present application, and are not intended to limit the specific implementation range of the present application. The scope of the present application includes but is not limited to the specific embodiments. Any equivalent changes made according to the shape and structure of the present application are within the protection scope of the present application.

Claims

1. An enterprise upstream and downstream collaborative risk control prediction method for supply chain data, characterized in that, The method comprises the following steps: obtaining transaction records between upstream enterprises and downstream enterprises on a supply chain, extracting payment time deviation values between the enterprises, and constructing a payment behavior sequence; calculating a fluctuation amplitude based on the payment behavior sequence, calculating a supply chain coordination credit rating based on the fluctuation amplitude and historical default records, and marking the corresponding enterprise as a risk source when the supply chain coordination credit rating continuously decreases; mapping the payment behavior sequence of the risk source enterprise to a multi-dimensional feature space using a recursive partitioning time window, extracting a periodic index and a mutation index of the payment behavior, and constructing a payment behavior portrait; constructing an enterprise payment behavior trajectory in the multi-dimensional feature space, identifying associated enterprises through the phase overlap degree of the payment behavior trajectory, and forming a risk transmission sequence based on the synchronous change intensity; predicting a risk trigger node using the evolution law of the payment behavior portrait in the multi-dimensional feature space, combining the risk transmission sequence, generating a risk prevention and control scheme including a credit limit control instruction and a behavior monitoring strategy; obtaining transaction records between upstream enterprises and downstream enterprises on a supply chain, extracting payment time deviation values between the enterprises, and constructing a payment behavior sequence comprises: obtaining transaction record information between upstream enterprises and downstream enterprises on a supply chain, the transaction record information including transaction amount and payment time; calculating the time deviation value of the actual payment time and the agreed payment time in the transaction record information, and performing standardization processing; calculating the correlation coefficient of the standardized time deviation value and the historical default records of the enterprise, and generating a payment feature value; constructing a payment behavior matrix based on the payment feature value and the transaction correlation degree of the enterprise in the supply chain, performing weighted calculation on the payment feature values between adjacent enterprises in the payment behavior matrix to obtain a payment behavior fluctuation value, and constructing a payment behavior sequence in time sequence according to the payment behavior fluctuation value; mapping the payment behavior sequence of the risk source enterprise to a multi-dimensional feature space using a recursive partitioning time window, extracting a periodic index and a mutation index of the payment behavior, and constructing a payment behavior portrait comprises: extracting the change characteristics of adjacent time points in the payment behavior sequence of the risk source enterprise to construct a feature matrix, calculating the fluctuation cumulative value of the trend inflection point in the fluctuation curve, and generating a fluctuation curve; calculating the fluctuation cumulative value of the trend inflection point in the fluctuation curve, comparing the fluctuation cumulative value with the historical fluctuation threshold to determine the candidate partition position; constructing a time window at the candidate partition position, determining the window boundary based on the aggregation degree distribution characteristics of the payment behavior in the time window; recursively partitioning the time window boundary to obtain time sequence segments, and merging adjacent segments according to the similarity of the payment behavior change law in the time sequence segments; constructing a transition window at the boundary of the merged segments, calculating the fluctuation slope and dispersion of the payment behavior in the transition window, adjusting the range of the transition window, and obtaining a recursive partitioning time window; extracting the time dimension feature, the amount dimension feature and the frequency dimension feature of the payment behavior in the recursive partitioning time window to construct a multi-dimensional feature space, extracting the periodic index and the mutation index of the payment behavior in the multi-dimensional feature space, and constructing a payment behavior portrait; The payment behavior trajectory of an enterprise is constructed in a multi-dimensional feature space, and the associated enterprises are identified through the phase overlap degree of the payment behavior trajectory, and a risk transmission sequence is formed based on the synchronization change intensity, including: The amount change and frequency change of adjacent time points in the payment behavior sequence of an enterprise are extracted, a change feature matrix is constructed, and the change feature matrix is mapped to a multi-dimensional feature space to construct a payment behavior trajectory; An initial time window is constructed for the payment behavior trajectory, the fluctuation amplitude of the trajectory in the initial time window is calculated, and the length of the initial time window is dynamically adjusted according to the fluctuation amplitude; In the adjusted time window, the angle change sequence of different enterprise payment behavior trajectories is calculated, and the phase overlap degree is obtained based on the angle change sequence; The fluctuation feature of the phase overlap degree is extracted, the fluctuation cumulative value is calculated, the fluctuation cumulative value is compared with a dynamic threshold value calculated based on historical data, a continuous overlap interval is identified, and associated enterprises are identified based on the continuous overlap interval; The change synchronization of the payment behavior trajectory of the associated enterprises in the continuous overlap interval is calculated, the synchronization change intensity is obtained, and a risk transmission sequence is formed based on the synchronization change intensity.

2. The method of claim 1, wherein, The fluctuation amplitude is calculated based on the payment behavior sequence, the supply chain collaborative credit rating is calculated based on the fluctuation amplitude and historical default records, and when the supply chain collaborative credit rating continuously decreases, the corresponding enterprise is marked as a risk source, including: The payment behavior sequence of a supply chain enterprise is obtained, and a sliding variance test is used to identify the mutation points of the time interval sequence; The payment behavior sequence is divided into multiple time periods based on the mutation points, and the payment behavior fluctuation amplitude in each time period is calculated; An abnormal payment time point is identified according to the deviation degree of the payment behavior fluctuation amplitude, and the historical default records of the transaction enterprise corresponding to the abnormal payment time point are obtained; The change trend and correlation coefficient of the payment behavior fluctuation amplitude and the historical default records are calculated respectively, and the correlation coefficient is used as the supply chain collaborative credit rating; The change trend of the supply chain collaborative credit rating is monitored, and when the supply chain collaborative credit rating continuously decreases, the corresponding enterprise is marked as a risk source.

3. The method of claim 1, wherein, The evolution rule of the payment behavior portrait in the multi-dimensional feature space is used to predict the risk trigger node in combination with the risk transmission sequence, and a risk prevention and control scheme including credit limit control instructions and behavior monitoring strategies is generated, including: The time sequence feature of the payment behavior portrait in the multi-dimensional feature space is extracted, the change trend of the time sequence feature in adjacent time windows is calculated, and the behavior abnormal mode is identified; The payment behavior feature of the associated enterprise is extracted based on the behavior abnormal mode, the divergence degree and convergence degree of the payment behavior feature are calculated, and when the divergence degree exceeds a preset degree threshold, the corresponding time point is marked as a risk trigger node; The feature of the risk trigger node is traced back, the evolution trajectory segment before the trigger is extracted, the cumulative intensity of the feature change in the evolution trajectory segment is analyzed, and a risk outbreak time window is determined; The influence range in the risk outbreak time window is calculated in combination with the position of the enterprise in the risk transmission sequence, and credit limit control instructions are set; The behavior monitoring strategy is generated according to the credit limit control instructions, and the risk prevention and control scheme including the credit limit control instructions and the behavior monitoring strategy is output.

4. The method of claim 1, wherein, In the adjusted time window, the angle change sequence of different enterprise payment behavior trajectories is calculated, and the phase overlap degree is obtained based on the angle change sequence, including: Obtaining the payment behavior trajectory vector sequence in the adjusted time window; Decoupling the payment behavior trajectory vector sequence into instantaneous amplitude and instantaneous phase, and identifying dynamic boundary points based on the mutation degree of the instantaneous amplitude and the instantaneous phase; According to the dynamic boundary points, the payment behavior trajectory vector sequence is divided into multiple adaptive segments, and the angle change amount of adjacent payment behavior trajectory vectors is calculated in the adaptive segments to form an angle change sequence; Multi-scale decomposition is performed on the angle change sequence to identify the mutation position, and the mutation position is combined with the angle change sequence to generate an enhanced angle change sequence; Nonlinear dynamic normalization is performed on the enhanced angle change sequence of any two enterprises, and the phase overlap degree is calculated based on the time sequence matching degree of the mutation position in the normalization process.

5. The enterprise upstream and downstream collaborative risk control prediction system for supply chain data, for implementing the method of any one of claims 1-4, characterized in that, The system comprises: A sequence construction module for obtaining transaction records between upstream and downstream enterprises in a supply chain, extracting payment time deviation values between enterprises, and constructing payment behavior sequences; A risk source identification module for calculating fluctuation amplitude based on the payment behavior sequence, calculating supply chain collaborative credit rating based on the fluctuation amplitude and historical default records, and marking the corresponding enterprise as a risk source when the supply chain collaborative credit rating continuously decreases; A portrait construction module for using recursive partitioning time windows to map the payment behavior sequence of the risk source enterprise to a multi-dimensional feature space, extracting periodic and mutation indicators of payment behavior, and constructing a payment behavior portrait; An associated enterprise identification module for constructing enterprise payment behavior trajectories in a multi-dimensional feature space, identifying associated enterprises through the phase overlap degree of the payment behavior trajectories, and forming a risk transmission sequence based on the synchronous change intensity; A scheme generation module for predicting risk trigger nodes using the evolution law of the payment behavior portrait in the multi-dimensional feature space, combining the risk transmission sequence, and generating a risk prevention and control scheme including credit limit control instructions and behavior monitoring strategies.

6. An electronic device, comprising: It includes: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the method of any one of claims 1 to 4.

7. A computer readable storage medium characterized in that, The computer readable storage medium stores computer program instructions, and the computer program instructions are executed by the processor to implement the steps in the method of any one of claims 1 to 4.

Images