A digital auditing method and system for the supply chain

Abstract

This invention relates to the field of information technology. To address the problem of overlooking time-sensitive anomaly signals when processing massive dynamic datasets, this invention provides a digital auditing method and system for the supply chain. The steps are as follows: First, collect real-time transaction and logistics information from each node of the supply chain to obtain a timestamped dynamic dataset and determine the initial distribution of anomaly signals. Second, process the time-series data using a sliding window algorithm to obtain the fusion result of the latest transactions and historical data. Third, if delivery delays exceed a threshold, identify potential risks and obtain quality fluctuation correlation indicators. Fourth, use an anomaly detection algorithm to analyze logistics anomalies and determine the location of evidence chain breaks. Fifth, automatically slide the audit window according to the break location to obtain a timeliness analysis framework. Sixth, if the framework shows risk deviations, use a clustering algorithm to obtain an optimized early warning sequence. Seventh, fuse the latest transactions and logistics anomalies to determine the complete evidence chain path. Finally, process legacy data according to the path to obtain real-time audit results. This method accurately captures the initial distribution of time-sensitive anomaly signals, avoiding obscuring them with historical data.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, in particular to a digital auditing method and system for a supply chain. BACKGROUND

[0002] As a key field to ensure the safety and compliance of enterprise operations, supply chain digital auditing is important in that it can monitor the entire chain process from raw material procurement to product delivery in real time, avoiding potential risks that may cause significant losses to the business. In a complex and changing business environment, this field directly affects the economic benefits and market competitiveness of enterprises. Current supply chain auditing methods often rely on fixed-cycle batch checks, which cannot adapt to the real-time influx of transaction records and logistics information at each node of the supply chain, resulting in an auditing process that lags behind the actual operational rhythm.

[0003] Traditional methods tend to overlook time-sensitive abnormal signals when dealing with massive dynamic data sets, such as early signs of supplier delivery delays being submerged in historical data accumulation, thus missing the opportunity for intervention. This limitation further magnifies the core technical difficulty, namely the lack of a dynamic adjustment mechanism for the auditing window, making it impossible to effectively integrate historical data with the latest transaction and logistics information. SUMMARY

[0004] To solve the problem of overlooking time-sensitive abnormal signals when dealing with massive dynamic data sets in the prior art, the present application provides a digital auditing method and system for a supply chain.

[0005] The technical solution adopted by the present application is as follows:

[0006] A digital auditing method for a supply chain, comprising the following steps:

[0007] Step S100, by collecting real-time transaction records and logistics information at each node of the supply chain, obtaining a dynamic data set containing timestamps, and determining the preliminary distribution of abnormal signals therein;

[0008] Step S200, according to the obtained dynamic data set, processing time series data using a sliding window algorithm to obtain the fusion result of the latest transaction information and historical data within the window;

[0009] Step S300, if the delivery delay signs in the fusion result exceed a preset threshold, determining it as a potential risk event and obtaining the corresponding quality fluctuation correlation indicators; if the delivery delay signs in the fusion result do not exceed the preset threshold, staying at this step;

[0010] Step S400, by obtaining the quality fluctuation correlation indicators, using an anomaly detection algorithm to analyze the pattern of logistics anomalies captured, and determining the occurrence position of the evidence chain break;

[0011] Step S500, according to the determined position of the evidence chain rupture, the mechanism adjusts the automatic sliding of the audit window, and obtains an updated timeliness data dominant analysis framework; if the updated framework does not show risk prediction deviation, the step is stopped;

[0012] Step S600, based on the timeliness data dominant analysis framework, an optimized prediction and early warning sequence is obtained;

[0013] Step S700, through the obtained optimized prediction and early warning sequence, the latest transaction information and logistics anomaly capture are fused to determine the complete path of the continuity evidence chain;

[0014] Step S800, according to the determined complete path of the continuity evidence chain, the remaining data of the fixed period inspection is processed to obtain the audit output result of real-time monitoring.

[0015] Further, in step S100, by collecting real-time transaction records and logistics information of each node of the supply chain, a dynamic data set containing time stamps is obtained, and the preliminary distribution of abnormal signals is determined, which includes the following sub-steps:

[0016] Collect real-time transaction records and logistics information of each node of the supply chain through the supply chain node interface to generate a dynamic data set containing time stamps;

[0017] Using time series analysis, the time sequence characteristics of transaction records and logistics information are extracted from the dynamic data set to obtain abnormal signals;

[0018] If the time sequence characteristics of the abnormal signals exceed the preset threshold, the abnormal signals are classified by a clustering algorithm to determine the abnormal category;

[0019] According to the abnormal category, a decision tree algorithm is used to analyze the relevance of transaction records and logistics information to obtain the distribution pattern of abnormal signals;

[0020] Through the distribution pattern, the frequency and intensity of abnormal signals in the dynamic data set are calculated to determine the preliminary distribution;

[0021] If the frequency or intensity of the preliminary distribution exceeds the preset threshold, the future trend of the abnormal signal is obtained by a time series prediction algorithm;

[0022] According to the trend, the data collection frequency is adjusted, the monitoring parameters of the supply chain node are optimized, and the preliminary distribution of the abnormal signal is finally determined.

[0023] Further, in step S200, according to the obtained dynamic data set, a sliding window algorithm is used to process time series data to obtain the fusion result of the latest transaction information and historical data in the window, which includes the following sub-steps:

[0024] For the obtained dynamic data set, a sliding window algorithm is used to traverse the local mode of the time series data contained therein, and statistical indicators of transaction information within the window are extracted to obtain statistical features of the local mode;

[0025] If the statistical features of the local mode deviate from the preset threshold, the principal component analysis algorithm is used to compress the feature dimension to obtain a feature set after dimension reduction;

[0026] According to the feature set after dimension reduction, the k-means clustering algorithm is used to group the local mode to obtain the clustering result of the transaction information;

[0027] Through the clustering result, the timestamp interval distribution of each group of transaction information is calculated to determine the periodic pattern of the time series;

[0028] If the timestamp interval of the periodic pattern exceeds the preset threshold, the moving average algorithm is used to smooth the time series data to obtain the smoothed sequence trend;

[0029] According to the smoothed sequence trend, the size and step length of the sliding window are adjusted to obtain an optimized time series local mode;

[0030] Through the optimized time series local mode, the dynamic features of the transaction information and the historical data fusion are extracted, and finally the fusion result of the latest transaction information and the historical data within the window is obtained.

[0031] Further, in step S300, if the delivery delay indication in the fusion result exceeds the preset threshold, it is determined that there is a potential risk event, and the corresponding quality fluctuation correlation index is obtained; if the delivery delay indication in the fusion result does not exceed the preset threshold, it stays in this step, which includes the following sub-steps:

[0032] For the obtained fusion result, if the delivery delay indication therein exceeds the preset threshold, the distribution characteristics of the abnormal points are extracted through time series analysis to determine the potential risk event;

[0033] According to the distribution characteristics of the abnormal points, the principal component analysis algorithm is used to compress the quality fluctuation data to obtain a feature set after dimension reduction;

[0034] Through the feature set after dimension reduction, the correlation strength between the nodes of the supply chain is calculated to obtain a dynamic correlation matrix between the nodes;

[0035] If the correlation strength in the dynamic correlation matrix is lower than the preset threshold, the moving average algorithm is used to smooth the time series data to obtain a smoothed trend sequence;

[0036] According to the smoothed trend sequence, the fluctuation frequency of the periodic pattern is extracted to determine the abnormal distribution pattern of the supply chain nodes;

[0037] By the abnormal distribution mode, the transaction information deviation value of each node is calculated to obtain a quantitative index of potential risk;

[0038] According to the quantitative index of potential risk, the window parameter of time series analysis is adjusted to obtain an optimized risk prediction model, and finally the corresponding quality fluctuation correlation index is obtained.

[0039] Further, in step S400, by the obtained index, the mode of logistics anomaly capture is analyzed by using an anomaly detection algorithm to determine the occurrence position of the evidence chain break, specifically including the following sub-steps:

[0040] For the obtained quality fluctuation correlation index, the associated logistics data is processed by using an anomaly detection algorithm, and an abnormal pattern is extracted by using an isolation forest algorithm to obtain an abnormal distribution feature;

[0041] According to the abnormal distribution feature, the transaction information deviation between the nodes of the supply chain is calculated to obtain a node abnormal value;

[0042] If the node abnormal value exceeds a preset threshold, the time trend of node transaction is extracted by using a time series decomposition method to determine an abnormal fluctuation mode;

[0043] According to the abnormal fluctuation mode, a clustering algorithm is used to group the nodes of the supply chain to obtain a node abnormal cluster;

[0044] Through the node abnormal cluster, the statistical deviation of transaction information in each cluster is calculated to determine the evidence chain break position;

[0045] According to the evidence chain break position, the sensitivity parameter of the anomaly detection algorithm is adjusted to obtain an optimized anomaly detection model;

[0046] By using the optimized anomaly detection model, the logistics data is reprocessed to finally determine the occurrence position of the evidence chain break.

[0047] Further, in step S500, according to the determined occurrence position of the evidence chain break, the audit window is automatically slid by adjusting the mechanism to obtain an updated time effectiveness data dominant analysis framework, specifically including the following sub-steps:

[0048] For the determined occurrence position of the evidence chain break, abnormal node data is obtained therefrom, and a time series analysis is used to extract the fluctuation period of node transaction to determine a periodic abnormal mode;

[0049] According to the periodic abnormal mode, a clustering algorithm is used to group the nodes of the supply chain to obtain an abnormal node cluster;

[0050] The transaction information deviation is obtained from the abnormal node cluster, and the deviation distribution is calculated by using statistical analysis to determine the deviation concentration area;

[0051] If the deviation concentration area exceeds the preset threshold, the time sequence characteristics of the abnormal node are extracted by a time sequence decomposition method, and the abnormal duration is determined;

[0052] According to the abnormal duration, the time span of the sliding window is adjusted, and the updated audit time period is obtained;

[0053] From the updated audit time period, transaction data is obtained, and an anomaly detection algorithm is used to process the data to determine the optimized evidence chain breaking position;

[0054] According to the optimized evidence chain breaking position, a timeliness dominant analysis logic is generated, and finally an updated timeliness data dominant analysis framework is obtained.

[0055] Further, in step S600, based on the timeliness data dominant analysis framework, the optimized prediction and early warning sequence is obtained, which specifically includes the following sub-steps:

[0056] For the risk prediction deviation displayed by the updated timeliness data dominant analysis framework, the abnormal signals are grouped by a clustering algorithm according to the abnormal signal grouping basis, and a signal cluster set is obtained;

[0057] From the signal cluster set, the signal intensity distribution is obtained, the distribution characteristics are calculated by a statistical analysis method, and the concentration distribution area is determined;

[0058] If the concentration distribution area exceeds the preset threshold, the dynamic change trend of the signal cluster is extracted by time series analysis, and the change trend is determined;

[0059] According to the change trend, the input parameters of the prediction model are adjusted, and an optimized prediction parameter set is obtained;

[0060] From the optimized prediction parameter set, a prediction and early warning sequence is generated by a regression analysis method, and early warning sequence data is obtained;

[0061] According to the early warning sequence data, the risk prediction framework is adjusted by a deviation correction method, and an updated risk prediction model is obtained;

[0062] From the updated risk prediction model, the priority ranking of abnormal signals is extracted, and finally an optimized prediction and early warning sequence is obtained.

[0063] Further, in step S700, by obtaining the sequence, the latest transaction information and logistics abnormality capture are fused to determine the complete path of the continuous evidence chain, which specifically includes the following sub-steps:

[0064] For the obtained optimized prediction and early warning sequence, transaction information sequence and logistics abnormal signal are extracted therefrom, multi-source data is integrated by a data fusion rule, and a fusion data set is obtained;

[0065] From the fusion data set, the transaction timestamp and the logistics state update are extracted, the time series analysis is used to determine the time series correlation characteristics, and a time series correlation set is obtained;

[0066] If the time series correlation set meets the abnormal trigger condition, the length of the evidence chain is analyzed through path consistency verification, and the continuity evidence path is determined;

[0067] According to the continuity evidence path, the transaction correlation identifier and the logistics node sequence are extracted by using the association rule mining algorithm, and an association path set is obtained;

[0068] From the association path set, the path consistency verification is grouped by using the clustering analysis method, and an optimized path cluster is obtained;

[0069] Through the optimized path cluster, the effectiveness of the complete path structure is judged by using the preset threshold, and the final complete path is determined;

[0070] From the final complete path, the matching relationship between the transaction correlation identifier and the logistics node sequence is extracted, and the complete path of the continuity evidence chain is finally determined.

[0071] Further, in step S800, according to the complete path, the remaining data of the fixed period check is processed, and the audit output result of real-time monitoring is obtained, which specifically includes the following sub-steps:

[0072] For the determined complete path of the continuity evidence chain, the corresponding fixed period check remaining data is extracted, the effective transaction record and the logistics state update are extracted by using the data cleaning rule, and a standardized data set is obtained;

[0073] If the standardized data set meets the preset integrity threshold, the transaction timestamp and the logistics node sequence are extracted by using time series analysis, and a time series correlation set is obtained;

[0074] According to the time series correlation set, the matching relationship between the transaction and the logistics node is extracted by using the association rule mining algorithm, and an association path set is obtained;

[0075] If the path consistency in the association path set meets the preset threshold, the paths are grouped by using clustering analysis, and an optimized path cluster is obtained;

[0076] From the optimized path cluster, the effectiveness of the path structure is judged by using the preset threshold, and an effective path set is obtained;

[0077] According to the effective path set, the continuity characteristics of the transaction timestamp and the logistics node sequence are extracted, and the audit output basic data of real-time monitoring is obtained;

[0078] From the audit output basic data, a structured transaction and logistics matching relationship is generated by using data mapping method, and finally the audit output result of real-time monitoring is obtained.

[0079] Based on the same inventive concept, the present application also discloses a digital audit system for a supply chain, which executes the aforementioned digital audit method for a supply chain, and comprises the following modules:

[0080] A data collection and preliminary anomaly distribution module, which inputs real-time transaction records and logistics information of each node of the supply chain, acquires a dynamic data set containing a time stamp through information collection, determines a preliminary distribution of abnormal signals, and outputs the dynamic data set and the preliminary distribution of abnormal signals.

[0081] A sliding window data fusion module, which inputs the dynamic data set, processes time series data by using a sliding window algorithm, obtains a fusion result of the latest transaction information and historical data within the window, and outputs the fusion result.

[0082] A risk judgment and quality index module, which inputs the fusion result, determines a potential risk event if a delivery delay sign exceeds a preset threshold, acquires a quality fluctuation correlation index, and outputs the potential risk event and the quality fluctuation correlation index.

[0083] An anomaly detection and fracture positioning module, which inputs the quality fluctuation correlation index, analyzes a logistics anomaly pattern by using an anomaly detection algorithm, determines a fracture position of an evidence chain, and outputs the fracture position of the evidence chain.

[0084] An audit window adjustment module, which inputs the fracture position, automatically slides an audit window by using an adjustment mechanism, obtains a time-effectiveness data dominant analysis framework, and outputs the analysis framework.

[0085] An early warning sequence optimization module, which inputs the analysis framework, groups abnormal signals by using a clustering algorithm if a risk prediction deviation is displayed, obtains a signal cluster set, and outputs an optimized early warning sequence.

[0086] A complete evidence chain determination module, which inputs the optimized early warning sequence, fuses the latest transaction information and logistics anomaly capture, determines a complete path of a continuous evidence chain, and outputs the complete path.

[0087] An audit result generation module, which inputs the complete path, processes fixed-period inspection residual data, obtains a real-time monitoring audit output result, and outputs the audit output result.

[0088] Each of the above modules can be realized by computer software.

[0089] The beneficial effects of the present application are: the supply chain risk audit method and system based on dynamic data set disclosed in the present application can accurately capture the preliminary distribution of time-sensitive abnormal signals by collecting real-time transaction records and logistics information of each node in the supply chain to obtain a dynamic data set, avoiding that early signs such as supplier delivery delay are covered by historical data, and providing a data basis for timely intervention; the sliding window algorithm is used to process time series data, effectively fusing the latest transaction information and historical data, and breaking the limitation that historical and latest information cannot be effectively fused; the abnormal detection algorithm is used to locate the evidence chain breaking position, and the audit window is automatically slid based on the position to make up for the short board of the missing dynamic adjustment mechanism of the audit window; finally, the real-time audit result is output by combining clustering optimization early warning sequence, determining complete evidence chain and processing residual data, which guarantees the timeliness and continuity of the audit throughout the whole process, greatly improves the accuracy of supply chain risk early warning, reduces the operation loss caused by missing intervention opportunity, and enhances the overall operation resilience of the supply chain. BRIEF DESCRIPTION OF DRAWINGS

[0090] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0091] Figure 1 The flowchart of the supply chain risk audit method based on dynamic data set is shown in the figure.

[0092] Figure 2 The block diagram of the supply chain risk audit system based on dynamic data set is shown in the figure. DETAILED DESCRIPTION

[0093] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate the orientation or positional relationship shown in the drawings, which are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0094] The following disclosure provides many different embodiments or examples for implementing different structures of the present application. In order to simplify the disclosure of the present application, the components and settings of a specific example are described below. Of course, they are only examples, and the purpose is not to limit the present application.

[0095] The embodiments of the invention will now be described in detail with reference to the accompanying drawings.

[0096] As attached Figure 1 As shown, the digital auditing method for the supply chain disclosed in this embodiment specifically includes the following steps:

[0097] Step S100: Collect real-time transaction records and logistics information from each node of the supply chain to determine the preliminary distribution of abnormal signals.

[0098] Real-time transaction records and logistics information from each node in the supply chain are collected via supply chain node interfaces to generate a dynamic dataset containing timestamps. Time series analysis is used to extract the temporal characteristics of transaction records and logistics information from the dynamic dataset to obtain anomaly signals. If the temporal characteristics of the anomaly signal exceed a preset threshold, a clustering algorithm is used to classify the anomaly signal and determine its category. Based on the anomaly category, a decision tree algorithm is used to analyze the correlation between transaction records and logistics information to obtain the distribution pattern of the anomaly signal. Based on the distribution pattern, the frequency and intensity of the anomaly signal in the dynamic dataset are calculated to determine the preliminary distribution. If the frequency or intensity of the preliminary distribution exceeds a preset threshold, a time series prediction algorithm is used to obtain the future trend of the anomaly signal. Based on the trend, the data collection frequency is adjusted, the monitoring parameters of the supply chain nodes are optimized, and the preliminary distribution of the anomaly signal is finally determined.

[0099] Specifically, when the supply chain node interface collects real-time transaction records and logistics information from each node in the supply chain, it can extract timestamp-containing data from the supply chain management system via API. For example, a warehouse node on an e-commerce platform uploads the logistics status and transaction order information of packages every minute, generating a dynamic dataset that records information such as order ID, shipping time, and logistics status. Time series analysis can use the sliding window method to extract the temporal characteristics of transaction records and logistics information. For example, analyzing the frequency of package status changes every hour, if 10 packages are processed per minute under normal conditions, but suddenly increase to 50 in a certain hour, it is considered an abnormal signal. When the abnormal signal exceeds a threshold (e.g., 30 packages / minute), subsequent processing is triggered.

[0100] In one possible implementation, anomalous signals are classified using a K-means clustering algorithm. For example, anomalous signals can be divided into high-frequency and low-frequency anomalies based on transaction volume and logistics delays. High-frequency anomalies might indicate duplicate orders, while low-frequency anomalies might indicate logistics congestion. After classification, a decision tree algorithm analyzes the correlation between transaction records and logistics information. For example, if a certain anomaly category shows a surge in transaction volume but no update in logistics status, the decision tree infers that it's due to duplicate orders caused by a system error, generating a distribution pattern indicating that the anomalies are concentrated in specific time periods, such as promotional events.

[0101] Specifically, distribution patterns can calculate the frequency and intensity of anomalous signals. For example, on a certain day, the frequency of anomalous signals might be 20 times per hour, with an intensity sufficient to affect 1000 orders, far exceeding the threshold of 10 times per hour or 500 orders. Time series forecasting algorithms such as ARIMA can predict anomalous trends for the next 24 hours based on historical anomalous data. For example, the prediction might show that the anomalous frequency will continue to increase in the early morning of the following day. Based on this trend, the data collection frequency can be adjusted, such as increasing it from once per minute to once every 30 seconds, optimizing monitoring parameters, and ultimately determining the preliminary distribution of anomalous signals.

[0102] Understandably, the optimized initial distribution of anomaly signals can improve supply chain efficiency. For example, after real-time adjustments to monitoring parameters, the system can issue warnings within 5 minutes of an anomaly occurring, instead of the original 15 minutes, reducing order processing delays. Furthermore, anomaly classification and prediction results can provide decision support for supply chain managers, such as strengthening system validation for duplicate order anomalies, reducing the error rate by 10%.

[0103] Preferably, through multi-dimensional analysis, the temporal characteristics, classification, and distribution patterns of abnormal signals form a closed-loop feedback, continuously optimizing the monitoring effect of supply chain nodes and ensuring the stability of logistics and transactions.

[0104] Step S200: Based on the dynamic dataset, obtain the fusion result of the latest transaction information and historical data.

[0105] For the acquired dynamic dataset, a sliding window algorithm is used to traverse the local patterns of the time series data it contains, extracting statistical indicators of transaction information within the window to obtain the statistical features of the local patterns. If the statistical features of the local patterns deviate from a preset threshold, principal component analysis is used to compress the feature dimensions, resulting in a dimensionality-reduced feature set. Based on the dimensionality-reduced feature set, k-means clustering is used to group the local patterns, obtaining the clustering results of the transaction information. Using the clustering results, the timestamp interval distribution of each group of transaction information is calculated to determine the periodicity pattern of the time series. If the timestamp interval of the periodic pattern exceeds a preset threshold, a moving average algorithm is used to smooth the time series data, obtaining the smoothed sequence trend. Based on the smoothed sequence trend, the size and step size of the sliding window are adjusted to obtain an optimized local time series pattern. Through the optimized local time series pattern, dynamic features of the fusion of transaction information and historical data are extracted, ultimately obtaining the fusion result of the latest transaction information and historical data within the window.

[0106] For example, for dynamic datasets, the sliding window algorithm can divide the time series of supply chain transactions it contains into fixed-size windows, such as one window per hour, and traverse local patterns to capture transaction fluctuations. In one possible implementation, statistical indicators are extracted within the window, including the mean and variance of transaction volume. If the mean transaction volume of a certain window reaches 500 transactions while the threshold is only 300 transactions, it is considered to be deviating from the normal range.

[0107] Specifically, principal component analysis (PCA) reduces the dimensionality of these indicators from 10 to 3, preserving key variances such as trading peaks and latency distributions, resulting in a dimensionality-reduced feature set for easier subsequent processing. For example, k-means clustering grouped local patterns into three classes based on the dimensionality-reduced set: peak trading groups, stable groups, and volatile groups, with the center point of each group reflecting differences in trading density.

[0108] It should be noted that the timestamp interval distribution is calculated through clustering results. For example, if the peak group interval is 5 minutes on average, it reveals periodic patterns such as daily promotional peaks. In one embodiment, if the periodic pattern interval exceeds a threshold of 10 minutes, a moving average algorithm is applied with a window size of 15 to smooth out noise and obtain a trend line showing a gradual increase in transactions.

[0109] Preferably, the sliding window size is increased from 1 hour to 30 minutes and the step size is halved based on the smoothing trend, thereby improving the sensitivity to sudden transactions. For example, the optimized local model integrates historical data and extracts dynamic features such as order matching rate, ultimately obtaining the fusion result of the latest transaction information and historical data within the window.

[0110] In one possible implementation, this process forms a tight chain from window traversal to clustering: statistical deviations trigger dimensionality reduction, clustering reveals cycles, smoothing optimizes the window, and finally, features are fused to generate a fusion result. Specifically, with multi-faceted support, the peak group interval distribution matches historical promotional data, smoothing trends predict the next day's peak, and support decisions such as increasing logistics resources.

[0111] Understandably, this method can accurately generate fusion results, improve the accuracy of supply chain monitoring, and reduce the risk of transaction delays.

[0112] Step S300: Based on the fusion results, obtain the quality fluctuation correlation index.

[0113] For the acquired fusion results, if delivery delay indicators exceed a preset threshold, time series analysis is used to extract the distribution characteristics of outliers to identify potential risk events. Based on the distribution characteristics of outliers, principal component analysis is used to compress the quality fluctuation data, obtaining a dimensionality-reduced feature set. Using this dimensionality-reduced feature set, the correlation strength between supply chain nodes is calculated, resulting in a dynamic correlation matrix. If the correlation strength in the dynamic correlation matrix is ​​lower than a preset threshold, a moving average algorithm is used to smooth the time series data, obtaining a smoothed trend sequence. Based on the smoothed trend sequence, the fluctuation frequency of cyclical patterns is extracted to determine the abnormal distribution patterns of supply chain nodes. Using these abnormal distribution patterns, the transaction information deviation value for each node is calculated, yielding a quantitative indicator of potential risk. Based on this quantitative indicator of potential risk, the window parameters of the time series analysis are adjusted to obtain an optimized risk prediction model, ultimately resulting in the corresponding quality fluctuation correlation indicator.

[0114] For example, in supply chain management, if delivery delays exceed a preset threshold (e.g., a 2-hour delay actually reaching 3 hours) in the integrated results, time series analysis can be used to extract the distribution characteristics of outliers. Time series analysis identifies outliers by observing the timestamps of transaction data, such as a warehouse node experiencing excessive delays for several consecutive days. Assuming a node handles 1000 orders daily, a sudden increase in the proportion of delayed orders to 20%, far exceeding the normal 5% threshold, can be identified as a potential risk event. Specifically, analyzing the temporal distribution of outliers reveals that delays are concentrated during the daily logistics peak period of 17:00-19:00, indicating scheduling bottlenecks.

[0115] In one possible implementation, principal component analysis (PCA) is used to compress quality fluctuation data. This data includes 10 dimensions such as order delay time, return rate, and inventory turnover. PCA extracts the first three dimensions, such as the combination of delay time and return rate, while retaining 80% of the original data variance. For example, if a warehouse node has a mean delay time of 2.5 hours and a return rate of 10%, dimensionality reduction generates a low-dimensional feature set that highlights the correlation between delay and returns, facilitating subsequent analysis of relationships between nodes.

[0116] For example, by using a dimensionality-reduced feature set, the correlation strength between supply chain nodes can be calculated to construct a dynamic correlation matrix. The correlation strength is calculated based on the order flow frequency and latency correlation between nodes. Suppose that warehouse A and distribution center B have an order flow frequency of 500 transactions per day, but their latency correlation is as low as 0.3, below the threshold of 0.5, indicating low collaboration efficiency between the two. The dynamic correlation matrix intuitively reflects the weak correlation between A and B, suggesting the need to optimize the logistics route.

[0117] Specifically, if the correlation strength is below a threshold, a moving average algorithm can be used to smooth the time series data. The moving average uses a 7-day window to smooth daily latency data, filtering out short-term fluctuations. For example, if a node's latency time series shows daily fluctuations of 1.8-3.2 hours, the smoothed trend shows the latency gradually increasing to 2.8 hours, revealing an escalating potential risk. The smoothed trend series helps identify long-term patterns, such as latency worsening with increasing order volume.

[0118] In one possible implementation, the fluctuation frequency of cyclical patterns is extracted by smoothing trend sequences. Assuming the analysis reveals that delay peaks occur twice a week, with a frequency consistent with promotional activities, the abnormal distribution pattern can be identified as promotion-driven. Further calculation of transaction information deviation values ​​is then performed; for example, if a node's delay deviation reaches 1.5 hours, exceeding the average of 0.5 hours, its risk level is quantified as high. Based on the quantitative indicators of potential risk, the window parameters of the time series analysis are adjusted (originally 1 hour, but can be shortened to 30 minutes if the risk indicators show high-frequency fluctuations), improving sensitivity to sudden delays and ultimately obtaining the corresponding quality fluctuation correlation indicators.

[0119] This method, through anomaly analysis, dimensionality reduction, smoothing, and window optimization, forms a complete chain, accurately pinpointing supply chain risk points and enhancing monitoring capabilities. In one embodiment, the dynamic correlation matrix between nodes is combined with cyclical patterns to reveal the impact of promotional activities on delays. For example, a 30% surge in orders on a promotional day leads to anomalies in the delay distribution at specific nodes, prompting the model to recommend increasing transportation resources. This multi-faceted analysis, from anomaly extraction to risk quantification, is logically rigorous and collectively supports the accuracy of quality fluctuation correlation indicators, improving supply chain management efficiency.

[0120] Step S400: Based on the quality fluctuation correlation index, determine the location where the chain of evidence breaks.

[0121] For the obtained quality fluctuation correlation indicators, anomaly detection algorithms are used to process the associated logistics data. Anomaly patterns are extracted using the Isolation Forest algorithm to obtain anomaly distribution characteristics. Based on these characteristics, the transaction information deviations between supply chain nodes are calculated to obtain node outliers. If a node outlier exceeds a preset threshold, a time series decomposition method is used to extract the time trend of node transactions to determine the anomaly fluctuation pattern. Based on the anomaly fluctuation pattern, a clustering algorithm is used to group the supply chain nodes, resulting in node anomaly clusters. Through these clusters, the statistical deviation of transaction information within each cluster is calculated to determine the location of the evidence chain break. Based on the evidence chain break location, the sensitivity parameters of the anomaly detection algorithm are adjusted to obtain an optimized anomaly detection model. Using the optimized anomaly detection model, the logistics data is reprocessed to ultimately determine the location of the evidence chain break.

[0122] For example, in supply chain management, the Isolation Forest algorithm can quickly identify potential risks by analyzing anomalies in the associated logistics data for indicators related to quality fluctuations. Isolation Forest is based on the principle of randomly partitioning data points; anomalies are more easily isolated due to their sparse distribution. Suppose a logistics node processes 5000 orders daily, and 20 orders have abnormally long delivery times, exceeding the normal average of 2 hours. The Isolation Forest algorithm marks these orders as anomalies, generating anomaly distribution characteristics that reflect the degree to which delivery times deviate from the normal pattern. This method can efficiently process high-dimensional data and is suitable for real-time monitoring of complex supply chains.

[0123] Specifically, based on the characteristics of abnormal distribution, the deviation of transaction information between nodes is calculated. For example, warehouse C sends 1000 orders to distribution center D daily, with a normal deviation of 0.5 hours. However, on one day, the deviation suddenly increases to 2 hours, exceeding the preset threshold of 1 hour. This indicates that there may be a problem with the logistics collaboration between C and D, such as insufficient dispatch of transport vehicles. The deviation calculation quantifies the degree of anomaly by comparing the difference between the actual delivery time and the historical average, providing a basis for subsequent analysis.

[0124] In one embodiment, time series decomposition is used to extract the time trend of node transactions. Suppose that time series data for node D shows that delivery delays are concentrated on Friday afternoons each week. Decomposition reveals that the trend component shows a gradual increase in delays, while the seasonal component reveals peak fluctuations every Friday. This suggests that the delays may be related to a surge in weekend orders. Time series decomposition breaks down the data into trend, seasonality, and residuals, helping to pinpoint anomalous fluctuation patterns.

[0125] For example, the K-means clustering algorithm can be used to group supply chain nodes, dividing them into clusters based on abnormal fluctuation patterns. Assuming an analysis of 20 nodes, the clustering results generate three clusters: high-risk, medium-risk, and low-risk. The high-risk cluster contains 5 nodes, such as warehouse C, whose average latency is 3 hours, significantly higher than the 1.5 hours of other clusters. Clustering, by grouping nodes based on the similarity of their anomalous characteristics, facilitates targeted optimization of resource allocation.

[0126] Specifically, by analyzing clusters of abnormal nodes, statistical deviations in transaction information within each cluster are calculated to initially pinpoint the location of breaks in the chain of evidence. For example, in a high-risk cluster, the order flow deviation between warehouse C and distribution center D reaches 2.5 hours, far exceeding the normal 0.8 hours, indicating a bottleneck in the logistics path from C to D, such as insufficient transportation capacity. Statistical deviations, by aggregating latency and order volume data from nodes within the cluster, reveal key nodes in the supply chain disruption.

[0127] In one embodiment, the sensitivity parameter of the Isolation Forest algorithm is adjusted to optimize the anomaly detection model. The original sensitivity was set to detect 5% of anomalies; if high-risk clusters show frequent anomalies, this can be increased to 10% to capture more potential risks. For example, the adjusted model identified a sudden increase in latency of 6 hours for node C on a promotional day, outperforming the original model's missed detection. This optimization improves the model's responsiveness to sudden anomalies.

[0128] For example, the optimized model reprocesses logistics data to ultimately pinpoint the location of the break in the chain of evidence. Hypothetical analysis reveals that the break in the path from C to D is concentrated between 16:00 and 18:00 daily. Combined with order volume data, this shows that promotional activities led to a 30% surge in shipments. The precise distribution of break locations supports scheduling optimization, such as increasing the number of vehicles available for evening deliveries. This multi-layered analysis, from anomaly detection to model optimization, forms a complete logical chain, significantly improving supply chain risk monitoring capabilities.

[0129] Step S500: Based on the location of the break in the chain of evidence, obtain the timeliness-driven analysis framework.

[0130] For identified locations where the chain of evidence breaks, abnormal node data is extracted. Time series analysis is used to extract the fluctuation cycle of node transactions and determine periodic anomaly patterns. Based on these periodic anomaly patterns, a clustering algorithm is used to group supply chain nodes, obtaining abnormal node clusters. Transaction information deviations are obtained from these clusters, and statistical analysis is used to calculate the deviation distribution and identify areas of concentrated deviation. If a concentrated deviation area exceeds a preset threshold, time series decomposition is used to extract the temporal characteristics of the abnormal nodes and determine the duration of the anomaly. Based on the duration of the anomaly, the time span of the sliding window is adjusted to obtain an updated audit period. Transaction data is obtained from the updated audit period, and anomaly detection algorithms are used to process the data and determine the optimized locations of the chain of evidence breaks. Based on the optimized locations of the chain of evidence breaks, a timeliness-driven analysis logic is generated, ultimately resulting in an updated timeliness-driven data-driven analysis framework.

[0131] For example, in supply chain management, for the location of a confirmed break in the chain of evidence, time series analysis can be used to extract data on anomalous nodes and uncover transaction fluctuation cycles. Time series analysis identifies patterns in anomalies by decomposing periodic patterns in the data. Suppose a distribution center E processes 6,000 orders daily. Analysis reveals that order processing times fluctuate significantly every Wednesday, with an average delay of 2.5 hours, far exceeding the normal 1 hour. Time series analysis reveals that the fluctuation cycle is weekly, suggesting that the anomaly may be related to the midweek peak in order volume. This method extracts periodic features to provide a basis for subsequent clustering.

[0132] For example, based on periodic anomaly patterns, the K-means clustering algorithm can be used to group supply chain nodes. The clustering algorithm divides nodes into different risk clusters based on the volatility characteristics of node transactions. Assuming an analysis of 30 nodes, four clusters are generated, with the high-risk cluster containing six nodes, such as warehouse F, whose average latency is 3 hours. Clustering, by grouping nodes based on the similarity of volatility cycles and latency levels, facilitates the identification of areas with concentrated anomalies. This grouping method helps to focus on problematic nodes and improve resource allocation efficiency.

[0133] Specifically, transaction information deviations are obtained from the cluster of abnormal nodes, and statistical analysis is used to calculate the deviation distribution. Statistical analysis identifies concentrated deviation areas by aggregating order delay and volume data. For example, assuming the order flow deviation from warehouse F to distribution center G reaches 2.8 hours, concentrated between 14:00 and 16:00 daily, exceeding the threshold by 1.5 hours, deviation distribution analysis shows that the anomaly mainly stems from insufficient vehicle scheduling during peak volume periods. This analysis clarifies the spatiotemporal distribution of the deviation, providing support for precise problem localization.

[0134] In one embodiment, if the concentrated area of ​​deviation exceeds a threshold, a time series decomposition method can extract the temporal characteristics of the abnormal nodes to determine the duration of the anomaly. The decomposition method splits the data into trend, seasonal, and residual components. Assuming node G data shows a delay lasting 4 hours on Thursday afternoon, the seasonal component reveals peak fluctuations, while the trend component shows a gradual increase in delay. This indicates that the anomaly is related to a surge in cargo volume on Thursday. Identifying the duration of the anomaly helps optimize monitoring strategies.

[0135] For example, by adjusting the time span of the sliding window based on the duration of the anomaly, an updated audit period can be obtained. The sliding window dynamically adjusts the analysis scope, focusing on periods with high anomaly incidence. Assuming the original window was 24 hours, it is adjusted to 12 hours, covering 14:00-02:00 daily. Audits reveal that anomalies at node G are concentrated on promotional days, increasing the delay to 5 hours. This adjustment improves the timeliness of data analysis and optimizes the accuracy of anomaly detection.

[0136] Specifically, transaction data is obtained from the updated audit period, and anomaly detection algorithms such as DBSCAN are used to process the data to determine the location of the optimized chain of evidence break. DBSCAN identifies anomalies through density clustering. For example, it is found that the break in the path from F to G is concentrated between 16:00 and 18:00 on the promotional day, with a 40% surge in order volume. This method efficiently processes irregularly distributed data and accurately locates the break.

[0137] In one embodiment, based on the optimized location of the evidence chain break, a timeliness-driven analysis logic is generated, ultimately resulting in an updated timeliness-driven data-driven analysis framework. This framework integrates time series features, clustering results, and deviation distributions to dynamically monitor abnormal trends. Assume the framework shows that the abnormal distribution along the F to G path fluctuates with promotional activities, with peak delays concentrated on weekends. This framework improves the response speed and accuracy of supply chain anomaly monitoring through multi-dimensional analysis.

[0138] Step S600: Based on the timeliness data-driven analysis framework, obtain the optimized prediction and early warning sequence.

[0139] To address the risk prediction biases revealed by the updated timeliness-driven data-driven analysis framework, a clustering algorithm is used to group abnormal signals according to their grouping criteria, resulting in a signal cluster set. The signal intensity distribution is obtained from these clusters, and statistical analysis methods are employed to calculate distribution characteristics and identify concentrated distribution areas. If a concentrated distribution area exceeds a preset threshold, time series analysis is used to extract the dynamic trends of the signal clusters and determine these trends. Based on these trends, the input parameters of the prediction model are adjusted to obtain an optimized set of prediction parameters. From this optimized set of prediction parameters, regression analysis is used to generate prediction and warning sequences, obtaining warning sequence data. Based on this warning sequence data, a bias correction method is used to adjust the risk prediction framework, resulting in an updated risk prediction model. From the updated risk prediction model, the priority ranking of abnormal signals is extracted, ultimately yielding an optimized prediction and warning sequence.

[0140] For example, in supply chain management, grouping anomaly signals is crucial for optimizing anomaly monitoring, especially when analyzing risk prediction deviations revealed by updated timeliness data-driven frameworks. Anomaly signals typically refer to abnormal fluctuations in order processing, such as delays and insufficient inventory. Grouping anomaly signals using clustering algorithms allows anomalies with similar characteristics to be categorized into signal clusters. Assuming a logistics network with 50 delivery nodes processing 100,000 orders daily, anomaly signals include order delays and surges in shipment volume. Using K-means clustering, based on delay time and shipment volume fluctuation magnitude, the anomaly signals are divided into three clusters: a high-frequency, high-intensity cluster, a medium-frequency, medium-intensity cluster, and a low-frequency, low-intensity cluster. The high-frequency, high-intensity cluster contains 8 nodes, such as delivery center H, with an average daily delay of 3 hours and shipment volume fluctuations exceeding 30%. This grouping method facilitates the identification of areas with concentrated anomalies, providing a foundation for subsequent analysis.

[0141] In one possible implementation, signal strength distribution is obtained from a set of signal clusters, and statistical analysis methods are used to calculate distribution characteristics. The signal strength distribution reflects the concentration of anomalies; for example, the concentration of anomalies can be determined by calculating the standard deviation and mean of delay times. Assuming the standard deviation of delay times for a high-frequency, high-intensity cluster is 1.5 hours and the mean is 3 hours, exceeding a preset threshold of 1 hour, statistical analysis shows that anomalies are concentrated between 15:00 and 17:00 daily, with a surge in cargo volume leading to scheduling pressure. Such distribution characteristics provide data support for subsequent dynamic analysis.

[0142] For example, time series analysis can be used to extract the dynamic trends of signal clusters and identify the periodic patterns of anomalies. Time series analysis decomposes data into trend and seasonal components, revealing the patterns of anomaly changes over time. Suppose that order data analysis of distribution center H shows significant delays every Friday afternoon; the trend component indicates that the delays are gradually worsening, while the seasonal component reveals a correlation with peak sales volume on promotional days. Based on this trend, the input parameters of the prediction model can be adjusted, such as increasing the weight of peak sales volume, resulting in an optimized set of prediction parameters and improving the model's sensitivity to anomalies.

[0143] In one possible implementation, regression analysis is used to generate a predictive warning sequence from the optimized set of prediction parameters. Regression analysis predicts the likelihood of future anomalies by fitting historical delay and volume data. Assuming the prediction model indicates that distribution center H may experience a 4-hour delay on a future promotional day, the warning sequence data identifies the high-risk period as 16:00-18:00 daily. This predictive warning sequence provides a precise time window for risk assessment.

[0144] For example, by adjusting the risk prediction framework using a deviation correction method, an updated risk prediction model is obtained. Deviation correction optimizes model parameters by comparing predicted and actual values. Assuming an actual delay of 4.5 hours and a predicted deviation of 0.5 hours, the corrected model more accurately identifies high-risk nodes. From the updated risk prediction model, the priority ranking of abnormal signals is extracted, ultimately yielding an optimized predictive warning sequence. For instance, distribution center H is marked as the highest priority, and the warning sequence shows that its anomalies are concentrated on promotional days, requiring priority allocation of scheduling resources. This approach improves the accuracy and response efficiency of supply chain anomaly monitoring through multi-dimensional analysis.

[0145] Step S700: Based on the predicted warning sequence, determine the complete path of the continuous evidence chain.

[0146] For the obtained optimized prediction and early warning sequence, transaction information sequences and logistics anomaly signals are extracted. Multi-source data are integrated using data fusion rules to obtain a fused data set. From the fused data set, transaction timestamps and logistics status updates are extracted, and time series analysis is used to determine temporal correlation features, resulting in a time series correlation set. If the time series correlation set meets the anomaly triggering conditions, the length of the evidence chain is analyzed through path consistency verification to determine the continuous evidence path. Based on the continuous evidence path, an association rule mining algorithm is used to extract transaction association identifiers and logistics node sequences, resulting in a correlation path set. From the correlation path set, cluster analysis is used to group the path consistency verification data, resulting in an optimized path cluster. Through the optimized path cluster, a preset threshold is used to judge the validity of the complete path structure, determining the final complete path. From the final complete path, the matching relationship between transaction association identifiers and logistics node sequences is extracted, ultimately determining the complete path of the continuous evidence chain.

[0147] For example, in supply chain management, extracting transaction information sequences and logistics anomaly signals from the optimized forecast and early warning sequences is a key step in optimizing monitoring. Transaction information sequences typically include order time, amount, and customer information, while logistics anomaly signals include delivery delays and inventory discrepancies. Data fusion rules integrate these multi-source data to form a unified dataset. For instance, an e-commerce platform's logistics network processes 200,000 orders daily, including transaction timestamps and logistics status updates such as order confirmation time and delivery node arrival time. Fusion rules can align transaction data with logistics status through timestamp alignment, generating a fused dataset that records the entire process of each order from order placement to delivery.

[0148] In one possible implementation, transaction timestamps and logistics status updates are extracted from the fused dataset, and time series analysis is used to determine temporal correlation features. Time series analysis focuses on data changes over time; for example, by analyzing the difference between order confirmation time and delivery arrival time, delay features can be extracted. Assuming a delivery node has a daily order confirmation time of 10:00 and a delivery arrival time of 16:00, a delay of 6 hours, analysis shows that the delay is concentrated during the peak period of 12:00-14:00. The time series correlation dataset records these features, reflecting the regularity of anomalies.

[0149] For example, if the time-series association set shows a delay exceeding 4 hours, an anomaly condition is triggered, requiring path consistency verification. Path consistency verification ensures the continuity of transactions and logistics nodes by analyzing the length of the evidence chain. For instance, an order must pass through warehouse A, sorting center B, and delivery station C from order placement to delivery. Verification reveals a 2-hour delay at sorting center B, with an evidence chain length of 3 nodes. Continuity evidence paths record the complete process, ensuring no data breaks.

[0150] In one possible implementation, an association rule mining algorithm is used to extract transaction association identifiers and logistics node sequences from continuous evidence paths. For example, the association strength between order ID123 and warehouse A and sorting center B is 0.9, indicating a high correlation. The association path set records the correspondence between multiple orders and nodes; for example, 1000 orders may involve 10 nodes. Cluster analysis groups the paths into high-frequency paths and low-frequency paths; for example, high-frequency paths are concentrated from warehouse A to distribution station C.

[0151] For example, by optimizing path clusters, a preset threshold is used to determine the validity of the path structure. Assuming the threshold requires paths to cover 80% of order volume, high-frequency paths meet this condition and are identified as the final complete path. The matching relationship between transaction association identifiers and logistics node sequences is extracted from the final complete path; for example, order ID 123 matches delivery station C, ultimately determining the complete path of the continuous evidence chain. This approach improves the accuracy and efficiency of supply chain monitoring through multi-dimensional analysis.

[0152] Step S800: Based on the complete path, obtain the audit output results of real-time monitoring.

[0153] For a defined, continuous chain of evidence, the corresponding periodic inspection data is extracted. Data cleaning rules are used to extract valid transaction records and logistics status updates, resulting in a standardized dataset. If the standardized dataset meets a preset integrity threshold, time series analysis is used to extract transaction timestamps and logistics node sequences, resulting in a time-series association set. Based on this time-series association set, association rule mining algorithms are used to extract matching relationships between transactions and logistics nodes, resulting in an association path set. If the path consistency in the association path set meets a preset threshold, cluster analysis is used to group the paths, resulting in an optimized path cluster. From the optimized path cluster, a preset threshold is used to determine the validity of the path structure, resulting in a valid path set. Based on the valid path set, the continuity features of transaction timestamps and logistics node sequences are extracted to obtain the basic data for real-time monitoring audit output. From this audit output basic data, a data mapping method is used to generate structured transaction and logistics matching relationships, ultimately yielding the real-time monitoring audit output results.

[0154] For example, in supply chain management, extracting the corresponding periodic legacy data from a defined, continuous chain of evidence and extracting useful information is a crucial step in optimizing monitoring. Legacy data typically includes historical order records and logistics status logs, which may contain noisy data, such as duplicate orders or invalid logistics statuses. Data cleaning rules remove invalid or redundant records by filtering valid fields, such as order IDs, transaction timestamps, and logistics node statuses. For instance, an e-commerce platform generates 300,000 order records daily, of which 5% are duplicates. Data cleaning rules can deduplicate by order ID, retaining unique records, and verify the integrity of logistics statuses, such as ensuring the existence of delivery node timestamps, generating a standardized dataset.

[0155] In one embodiment, the standardized dataset must meet an integrity threshold, such as requiring 90% of order records to contain complete transaction timestamps and logistics node information. If this condition is met, time series analysis can be used to extract the temporal characteristics of the transaction timestamps and logistics node sequences. For example, analyzing an order from its placement time of 10:00 to its delivery node arrival time of 15:00, extracting the time difference feature reveals a concentration of delivery delays during the peak period of 13:00-14:00. The time series association dataset records these characteristics, reflecting the temporal relationship between orders and logistics nodes.

[0156] For example, association rule mining algorithms extract matching relationships between transactions and logistics nodes from a time-series association set. Suppose that the association strength between order ID 456 and warehouse X and delivery station Y is 0.85, indicating a high correlation. The association path set records the correspondence between multiple orders and nodes; for example, 2000 orders may involve 8 logistics nodes. If the path consistency meets a preset threshold, such as 80% of the order paths covering the main nodes, further processing is performed.

[0157] In one embodiment, cluster analysis groups the set of associated paths into high-frequency and low-frequency paths. For example, a high-frequency path is from warehouse X to delivery station Y, covering 70% of the order volume. The optimized path cluster filters out efficient paths by analyzing path frequency. A preset threshold requires paths to cover 85% of the order volume; high-frequency paths meet this condition and are selected as the effective path set.

[0158] For example, the set of valid paths is used to extract continuous features, such as the complete trajectory of order ID456 from order placement to delivery station Y, to obtain the basic data for real-time monitoring audit output. From the audit output basic data, a data mapping method is used to generate structured transaction and logistics matching relationships, such as generating a matching record between order ID456 and delivery station Y, ultimately obtaining the audit output results for real-time monitoring. The final audit output results record the complete path from order placement to delivery, ensuring the accuracy and efficiency of monitoring.

[0159] As attached Figure 2As shown, based on the same inventive concept, this embodiment also discloses a digital auditing system for the supply chain, which executes the aforementioned digital auditing method for the supply chain, and includes the following modules:

[0160] The data acquisition and preliminary anomaly distribution module 1 takes into account real-time transaction records and logistics information of each node in the supply chain, acquires a dynamic dataset with timestamps through the collected information, determines the preliminary distribution of anomaly signals, and outputs the dynamic dataset and the preliminary distribution of anomaly signals.

[0161] Sliding window data fusion module 2 takes a dynamic dataset as input, processes the time-series data using the sliding window algorithm, obtains the fusion result of the latest transaction information and historical data within the window, and outputs the fusion result.

[0162] Risk assessment and quality indicator module 3: Input the fusion result; if the delivery delay signs exceed the preset threshold, determine the potential risk event; obtain the quality fluctuation correlation index; output the potential risk event and the quality fluctuation correlation index.

[0163] Anomaly detection and breakage location module 4 takes quality fluctuation correlation indicators as input, uses anomaly detection algorithm to analyze logistics anomaly patterns, determines the location of evidence chain breakage, and outputs the location of evidence chain breakage.

[0164] Audit window adjustment module 5: Input the break position, and the audit window will automatically slide through the adjustment mechanism to obtain the timeliness data-driven analysis framework and output the analysis framework;

[0165] The early warning sequence optimization module 6 takes the input analysis framework and, if it shows a risk prediction deviation, it uses a clustering algorithm to group the abnormal signals to obtain a signal cluster set and outputs an optimized early warning sequence.

[0166] The complete evidence chain determination module 7 takes the optimized early warning sequence as input, integrates the latest transaction information and logistics anomaly detection, determines the complete path of the continuous evidence chain, and outputs the complete path;

[0167] The audit result generation module 8 takes the complete path as input, processes legacy data checked at fixed intervals, obtains real-time monitoring audit output results, and outputs the audit output results.

Claims

1. A method for digitalized audit of a supply chain, characterized by, The method comprises the following steps: Step S100, acquiring a dynamic data set containing a time stamp by collecting real-time transaction records and logistics information of each node in the supply chain, and determining a preliminary distribution of abnormal signals therein; Step S200, processing time series data by using a sliding window algorithm according to the acquired dynamic data set, and obtaining a fusion result of the latest transaction information and historical data in the window; Step S300, if a delivery delay sign in the fusion result exceeds a preset threshold, judging that a potential risk event occurs, and obtaining a quality fluctuation correlation index corresponding to the potential risk event; if the delivery delay sign in the fusion result does not exceed the preset threshold, staying at this step; Step S400, analyzing a mode of logistics abnormality capture by using an abnormality detection algorithm according to the obtained quality fluctuation correlation index, and determining a position where a chain of evidence is broken; Step S500, adjusting a mechanism to automatically slide an audit window according to the determined position where the chain of evidence is broken, and obtaining an updated time-sensitive data dominant analysis framework; if the updated framework does not show a risk prediction deviation, staying at this step; Step S600, obtaining an optimized prediction and early warning sequence based on the time-sensitive data dominant analysis framework; Step S700, determining a complete path of a continuous chain of evidence by fusing the latest transaction information and the logistics abnormality capture according to the obtained optimized prediction and early warning sequence; Step S800, processing legacy data of fixed-period inspection according to the determined complete path of the continuous chain of evidence, and obtaining an audit output result of real-time monitoring; In the step S100, the dynamic data set containing the time stamp is acquired by collecting the real-time transaction records and the logistics information of each node in the supply chain, and the preliminary distribution of the abnormal signals is determined, which specifically comprises the following sub-steps: The real-time transaction records and the logistics information of each node in the supply chain are collected through a supply chain node interface, and the dynamic data set containing the time stamp is generated; The time sequence characteristics of the transaction records and the logistics information are extracted from the dynamic data set by using time series analysis, and the abnormal signals are obtained; If the time sequence characteristics of the abnormal signals exceed a preset threshold, the abnormal signals are classified by using a clustering algorithm, and the abnormal categories are determined; The correlation between the transaction records and the logistics information is analyzed by using a decision tree algorithm according to the abnormal categories, and a distribution mode of the abnormal signals is obtained; The frequency and the intensity of the abnormal signals in the dynamic data set are calculated through the distribution mode, and the preliminary distribution is determined; If the frequency or the intensity of the preliminary distribution exceeds a preset threshold, the future change trend of the abnormal signals is acquired by using a time series prediction algorithm; According to the change trend, the data collection frequency is adjusted, the monitoring parameters of the supply chain nodes are optimized, and the preliminary distribution of the abnormal signals is finally determined; In the step S500, the audit window is automatically slid by adjusting the mechanism according to the determined position where the chain of evidence is broken, and the updated time-sensitive data dominant analysis framework is obtained, which specifically comprises the following sub-steps: For the determined position where the chain of evidence is broken, abnormal node data is acquired therefrom, a fluctuation period of node transaction is extracted by using time series analysis, and a periodic abnormal mode is determined; According to the periodic anomaly mode, the clustering algorithm is used to group the supply chain nodes to obtain an abnormal node cluster; From the abnormal node cluster, the transaction information deviation is obtained, the deviation distribution is calculated by statistical analysis, and the deviation concentration area is judged; If the deviation concentration area exceeds the preset threshold, the time series decomposition method is used to extract the time sequence characteristics of the abnormal node to determine the abnormal duration; According to the abnormal duration, the time span of the sliding window is adjusted to obtain the updated audit time period; From the updated audit time period, the transaction data is obtained, and the data is processed by the anomaly detection algorithm to determine the optimized evidence chain breaking position; According to the optimized evidence chain breaking position, a time-effectiveness-dominated analysis logic is generated, and finally an updated time-effectiveness data-dominated analysis framework is obtained; In step S600, based on the time-effectiveness data-dominated analysis framework, the optimized prediction and early warning sequence is obtained, which specifically includes the following sub-steps: According to the risk prediction deviation displayed by the updated time-effectiveness data-dominated analysis framework, the abnormal signals are grouped by clustering algorithm according to the abnormal signal grouping basis to obtain a signal cluster set; From the signal cluster set, the signal intensity distribution is obtained, the distribution characteristics are calculated by statistical analysis method, and the concentrated distribution area is judged; If the concentrated distribution area exceeds the preset threshold, the dynamic change trend of the signal cluster is extracted by time series analysis to determine the change trend; According to the change trend, the input parameters of the prediction model are adjusted to obtain an optimized prediction parameter set; From the optimized prediction parameter set, a prediction and early warning sequence is generated by regression analysis method to obtain early warning sequence data; According to the early warning sequence data, the risk prediction framework is adjusted by the deviation correction method to obtain an updated risk prediction model; From the updated risk prediction model, the priority ranking of abnormal signals is extracted, and finally an optimized prediction and early warning sequence is obtained.

2. The method for digital audit of a supply chain according to claim 1, characterized in that, In step S200, according to the obtained dynamic data set, the sliding window algorithm is used to process the time series data to obtain the fusion result of the latest transaction information and historical data in the window, which specifically includes the following sub-steps: According to the obtained dynamic data set, the sliding window algorithm is used to traverse the local mode of the time series data contained therein, and the statistical indicators of the transaction information in the window are extracted to obtain the statistical characteristics of the local mode; If the statistical characteristics of the local mode deviate from the preset threshold, the feature dimension is compressed by principal component analysis algorithm to obtain a reduced feature set; According to the reduced feature set, the k-means clustering algorithm is used to group the local mode to obtain the clustering result of the transaction information; Through the clustering result, the timestamp interval distribution of each group of transaction information is calculated to determine the periodic mode of the time series; If the timestamp interval of the periodic mode exceeds the preset threshold, the moving average algorithm is used to smooth the time series data to obtain the smoothed sequence trend; According to the smoothed sequence trend, the size and step length of the sliding window are adjusted to obtain the optimized time series local mode; Through the optimized time series local mode, the dynamic characteristics of the transaction information and historical data fusion are extracted, and finally the fusion result of the latest transaction information and historical data in the window is obtained.

3. The method for digital audit of a supply chain according to claim 1, wherein, In the step S300, if the delivery delay sign in the fusion result exceeds the preset threshold, a potential risk event is determined, and the corresponding quality fluctuation correlation index is obtained; if the delivery delay sign in the fusion result does not exceed the preset threshold, the step is stayed, and the following sub-steps are specifically included: For the obtained fusion result, if the delivery delay sign exceeds the preset threshold, the distribution characteristics of the abnormal points are extracted through time series analysis to determine a potential risk event; According to the distribution characteristics of the abnormal points, principal component analysis algorithm is used to compress the quality fluctuation data to obtain a feature set after dimension reduction; The correlation strength between the supply chain nodes is calculated through the feature set after dimension reduction to obtain a dynamic correlation matrix between the nodes; If the correlation strength in the dynamic correlation matrix is lower than the preset threshold, moving average algorithm is used to smooth the time series data to obtain a smoothed trend sequence; According to the smoothed trend sequence, the fluctuation frequency of the periodic pattern is extracted to determine the abnormal distribution pattern of the supply chain nodes; Through the abnormal distribution pattern, the transaction information deviation value of each node is calculated to obtain a quantitative index of the potential risk; According to the quantitative index of the potential risk, the window parameter of the time series analysis is adjusted to obtain an optimized risk prediction model, and finally the corresponding quality fluctuation correlation index is obtained.

4. The method for digitalized audit of a supply chain according to claim 1, characterized in that, In the step S400, the obtained index is used to analyze the mode of the logistics anomaly capture by using an anomaly detection algorithm to determine the occurrence position of the evidence chain breakage, and the following sub-steps are specifically included: For the obtained quality fluctuation correlation index, an anomaly detection algorithm is used to process the associated logistics data, an abnormal pattern is extracted through an isolation forest algorithm, and abnormal distribution characteristics are obtained; According to the abnormal distribution characteristics, the transaction information deviation between the supply chain nodes is calculated to obtain a node abnormal value; If the node abnormal value exceeds the preset threshold, the time trend of the node transaction is extracted through a time series decomposition method to determine an abnormal fluctuation pattern; According to the abnormal fluctuation pattern, a clustering algorithm is used to group the supply chain nodes to obtain a node abnormal cluster; Through the node abnormal cluster, the statistical deviation of the transaction information in each cluster is calculated to determine the evidence chain breakage position; According to the evidence chain breakage position, the sensitivity parameter of the anomaly detection algorithm is adjusted to obtain an optimized anomaly detection model; Through the optimized anomaly detection model, the logistics data is reprocessed to finally determine the occurrence position of the evidence chain breakage.

5. The method for digitalized audit of a supply chain according to claim 1, characterized in that, In the step S700, the obtained sequence is used to fuse the latest transaction information and the logistics anomaly capture to determine the complete path of the continuous evidence chain, and the following sub-steps are specifically included: For the obtained optimized prediction and early warning sequence, the transaction information sequence and the logistics anomaly signal are extracted therefrom, multi-source data is integrated by using a data fusion rule, and a fusion data set is obtained; From the fusion data set, the transaction timestamp and the logistics state update are extracted, time series analysis is used to determine the time sequence correlation characteristics, and a time sequence correlation set is obtained; If the time sequence correlation set satisfies the abnormal trigger condition, the length of the evidence chain is analyzed through path consistency verification to determine the continuous evidence path; According to the continuous evidence path, a correlation rule mining algorithm is used to extract the transaction correlation identifier and the logistics node sequence to obtain a correlation path set; From the association path set, the path consistency verification is grouped by using a clustering analysis method to obtain an optimized path cluster; By optimizing the path cluster, the effectiveness of the complete path structure is judged by using a preset threshold to determine the final complete path; From the final complete path, the matching relationship between the transaction association identifier and the logistics node sequence is extracted to finally determine the complete path of the continuity evidence chain.

6. The method for digital audit of a supply chain according to any of claims 1-5, characterized in that, In the step S800, according to the complete path, the remaining data of the fixed period check is processed to obtain the audit output result of real-time monitoring, specifically including the following sub-steps: For the determined complete path of the continuity evidence chain, the corresponding fixed period check remaining data is extracted, the effective transaction record and logistics state update are extracted by using the data cleaning rule to obtain a standardized data set; If the standardized data set meets the preset integrity threshold, the transaction timestamp and the logistics node sequence are extracted by time series analysis to obtain a time series association set; According to the time series association set, the matching relationship between the transaction and the logistics node is extracted by using an association rule mining algorithm to obtain an association path set; If the path consistency in the association path set meets the preset threshold, the paths are grouped by clustering analysis to obtain an optimized path cluster; From the optimized path cluster, the effectiveness of the path structure is judged by using a preset threshold to obtain an effective path set; According to the effective path set, the continuity features of the transaction timestamp and the logistics node sequence are extracted to obtain the audit output basic data of real-time monitoring; From the audit output basic data, a structured transaction and logistics matching relationship is generated by using a data mapping method to finally obtain the audit output result of real-time monitoring.

7. A digitized audit system for a supply chain, characterized by, The system comprises the following modules for performing the digital audit method for the supply chain as claimed in any one of claims 1-6: A data acquisition and abnormality preliminary distribution module inputs real-time transaction records and logistics information of each node of the supply chain, acquires a dynamic data set containing a timestamp by acquiring information, determines an abnormal signal preliminary distribution, and outputs the dynamic data set and the abnormal signal preliminary distribution; A sliding window data fusion module inputs the dynamic data set, processes time series data by using a sliding window algorithm to obtain a fusion result of the latest transaction information and historical data in the window, and outputs the fusion result; A risk judgment and quality index module inputs the fusion result, determines a potential risk event if a delivery delay indication exceeds a preset threshold, acquires a quality fluctuation correlation index, and outputs the potential risk event and the quality fluctuation correlation index; An abnormality detection and fracture positioning module inputs the quality fluctuation correlation index, analyzes a logistics abnormality mode by using an abnormality detection algorithm, determines an evidence chain fracture position, and outputs the evidence chain fracture position; An audit window adjustment module inputs the fracture position, automatically slides the audit window by an adjustment mechanism to obtain a time effectiveness data dominant analysis framework, and outputs the analysis framework; An early warning sequence optimization module inputs the analysis framework, groups abnormal signals by a clustering algorithm if a risk prediction deviation is displayed to obtain a signal cluster set, and outputs an optimized early warning sequence. A complete evidence chain determination module, input optimization early warning sequence, fusion latest transaction information and logistics anomaly capture, determine the continuity of the complete path of the evidence chain, output the complete path; An audit result generation module, input complete path, process fixed period check remaining data, get real-time monitoring audit output result, output audit output result.

Images