Supply chain full-link risk early warning platform and method based on big data

Abstract

The invention discloses a supply chain full-link risk early warning platform and method based on big data. The platform comprises a supply chain division module, a data acquisition module, a risk assessment module, an alternative supplier module, a risk path conduction module and a risk plan module. Supply chain information is divided according to link nodes, then multi-source data is collected, supplier quality information is generated in combination with supplier data, consumption behavior data and public opinion data, suppliers are bound, the number of alternative suppliers is dynamically adjusted based on the real-time fluctuation rate, and risk conduction paths containing path nodes of different risk levels are constructed. And finally, generating a risk plan adaptive to the link scene. According to the invention, through multi-dimensional data fusion and a dynamic threshold mechanism, the cascade effect capture capability of supply chain risk conduction is enhanced, and collaborative improvement of risk early warning timeliness and plan accuracy is realized.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a supply chain full-link risk warning platform and method based on big data. Background Art

[0002] Due to the globalization of the economy and the iterative update of technology, risk warning for the supply chain has emerged, helping enterprises to control risks for relatively critical supply demands. However, the existing risk prediction still has the following limitations: there are various types of data throughout the supply chain. Currently, risk prediction can only achieve data processing and prediction for a unified category, resulting in the singularity of risk assessment; the risk prediction of the supply chain relies on historical data processing and does not consider temporary and sudden situations. In some fields with time-sensitive requirements, especially the e-commerce field, due to sudden situations such as marketing and current events, historical data is not referenceable, making the risk prediction inaccurate. Summary of the Invention

[0003] In view of the above problems, the present invention provides a supply chain full-link risk warning platform and method based on big data, which solves the problems that the existing risk prediction has a single risk assessment in application and cannot accurately predict emergencies.

[0004] To achieve the above object, in the first aspect, the present application provides a supply chain full-link risk warning platform based on big data, including a supply chain division module, a data collection module, a risk assessment module, an alternative supplier module, a risk path conduction module, and a risk plan module. The supply chain division module is used to obtain supply chain information and divide the supply chain information according to link nodes to obtain multiple link node information. The data collection module is used to obtain multi-source data of each link node information, and the multi-source data includes supplier data, consumer behavior data, logistics timeliness data, geographical data, and public opinion data. The risk assessment module is used to generate risk information of each link node information according to the multi-source data, and generate supply quality information according to the consumer behavior data and public opinion data, and bind the supply quality information to the supplier. The alternative supplier module is used to screen a preset number of alternative suppliers in the database according to the risk information and the link node information, and update the alternative suppliers to the link node information, and the preset number is configured to be generated according to the real-time volatility of the current link node. The risk path conduction module is used to generate a risk conduction path according to the risk information and the supply chain information, and the risk conduction path includes multiple path nodes with different risk levels. The risk plan module is used to generate a risk plan according to the risk information, the risk conduction path, and the updated link node information.

[0005] In some embodiments, the data acquisition module includes a data standardization unit, a spatio-temporal association unit, and a real-time update unit. The data standardization unit is used to perform standardization processing on multi-source data to obtain standardized data, including: extracting first key fields from supplier data and converting the first key fields into first structured data entries; processing the return records in the consumer behavior data to obtain the category return rate, and performing NLP sentiment analysis on the text evaluations in the consumer behavior data to generate sentiment polarity scores; constructing a regional supplier mapping table from regional data, supplier data, and regional policies; extracting second key fields from the public opinion data and converting the second key fields into second structured data entries; converting the logistics timeliness data into third structured data entries; The spatio-temporal association unit is used to associate the standardized data with the link nodes; The real-time update unit is used to achieve dynamic update of multi-source data through a stream processing framework.

[0006] In some embodiments, the risk path conduction module includes a node association unit, a risk quantification unit, and a path adjustment unit. The node association unit is used to construct the logical dependency relationships of multiple link nodes to obtain a composite topology structure, including: generating an initial topology structure according to the node connection rules in the supply chain information; calculating the hierarchical associations between each pair of link nodes, constructing multi-directional association lines on the initial topology structure, and generating a composite topology structure; initializing the risk conduction coefficients for multiple link nodes in the composite topology structure; The risk quantification unit is used to perform risk quantification on multiple link nodes according to the risk information and the composite topology structure to obtain the risk increment of each link node and generate the final risk weight of each link node; The path adjustment unit is used to generate risk conduction paths with different risk levels according to the final risk weights and the composite topology structure.

[0007] In a second aspect, the present invention also provides a supply chain full-link risk warning method based on big data, which is applicable to the risk warning platform in the first aspect. The method includes: Obtaining supply chain information, dividing the supply chain information according to link nodes to obtain multiple link node information; Regularly obtaining multi-source data of each link node information according to a preset acquisition frequency and performing preprocessing, and mapping and storing the preprocessed multi-source data with the link nodes; Generate risk information for each link node information based on the preprocessed multi-source data, and generate supply product quality information based on consumption behavior data and public opinion data, and bind the supply product quality information to the supplier; Screen a preset number of alternative suppliers in the database according to the risk information and the link node information, and update the alternative suppliers to the link node information, and the preset number is configured to be generated according to the real-time volatility of the current link node; Generate a risk conduction path according to the risk information and the supply chain information, and the risk conduction path includes path nodes of multiple different risk levels; Generate a risk plan according to the risk information, the risk conduction path and the updated link node information.

[0008] In some embodiments, obtaining multi-source data of each link node information and performing preprocessing, and mapping and storing the preprocessed multi-source data with the link node includes: Extract the first key field from the supplier data, and convert the first key field into the first structured data entry; Process the return records in the consumption behavior data to obtain the category return rate, and perform NLP sentiment analysis on the text evaluation in the consumption behavior data to generate sentiment polarity scores; Construct a regional supplier mapping table from the regional data, the supplier data, and the regional policies; Extract the second key field from the public opinion data, and convert the second key field into the second structured data entry; Convert the logistics timeliness data into the third structured data entry; Generate node data information in a preset format according to the first structured data entry, the second structured data entry, the third structured data entry, the regional supplier mapping table, and the sentiment polarity score; Map and store the node data information with the link node information; And, evaluate whether the node data information meets the preset fluctuation condition, if not, adjust the preset collection frequency of the multi-source data.

[0009] In some embodiments, generating risk information for each link node information based on the preprocessed multi-source data includes: Sort the influence coefficients of the preprocessed multi-source data associated with the link node information according to the AHP algorithm to obtain the influence coefficients of each data type in the multi-source data; Convert the risk weights of each data type through a linear normalization function according to the influence coefficients, and record them as the initial risk weights; Calculate the risk probability for each data type, where the risk probability is configured to be calculated using the Poisson distribution model based on the number of risk occurrences in the historical data information of the same data type at the same link node; Fuse the risk probability with the initial risk weight to form the basic risk weight; Extract the burst information from the multi-source data through the CUSUM algorithm, where the burst information includes policy burst features, promotion burst features, and regional burst features; Convert the burst information into a burst risk weight using the fuzzy logic algorithm; Fuse the basic risk weight and the burst risk weight to obtain the risk information.

[0010] In some embodiments, according to the link node information, sort the preprocessed multi-source data associated therewith according to the influence coefficient by the AHP algorithm, and the influence coefficient of each data type in the multi-source data includes: Construct a judgment matrix corresponding to the node type of the current link node, and the matrix elements of the judgment matrix are generated by the relative importance of the data types of the multi-source data; Calculate the maximum eigenvalue of the judgment matrix and its associated eigenvector, and each component of the eigenvector corresponds to a data type; Normalize the eigenvector to obtain the influence coefficients of multiple data types; Calculate the risk probability for each data type, where the risk probability is configured to be calculated using the Poisson distribution model based on the number of risk occurrences in the historical data information of the same data type at the same link node, including: Statistically calculate the average number of risk events of the current data type per unit time according to the historical data information; Calculate the probability of at least one risk event occurring in the current time window through the Poisson distribution probability function as the risk probability, which is represented by formula (1), and formula (1) is as follows: ; In formula (1), is the risk probability, is the number of occurrences of the risk event, is the base of the natural logarithm, is the average number of risk events per unit time; Extract the burst information from the multi-source data through the CUSUM algorithm, where the burst information includes policy burst features, promotion burst features, and regional burst features, including: Construct a cumulative statistic for the time series data stream in the multi-source data, and mark it as a mutation point when the cumulative statistic exceeds the preset control limit, which is represented by formula (2), and the formula (2) is as follows: ; In formula (2), is the cumulative statistic at moment, is the cumulative statistic at moment, is the observed value at moment, is the reference mean value, is the allowable deviation parameter; Generate at least one of the policy burst feature, promotion burst feature, or regional burst feature according to the data type classification corresponding to the mutation point; Converting the burst information into a burst risk weight using a fuzzy logic algorithm includes: Establish a fuzzy rule base, where the input variables of the fuzzy rule base are the intensity of the burst feature and the duration of the burst feature, and the output variable of the fuzzy rule base is the burst risk weight; Convert the fuzzy inference result into a numerical burst risk weight through defuzzification operation.

[0011] In some embodiments, generating supply product quality information according to consumption behavior data and public opinion data, and binding the supply product quality information to the supplier includes: Perform data processing on the return records in the consumption behavior data, and count the category return rates of each supplier; Extract the text evaluations in the consumption behavior data for NLP sentiment analysis to generate the quality satisfaction score of the supplier; Extract keyword fields from the public opinion data to identify quality events associated with the supplier; Calculate the public opinion risk value according to the occurrence frequency and spread range of the quality event; Weightedly fuse the category return rate, quality satisfaction score, and public opinion risk value to generate a comprehensive quality score; Divide the quality level according to the comprehensive quality score to obtain the supply product quality information; Associate and store the supply product quality information with the corresponding supplier; And, when the quality level is lower than the preset level threshold, trigger the screening mechanism for alternative suppliers; And, update the supply product quality information according to the newly added consumption behavior data and public opinion data at a preset cycle.

[0012] In some embodiments, screen a preset number of alternative suppliers in the database according to the risk information and link node information, and update the alternative suppliers to the link node information, and the preset number is configured to be generated according to the real-time volatility of the current link node, including: Calculate the real-time volatility of the current link node, which is obtained by converting the degree of change of key indicators in the link node information; Dynamically adjust the preset quantity according to the magnitude of the real-time volatility; Sort the suppliers to be replaced according to preset screening conditions in the database. The preset screening conditions include geographical coverage matching degree, supply quality grade, historical delivery on-time rate, and cost premium; Select the top-ranked preset quantity of suppliers to be replaced, denoted as alternative suppliers, and update them to the link node information.

[0013] In some embodiments, generate a risk conduction path according to risk information and supply chain information. The risk conduction path includes multiple path nodes with different risk levels, including: Generate an initial topological structure according to the node connection rules in the supply chain information; Calculate the hierarchical correlation degree between each link node. The hierarchical correlation degree is configured to be generated by the physical connection strength, data interaction frequency, and business dependence degree between multiple link nodes; Construct multi-directional association lines on the initial topological structure according to the hierarchical correlation degree to generate a composite topological structure; Initialize the risk conduction coefficient for multiple link nodes in the composite topological structure. The risk conduction coefficient is configured to be obtained through the following steps: Calculate the conduction influence weight of the previous link node and the subsequent link node of the current link node; And calculate the collaborative risk weight of other link nodes at the same level as the current link node; And calculate the indirect conduction attenuation weight of multiple link nodes across levels and the current link node; Fuse the conduction influence weight, collaborative risk weight, and indirect conduction attenuation weight to obtain the risk conduction coefficient; Perform risk quantification on multiple link nodes according to the risk information and the composite topological structure to obtain the risk increment of each link node, and generate the final risk weight of each link node, which is represented by formula (3). Formula (3) is as follows: ; In formula (3), is the final risk weight of the current link node, is the initial risk value of the current link node, is the risk conduction coefficient, is the risk increment; Generate a risk conduction path with different risk levels according to the final risk weight and the composite topological structure.

[0014] Different from the prior art, the above technical solution provides a supply chain full-link risk early warning platform and method based on big data. The platform includes a supply chain division module, a data collection module, a risk assessment module, an alternative supplier module, a risk path conduction module, and a risk pre-plan module. By dividing the supply chain information according to link nodes and collecting multi-source data, combining supplier data, consumer behavior data, and public opinion data to generate supply product quality information and bind suppliers, dynamically adjusting the number of alternative suppliers based on real-time volatility, constructing a risk conduction path including path nodes of different risk levels, and finally generating a risk pre-plan adapted to the link scenario. The above technical solution strengthens the ability to capture the cascading effect of supply chain risk conduction through multi-dimensional data fusion and dynamic threshold mechanism, and realizes the coordinated improvement of the timeliness of risk early warning and the accuracy of the pre-plan.

[0015] The above relevant records of the invention content are only an overview of the technical solution of this application. In order to enable those of ordinary skill in the art to more clearly understand the technical solution of this application, and then can be implemented according to the content recorded in the description and the drawings, and in order to make the above objects, other objects, features and advantages of this application more easily understood, the following is described in conjunction with the specific implementation manners and drawings of this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings are only used to illustrate the principles, implementation methods, applications, features, and effects of the specific implementation manners of the present invention and other related contents, and should not be considered as a limitation to this application.

[0017] In the accompanying drawings of the specification: Figure 1 It is a structural schematic diagram of the risk early warning platform described in the specific implementation manner; Figure 2 It is a method step diagram of steps S101 to S106 of the risk early warning method described in the specific implementation manner; Figure 3 It is a method step diagram of steps S201 to S207 of the risk early warning method described in the specific implementation manner; Figure 4 It is a method step diagram of steps S301 to S307 of the risk early warning method described in the specific implementation manner; Figure 5 It is a method step diagram of steps S401 to S407 of the risk early warning method described in the specific implementation manner.

[0018] The descriptions of the reference numerals involved in the above drawings are as follows: 1. Risk early warning platform; 11. Supply chain division module; 12. Data collection module; 13. Risk assessment module; 14. Alternative supplier module; 15. Risk path conduction module; 16. Risk pre-plan module. Detailed implementation manners

[0019] To describe in detail the possible application scenarios, technical principles, specific implementable solutions, achievable purposes and effects of this application, etc., the following will be described in detail with reference to the specific examples listed and in conjunction with the accompanying drawings. The embodiments described herein are only used to more clearly illustrate the technical solutions of this application, so they are only examples and cannot be used to limit the protection scope of this application.

[0020] Referring to "embodiments" herein means that the specific features, structures or characteristics described in connection with the embodiments can be included in at least one embodiment of this application. The term "embodiment" appearing in various positions in the specification does not necessarily refer to the same embodiment, nor does it particularly limit its independence or relevance to other embodiments. In principle, in this application, as long as there is no technical contradiction or conflict, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0021] Unless otherwise defined, the meanings of the technical terms used herein are the same as those generally understood by those skilled in the technical field to which this application belongs; the use of the relevant terms herein is only for describing specific embodiments and is not intended to limit this application.

[0022] In the description of this application, the phrase "and / or" is an expression used to describe the logical relationship between objects, indicating that there can be three relationships. For example, A and / or B means: there is A, there is B, and there is both A and B at the same time. In addition, the character " / " in this article generally represents an "or" logical relationship between the associated objects before and after.

[0023] In this application, terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual quantity, primary or secondary, or order relationship between these entities or operations.

[0024] Without more limitations, in this application, the open expressions such as "including", "comprising", "having" or other similar expressions used in the statements are intended to cover non-exclusive inclusion. These expressions do not exclude that there may be other elements in the process, method or product including the said elements, so that the process, method or product including a series of elements may not only include those defined elements, but also include other elements not explicitly listed, or also include elements inherent to this process, method or product.

[0025] Similar to the understanding in the "Examination Guidelines", in this application, expressions such as "greater than", "less than", "exceeding", etc. are understood to exclude the base number; expressions such as "above", "below", "within", etc. are understood to include the base number. In addition, in the description of the embodiments of this application, the meaning of "a plurality of" is two or more (including two). Similar expressions related to "many", such as "multiple groups", "multiple times", etc., are understood in this way unless otherwise specifically defined.

[0026] In the description of the embodiments of this application, the spatially related expressions used, such as "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "perpendicular", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., the indicated orientation or positional relationship is based on the orientation or positional relationship shown in the specific embodiment or the accompanying drawings. It is only for the convenience of describing the specific embodiments of this application or for the reader to understand, rather than indicating or implying that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, it should not be construed as a limitation on the embodiments of this application.

[0027] The processor described in the embodiments of this application can be implemented by hardware, firmware, software, or a combination thereof. It can use circuits, one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), central processing units (CPUs), controllers, microcontrollers, microprocessors, etc. It also includes other physical, biological, or chemical structures that can achieve functions similar to or equivalent to the above-listed processors, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute some steps, all steps, or any combination of the steps mentioned in the computer programs or methods involved in the various embodiments of this application.

[0028] The computer program involved in the embodiments can be stored in a computer-readable storage medium, which includes but is not limited to magnetic disks, magnetic tapes, magnetic cards, floppy disks, flash memories, optical discs, optical cards, read-only memories (ROMs), random access memories (RAMs), erasable programmable ROMs (EPROMs), and electrically erasable programmable ROMs (EEPROMs), etc. It also includes other biological, physical, or chemical structures that can achieve the same or equivalent functions as the above-listed storage media, such as units with information storage capabilities like DNA, RNA, proteins, etc. In a specific embodiment, the storage medium involved can be one of the above medium types or a combination of the above medium types. In different embodiments, the computer program involved in the embodiments can be stored centrally in a single medium or distributed among multiple media. The memory containing the computer-readable storage medium can be a non-volatile memory or a random access memory. These computer-readable storage media can be built into the device or connected to the device involved in the embodiments as an external device or a part of an external device. In some embodiments, the memory with the computer-readable storage medium is deployed locally; in other embodiments, a scheme of deploying the memory away from the processor can also be adopted, such as a network-attached memory accessed via an RF circuit or an external port and a communication network, where the communication network can be the Internet, one or more internal networks, local area networks (LANs), wide area wireless networks (WLANs), storage area networks (SANs), etc., or a suitable combination thereof, as long as the computer device can access the memory. In addition, the computer program involved in the embodiments can be stored in plaintext / ciphertext form or designed as training data and integrated and recombinantly stored implicitly in the parameter states of a deep neural network or other machine learning models through model training.

[0029] Please refer to Figure 1 , in a first aspect, this embodiment provides a big data-based full-link risk warning platform 1 for the supply chain, including a supply chain division module 11, a data collection module 12, a risk assessment module 13, an alternative supplier module 14, a risk path conduction module 15, and a risk plan module 16. The supply chain division module 11 is used to obtain supply chain information and divide the supply chain information according to link nodes to obtain multiple link node information. The data collection module 12 is used to obtain multi-source data of each link node information, and the multi-source data includes supplier data, consumer behavior data, logistics timeliness data, geographical data, and public opinion data. The risk assessment module 13 is used to generate risk information for each link node information based on the multi-source data, and generate supply quality information based on the consumer behavior data and public opinion data, and bind the supply quality information to the supplier. The candidate supplier module 14 is used to screen a preset number of candidate suppliers in the database according to the risk information and the link node information, and update the candidate suppliers to the link node information, and the preset number is configured to be generated according to the real-time volatility of the current link node; The risk path transmission module 15 is used to generate a risk transmission path according to the risk information and the supply chain information. The risk transmission path includes a plurality of path nodes of different risk levels. The risk plan module 16 is used to generate a risk plan based on risk information, risk transmission paths and updated link node information.

[0030] In the supply chain segmentation module 11, the link node information includes merchant management nodes, procurement and inventory nodes, order flow nodes, warehousing and sorting nodes, logistics distribution nodes, after-sales nodes and consumer feedback nodes. Risks are segmented according to link node information to achieve refined early warning. Among them, the consumer feedback node is used to collect user evaluation texts and return records to generate supplier quality information.

[0031] In the risk assessment module 13, preferably, the supplier data in the multi-source data includes qualification documents, on-time delivery rate and quality sampling results; consumer behavior data refers to data such as consumer purchase style, purchase quantity, purchase age, transaction success rate, etc.; public opinion data can be divided into three categories: current affairs public opinion, activity public opinion and negative review public opinion, among which, current affairs public opinion refers to emergencies, such as events caused by local policies, public panic caused by consumer accidents and reduction of this demand, etc., activity public opinion is such as a certain promotional event, and negative review public opinion refers to consumers' negative review data, including but not limited to negative logistics reviews, negative product quality reviews, etc., that is, supplier quality information. Preferably, regional data can construct a regional supplier mapping table through the mapping relationship between the supplier's location and the business coverage area.

[0032] In the alternative supplier module 14, the preset quantity is dynamically adjusted according to the real-time volatility, which can be understood as the alternative supplier module 14 automatically calculates the required number of alternative suppliers based on the real-time volatility of the current link node (such as the inventory turnover rate deviation value, the logistics time fluctuation range), and the higher the real-time volatility, the larger the preset quantity. Preferably, the screening conditions of the alternative suppliers include geographic coverage matching, quality grade and historical delivery on-time rate, wherein the geographic coverage matching is calculated by the overlap rate between the supplier's service scope and the business area of ​​the current node.

[0033] In the risk path transmission module 15, preferably, the generation of the risk transmission path depends on the logical dependency between the link node information of the composite topology structure modeling. Further, when the supplier quality information cannot meet the requirements, the alternative supplier screening mechanism is triggered to realize the dynamic risk path and improve the resilience.

[0034] The risk response plan module 16 generates response measures based on the associated impacts of different-level nodes in the risk conduction path. For example, if the risk information is that the inventory of a popular product is insufficient during Prime Day and the alternative supplier cannot replenish the stock within 7 days, the procurement node predicts the demand based on historical flash sale data and triggers the alternative supplier in advance (such as overseas warehouse in-stock). If the risk information is that a large number of negative reviews have emerged for a product due to "battery fire" and the product faces the risk of being taken off the shelves, the consumer feedback node monitors keywords in real time and triggers the quality risk response plan (such as suspending procurement and contacting the supplier for inspection). If it is necessary to further focus on a certain type of e-commerce (such as cross-border e-commerce / DTC brand), the corresponding nodes can be refined (such as customs clearance, overseas warehouse).

[0035] The full-link risk warning platform 1 for the supply chain provided in this embodiment realizes the closed-loop control of risk identification, conduction modeling, and dynamic response through a modular architecture. The supply chain division module 11 divides the supply chain information into independent link nodes according to business processes (such as procurement nodes, logistics nodes), and each node is associated with specific data sources and risk indicators. The data collection module 12 integrates multi-source data, constructs structured data entries through standardized processing, and provides input for risk assessment. The risk assessment module 13 generates risk information for link nodes based on multi-source data fusion, extracts quality characteristics from consumer behavior and public opinion data to generate supply product quality information, and realizes the dynamic binding of suppliers and risk indicators. The alternative supplier module 14 dynamically adjusts the number of alternative suppliers according to the real-time volatility of the current node, and screens the optimal alternative resources according to conditions such as geographical coverage matching degree and quality level. The risk path conduction module 15 models the logical dependence relationship between nodes through a composite topology structure (such as the business dependence intensity between the procurement node and the warehousing node), quantifies the risk cross-node conduction effect, and generates conduction paths with different risk levels. The risk response plan module 16 outputs multi-level response strategies based on the node risk weights and alternative supplier information in the conduction path, such as the coordinated execution of supplier switching and inventory allocation.

[0036] Furthermore, the above technical solution can be understood in combination with the following example: In the cross-border e-commerce scenario (such as Amazon, Taobao, etc.), when the customs clearance node triggers a risk due to a sudden policy change, the geographical supplier mapping table screens alternative suppliers that meet the new policy. The risk path conduction module 15 calculates the indirect conduction attenuation weight of this event on the overseas warehouse node and generates a cross-level risk conduction path. The risk response plan module 16 outputs a composite plan including customs clearance material update, overseas warehouse allocation, and supplier switching in combination with the inventory turnover data of the warehousing node and the replenishment timeliness of the alternative supplier.

[0037] In this embodiment, multi-dimensional risk early warning and dynamic response are realized through a modular architecture. The supply chain division module 11 disassembles the supply chain into independent link nodes according to business processes, realizing refined positioning of risk sources; the data acquisition module 12 integrates multi-source data, enhancing the comprehensiveness of the risk assessment dimension; the risk assessment module 13 fuses multi-source data to generate link node risk information and realizes dynamic binding of suppliers and risk indicators, enhancing the accuracy of risk identification; the alternative supplier module 14 dynamically adjusts the number of alternative suppliers and further screens the optimal alternative resources, improving the scenario adaptability of supplier resources; the risk path conduction module 15 accurately captures the cascade risk diffusion path; the risk pre-plan module 16 ensures a high degree of fit between the risk pre-plan and the real-time business scenario, effectively improving the accuracy of risk identification, and realizing the coordinated optimization of the timeliness of risk early warning and the accuracy of risk pre-plans.

[0038] In some embodiments, the data acquisition module 12 includes a data standardization unit, a spatio-temporal association unit, and a real-time update unit. The data standardization unit is used to perform standardization processing on multi-source data to obtain standardized data, including: Performing first keyword field extraction on supplier data and converting the first keyword field into a first structured data entry; Performing data processing on the return records in the consumer behavior data to obtain the category return rate, and performing NLP sentiment analysis on the text evaluations in the consumer behavior data to generate sentiment polarity scores; Constructing a regional supplier mapping table from the regional data, supplier data, and regional policies; Performing second keyword field extraction on the public opinion data and converting the second keyword field into a second structured data entry; Converting the logistics timeliness data into a third structured data entry; The spatio-temporal association unit is used to associate the standardized data with the link nodes; The real-time update unit is used to realize dynamic update of multi-source data through a stream processing framework.

[0039] In this embodiment, the data standardization unit refers to a functional module that performs unified format conversion and key information extraction on multi-source heterogeneous data. The first keyword field extraction is to convert the unstructured content in the supplier data into structured data entries. For example, parsing a PDF format inspection report into a data table containing the inspection date and results. The category return rate is obtained by classifying and counting the return records in the consumer behavior data by commodity category. For example, counting the proportion of returns caused by "logistics damage" in the "household appliances" category; the NLP sentiment analysis of the text evaluation refers to semantic parsing of the user comment text and generating sentiment polarity scores (such as numerical scores in the range of -1 to 1) to quantify consumer satisfaction.

[0040] The regional supplier mapping table is constructed by associating the geographical location of the supplier, the business coverage area, and regional policies. For example, suppliers meeting the import qualifications of a certain country are bound to the target clearance nodes.

[0041] Preferably, the extraction of the second keyword field is for the structured storage of the description of hot events, the scope of dissemination, and associated entities (such as brand names) in the public opinion data.

[0042] The third structured data entry is to convert the original trajectory information (such as "received", "in transit") in the logistics timeliness data into unified fields (such as node status, estimated delay days) to facilitate cross-node timeliness comparison and analysis.

[0043] The spatio-temporal association unit is used to dynamically bind the standardized data to the link nodes according to the business logic. For example, the on-time delivery rate data of a certain supplier is associated with the procurement node, or the logistics delay data of a certain region is mapped to the corresponding distribution node.

[0044] The real-time update unit realizes the dynamic synchronization of data through a stream processing framework. For example, when the supplier qualification document changes, the weight calculation of the risk assessment module 13 is triggered in real time; or when the keyword "new customs policy" appears in the public opinion data, the regional supplier mapping table is updated immediately and the conduction risk of the associated nodes is marked.

[0045] This embodiment can be understood as follows: eliminating the format barriers of multi-source heterogeneous data through data standardization to construct a unified risk assessment input; ensuring the accurate matching of data with business nodes through spatio-temporal association; and guaranteeing the timeliness of risk indicators through a real-time update mechanism. For example, when the quality inspection results of a certain supplier change due to the update of the quality inspection report, the data standardization unit extracts the latest pass rate and updates the structured entry, the real-time update unit triggers a re-assessment of the risk at the procurement node, and the spatio-temporal association unit synchronizes the updated data to the relevant link nodes, ultimately supporting the dynamic response of risk early warning. The data standardization unit eliminates the data integration barrier, the spatio-temporal association unit realizes the accurate mapping between business links and data sources, and the real-time update unit ensures the timeliness of risk assessment indicators, thereby improving the real-time response accuracy of risk early warning and the cross-node data collaboration efficiency, and strengthening the dynamic control ability of supply chain risks.

[0046] In some embodiments, the risk path conduction module 15 includes a node association unit, a risk quantification unit, and a path adjustment unit. The node association unit is used to construct the logical dependency relationship of multiple link nodes to obtain a composite topology structure, including: Generating an initial topology structure according to the node connection rules in the supply chain information; Calculating the hierarchical association between each link node, constructing multi-directional association lines on the initial topology structure, and generating a composite topology structure; Initialize the risk conduction coefficient for multiple link nodes in the composite topology structure; The risk quantification unit is used to quantify the risks of multiple link nodes according to the risk information and the composite topology structure, obtain the risk increment of each link node, and generate the final risk weight of each link node; The path adjustment unit is used to generate risk conduction paths with different risk levels according to the final risk weights and the composite topology structure.

[0047] In this embodiment, the initial topology structure represents the basic service flow between nodes; preferably, calculating the hierarchical association between each link node can be understood as quantifying the potential of the conduction impact between nodes by analyzing the physical connection strength between link nodes (such as the number of transportation lines between logistics nodes and distribution nodes), the data interaction frequency (such as the number of inventory synchronization times between order nodes and warehouse nodes), and the degree of business dependence (such as the proportion of raw material supply from supplier nodes to production nodes); constructing multi-directional association lines on the initial topology structure can be understood as superimposing cross-level and cross-business functional connection paths (such as the reverse logistics channel where the after-sales node is directly associated with the supplier node) on the initial topology structure to form a composite topology structure to support multi-dimensional risk conduction modeling.

[0048] The risk conduction coefficient is a parameter that quantifies the risk transfer ability between nodes in the composite topology structure. Preferably, it is obtained by fusing the conduction impact weight of the previous node on the current node (such as the impact ratio of procurement delay on warehouse out-of-stock), the collaborative risk weight of nodes at the same level (such as the superposition effect of multiple logistics nodes being delayed simultaneously), and the indirect conduction attenuation weight across levels (such as the cross-regional impact attenuation rate of sudden changes in customs clearance policies on overseas warehouse nodes).

[0049] The risk increment is to calculate the risk diffusion value of the node to the associated node based on the risk information of the current node (such as the probability of supplier delivery delay) and the risk conduction coefficient in the composite topology structure; the final risk weight can be used to divide the risk levels of nodes.

[0050] The path adjustment unit generates a risk conduction path according to the final risk weight and the associated lines in the composite topology structure. For example, high-risk nodes with weights exceeding the threshold are marked as key path nodes, and the conduction chain is traced along the multi-directional association lines to form a visual path containing nodes with different risk levels.

[0051] This embodiment can be understood as follows: By modeling the multi-dimensional correlation relationships between nodes through a composite topological structure, the limitations of the unidirectional conduction model are broken through; the risk quantification unit combines static risk values with dynamic conduction effects to accurately depict the cascading risk diffusion path; the path adjustment unit dynamically optimizes the conduction path based on the weight threshold to improve the pertinence of risk positioning and intervention. For example, in the cross-border e-commerce scenario, when the customs clearance node triggers a risk due to policy adjustments, the node association unit constructs a cross-level association line between it and the overseas warehouse node, the risk quantification unit calculates the conduction increment of customs clearance delays on warehousing inventory, and the path adjustment unit generates a conduction path including customs clearance, warehousing, and logistics nodes, providing a basis for the risk pre-plan module 16. This embodiment models the logical dependency relationships and conduction effects between link nodes in multiple dimensions through a composite topological structure, dynamically quantifies the cascading risk weights of nodes in combination with risk increments, and generates multi-level risk conduction paths based on the weight threshold, realizing the accurate positioning and dynamic intervention of the risk conduction link, and comprehensively improving the real-time performance of supply chain cascading risk early warning and the cross-node collaborative prevention and control ability.

[0052] Please refer to Figure 2 , in the second aspect, this embodiment also provides a big data-based supply chain full-link risk early warning method, which is applicable to the risk early warning platform in the first aspect. The method includes: S101. Obtain supply chain information, divide the supply chain information according to link nodes, and obtain multiple link node information; S102. Regularly obtain multi-source data of each link node information at a preset collection frequency and perform preprocessing, and map and store the preprocessed multi-source data with the link nodes; S103. Generate risk information for each link node information based on the preprocessed multi-source data, and generate supply commodity quality information based on consumption behavior data and public opinion data, and bind the supply commodity quality information to the supplier; S104. Screen a preset number of alternative suppliers in the database according to the risk information and the link node information, and update the alternative suppliers to the link node information. The preset number is configured to be generated according to the real-time volatility of the current link node; S105. Generate a risk conduction path according to the risk information and the supply chain information. The risk conduction path includes multiple path nodes with different risk levels; S106. Generate a risk pre-plan according to the risk information, the risk conduction path, and the updated link node information.

[0053] In step S101, the link node division is completed through preset business rules. The preset business rules can be the upstream and downstream dependency relationships between order transfer nodes and warehousing sorting nodes. For example, the customs clearance node and the overseas warehouse node of the Amazon platform are divided as independent units.

[0054] In step S102, the format differences of heterogeneous data are eliminated through preprocessing. Furthermore, the multi-source data can be converted into a unified structured data entry through standardization to ensure the input consistency and calculation accuracy of the risk assessment module. The preprocessed multi-source data is bound and stored with the corresponding link node through the spatiotemporal association unit in the aforementioned embodiment, for example, the on-time delivery rate of a supplier is mapped to the procurement node.

[0055] In step S103, risk information is a quantitative indicator of node risk calculated based on multi-source data. The process of generating risk information can be understood as calculating the risk probability of link node information based on pre-processed multi-source data, fusing historical risk rules and real-time emergency characteristics through weights, and outputting dynamic risk values. For example, when the timeliness deviation rate of a logistics node exceeds the threshold due to sudden weather, a high risk level is generated in combination with the historical delay probability. Preferably, the process of generating and binding supplier quality information can be understood as integrating the return rate, evaluation sentiment polarity and public opinion propagation range in consumer behavior data, weightedly calculating the comprehensive quality score, dividing the levels according to the threshold, and binding to the supplier's unique identifier. For example, a supplier's quality level is downgraded and bound to the supplier's file because the negative review sentiment score is lower than -0.5 and the spread exceeds 100,000 times.

[0056] In step S104, the higher the real-time volatility, the worse the node business stability, and the more alternative suppliers need to be added to cope with potential disruption risks and improve the supply chain's resilience to risks. Preferably, the screening conditions include geographic coverage matching (such as the overlap rate between the supplier's service area and the node's business area), quality level (such as ISO-certified supplier priority) and historical delivery on-time rate (such as the fulfillment rate of the past 90 days ≥ 95%).

[0057] In step S105, the process of generating a risk transmission path can be understood as analyzing the business relevance between the path nodes (such as the impact of the supply interruption of the procurement node on the storage node), combining the risk information (such as the probability of supplier delay) to simulate the risk diffusion direction, and forming a transmission chain from the risk source to the affected node. Preferably, each path node in the risk transmission path is graded according to the final risk weight, for example, divided into high risk, medium risk, and low risk, with high-risk nodes being directly impacted links, and medium-risk nodes and low-risk nodes being indirectly affected areas, so as to locate the core risk links and formulate priority intervention strategies.

[0058] In step S106, the risk plan is generated based on the risk level of the node in the risk transmission path and the information of the alternative suppliers. For example, when the customs clearance node causes high risk due to sudden policy changes, the plan module combines the replenishment time of the alternative supplier and the inventory data of the storage node to output the collaborative plan for customs clearance material update, overseas warehouse transfer and supplier switching.

[0059] In this embodiment, a collaborative mechanism of dynamic division of link nodes, cross-node risk conduction modeling, and dynamic update of alternative suppliers is adopted to achieve full-link risk prevention and control in the supply chain. The specific step process can be understood as follows: The supply chain is segmented into independent nodes according to business stages, and each node is associated with supply product quality information (such as quality sampling rate, policy compliance, etc.) and alternative suppliers in the database; in supplier management, a regional supplier mapping table can be constructed through regional data and public opinion data, access the impact of marginal policies (such as import and export restrictions caused by trade wars), and dynamically adjust the supplier qualification matching strategy. The screening of alternative suppliers is dynamically determined based on real-time volatility to a preset quantity, and the best ones are selected according to demand and cost performance. For example, when the tariff policy suddenly changes, regional suppliers that meet the new regulations are called to replace the original large-scale partners. The risk conduction path integrates public opinion risks (such as the spread heat of policy keywords), enterprise operation risks (such as supplier financial crises), and quality risks (such as unqualified sampling rates), and constructs a cascade diffusion model through the business dependence intensity between nodes (such as the inventory supply relationship between the procurement node and the warehousing node), divides high-risk, medium-risk, and low-risk level nodes, and locates the core risk sources. Finally, the final plan module generates short-term emergency plans (such as rapid update of customs clearance materials, emergency replenishment of regional suppliers) according to the priority of the conduction path and alternative supplier resources, realizing the dynamic adaptation of risk response and business scenarios.

[0060] Please refer to Figure 3 , in some embodiments, obtaining multi-source data of each link node information and performing preprocessing, and mapping and storing the preprocessed multi-source data with the link node includes: S201. Extract the first key field from the supplier data and convert the first key field into the first structured data entry; S202. Process the return records in the consumer behavior data to obtain the category return rate, and perform NLP sentiment analysis on the text evaluation in the consumer behavior data to generate sentiment polarity scores; S203. Construct a regional supplier mapping table from the regional data, supplier data, and regional policies; S204. Extract the second key field from the public opinion data and convert the second key field into the second structured data entry; S205. Convert the logistics timeliness data into the third structured data entry; S206. Generate node data information in a preset format from the first structured data entry, the second structured data entry, the third structured data entry, the regional supplier mapping table, and the sentiment polarity scores; S207. Map and store the node data information with the link node information; And, evaluate whether the node data information meets the preset fluctuation condition, if not, adjust the preset acquisition frequency of the multi-source data.

[0061] In step S201, the extraction of the first key fields refers to identifying and extracting the core attributes related to risk assessment from the supplier data, such as the expiration date of certifications in the supplier qualification documents, the proportion of non-conforming items in the quality spot-check results, and the historical fluctuation range of the on-time delivery rate. The first key fields are converted into the first structured data entries by parsing unstructured documents. For example, the "on-time delivery rate" field is standardized into a numerical variable to facilitate quantitative analysis by subsequent modules.

[0062] In step S202, the category return rate is calculated by counting the proportion of return records of specific product categories in the total sales orders. Preferably, the NLP sentiment analysis of text evaluation identifies the sentiment tendency of the review text through a pre-trained model and outputs a sentiment polarity score (such as -0.8 indicating strong negative sentiment), which is used to reflect the satisfaction of consumers with the product or service.

[0063] In step S203, preferably, the regional supplier mapping table is constructed by associating the supplier registration location, service coverage area, and regional policies. For example, suppliers with EU CE certifications are bound to the compliance requirements of the target clearance nodes to ensure quick matching of alternative resources when regional policies change.

[0064] In step S204, the extraction of the second key fields focuses on the event types, dissemination heat, and associated entities in the public opinion data, and is converted into the second structured data entries. For example, the "trade war" public opinion is marked as the "current affairs public opinion" type and associated with the list of affected suppliers.

[0065] In step S205, the original status in the logistics timeliness data is converted into the third structured data entries. For example, the deviation days between the actual transportation duration and the promised timeliness are calculated according to the logistics node timings to evaluate the timeliness risk of the logistics nodes.

[0066] In step S206, the preset format requirements integrate the structured entries from different data sources into a unified JSON or database table structure. For example, the supplier qualifications, category return rate, and logistics deviation days are aligned by field names to eliminate data heterogeneity.

[0067] In step S207, the mapping storage associates and stores the node data information in the form of key-value pairs by assigning a unique identifier to each link node. For example, the logistics timeliness deviation data is bound to the distribution node ID. Preferably, when evaluating the preset fluctuation conditions, the preset fluctuation conditions are configured as the information entropy or the time series fluctuation amplitude threshold of the node data. For example, when the standard deviation of the sentiment polarity scores collected at a certain node for 3 consecutive times exceeds 0.5, it is determined that the fluctuation is abnormal, and the collection frequency is adjusted from once per hour to once every 15 minutes to improve data real-time performance.

[0068] It should be noted that among the various link nodes divided in the entire supply chain link, supplier data covers multiple types of supply resources corresponding to each link. For example, in the procurement and inventory nodes, supplier data includes multiple optional suppliers for the same material to ensure redundant raw material supply; in the logistics distribution nodes, supplier data involves multiple logistics carriers in different regions to achieve flexible scheduling of transportation routes; in the warehousing and sorting nodes, supplier data can be subdivided into self-operated warehousing facilities and third-party warehousing service providers, and the third-party service providers are further classified and managed according to their operation scale and automation level; in the after-sales nodes, supplier data includes self-operated service teams and external cooperative service outlets, and the external outlets are classified and configured according to service qualifications (such as brand-authorized repairers, regional comprehensive service points). By hierarchically managing supplier data at each node according to resource types and service capabilities, it supports the alternative supplier module to quickly match the optimal alternative solution in risk scenarios.

[0069] In this embodiment, through the standardized processing of multi-source heterogeneous data and the adjustment of dynamic acquisition frequencies, the real-time performance and data consistency of risk early warning are improved: the first structured data entry ensures that core indicators such as supplier qualifications and logistics timeliness can be quantitatively analyzed; the geographical supplier mapping table enhances the cross-regional risk response ability; the mapping of node data information and link nodes guarantees the deep coupling of business scenarios and risk indicators, supporting the modeling of risk conduction paths; the preset fluctuation conditions trigger the dynamic optimization of the acquisition strategy to ensure the timely capture of abnormal fluctuations. The above technical solutions achieve the precise screening of alternative suppliers and the scenario adaptability of pre-plan generation through the hierarchical management of supplier resources at each node and the integration of multi-dimensional data, effectively improving the agility and reliability of the entire link risk prevention and control.

[0070] Please refer to Figure 4 , in some embodiments, the risk information for generating each link node information based on the preprocessed multi-source data includes: S301. Sort the influence coefficients of the preprocessed multi-source data associated with the link node information according to the AHP algorithm to obtain the influence coefficients of each data type in the multi-source data; S302. Convert the risk weights of each data type through a linear normalization function according to the influence coefficients, denoted as the initial risk weights; S303. Calculate the risk probability of each data type, and the risk probability is configured to be calculated using the Poisson distribution model through the number of risk occurrences in the historical data information of the same data type at the same link node; S304. Integrate the risk probability and the initial risk weights to form the basic risk weights; S305. Extract the sudden information in the multi-source data through the CUSUM algorithm, and the sudden information includes policy sudden characteristics, promotion sudden characteristics, and geographical sudden characteristics; S306. Convert the sudden information into a sudden risk weight using a fuzzy logic algorithm; S307. Integrate the basic risk weight and the sudden risk weight to obtain risk information.

[0071] In step S301, preferably, the AHP algorithm quantifies the influence degree of multi-source data on the risk of link nodes by constructing a hierarchical structure model, where the link nodes are used as the target layer, and the associated multi-source data types are used as the criterion layer. A judgment matrix is constructed through expert experience and historical data to calculate the influence coefficients of each data type.

[0072] In step S302, the linear normalization function is used to map the influence coefficients to a standardized interval and ensure that the sum of weights is a fixed value to form an initial risk weight.

[0073] In step S303, preferably, the risk probability can be obtained by statistically calculating the historical risk occurrence frequency through a Poisson distribution model. For example, the average occurrence rate of past risk events of a certain data type under the same link node is used as the estimation basis for the current risk probability.

[0074] In step S304, the fusion of the risk probability and the initial risk weight can be achieved through methods such as weighted geometric mean, Euclidean distance weighting, attention mechanism fusion, or neural network model fusion. Preferably, the weighted geometric mean method is adopted to dynamically integrate the initial risk weight and the risk probability according to a preset ratio. Specifically, the initial risk weight reflects the static influence degree of the data type on the node, while the risk probability represents the dynamic frequency of historical risk occurrence. By setting the weighting coefficients of the basic weight ratio and the risk probability ratio, the geometric mean is calculated as the basic risk weight, which not only retains the benchmark weight given by expert experience but also strengthens the decision-making priority of high-frequency risk data.

[0075] In step S305, the CUSUM algorithm can detect sudden fluctuations by monitoring the cumulative deviation of time-series data, such as a sharp increase in the order volume during a promotion period or an abnormal jump in the call frequency of the policy API interface. The sudden information includes policy sudden characteristics, promotion sudden characteristics, and regional sudden characteristics. Among them, the policy sudden characteristics can monitor the text change frequency of the customs policy API; the promotion sudden characteristics can detect the second derivative mutation points of the sales time series to accurately capture flash sales activities; the regional sudden characteristics can analyze the spatio-temporal clustering of natural disaster warning signals.

[0076] In step S306, preferably, the fuzzy logic algorithm converts unstructured sudden characteristics (such as the policy text update frequency) into membership degree scores and maps them to sudden risk weights based on a preset rule base.

[0077] In step S307, when fusing the basic risk weight and the sudden risk weight, methods such as weighted geometric mean, Euclidean distance weighting, attention mechanism fusion, or neural network model fusion can be selected. Preferably, the attention mechanism is adopted to dynamically allocate the fusion ratio of the basic and sudden weights according to the intensity of the sudden features. Specifically, by constructing an attention scoring model, features such as the policy mutation frequency, promotion scale, or disaster warning level in the sudden information are analyzed, the attention coefficient of the sudden risk weight is calculated, and the two types of weights are weighted and summed according to the attention coefficient. For example, when there is a policy mutation, the proportion of the sudden weight in the final risk information is increased to achieve the adaptive adjustment of the risk weight with the scenario.

[0078] The technical solution of this embodiment can be understood as follows: The risk information reflects the multi-dimensional risk characteristics of multi-source data in the link node, including each risk probability corresponding to the multi-source data, which can be understood as the risk weight of this data type and is evaluated according to the link node information. Among them, the basic risk weight is obtained by hierarchically evaluating the influence degree of the multi-source data types associated with the link node. First, the AHP algorithm is used to sort the influence coefficients of the data types to quantify the static weight; then, the Poisson distribution model is combined to statistically analyze the historical risk occurrence probability, and the comprehensive weight is generated through weighted geometric mean fusion.

[0079] The sudden risk weight is aimed at scenarios such as policy mutation, promotion peak, or regional disaster. The CUSUM algorithm is used to extract the sudden features, and after fuzzy logic conversion, a dynamic weight is formed. The final risk information dynamically adjusts the fusion ratio of the two types of weights through the attention mechanism. For example, when there is a policy mutation, the proportion of the sudden weight is increased to achieve the adaptive balance between the normal risk law and the sudden scenario response. For example, in the procurement node, which is a link node, the basic weight of the logistics timeliness data reflects the risk of the regular transportation efficiency. After superimposing the sudden weight of the tariff policy mutation, a composite risk index including the supplier switching priority and the emergency customs clearance strategy is generated, which is the risk information.

[0080] This embodiment improves the accuracy and scenario adaptability of risk information generation through a hierarchical evaluation and dynamic sudden response mechanism. The combination of the AHP algorithm and the Poisson distribution quantifies the historical influence weight of multi-source data and strengthens the decision-making priority of high-frequency risk factors; the CUSUM algorithm and fuzzy logic accurately capture sudden features such as policy mutation and promotion peak and convert them into dynamic risk weights; the attention mechanism adaptively adjusts the fusion ratio of the basic and sudden weights according to the sudden intensity to achieve the dynamic balance between the normal risk law and the sudden scenario response, and finally forms multi-dimensional risk information to support the precise triggering of the risk pre-plan.

[0081] In some embodiments, according to the link node information, the preprocessed multi-source data associated with it is sorted according to the AHP algorithm to obtain the influence coefficient of each data type in the multi-source data, including: Construct a judgment matrix corresponding to the node type of the current link node, and the matrix elements of the judgment matrix are generated from the relative importance of the data types of multi-source data; Calculate the maximum eigenvalue of the judgment matrix and its associated eigenvector, and each component of the eigenvector corresponds to a data type; Normalize the eigenvector to obtain the influence coefficients of multiple data types; Calculate the risk probability of each data type, and the risk probability is configured to be calculated using the Poisson distribution model based on the number of risk occurrences in the historical data information of the same data type of the same link node, including: Statistically calculate the average number of risk events of the current data type per unit time based on the historical data information; Calculate the probability of at least one risk event occurring within the current time window using the Poisson distribution probability function as the risk probability, which is represented by formula (1), and formula (1) is as follows: ; In formula (1), is the risk probability, is the number of occurrences of risk events, is the base of the natural logarithm, is the average number of risk events per unit time; Extract the burst information in the multi-source data through the CUSUM algorithm, and the burst information includes policy burst characteristics, promotion burst characteristics, and regional burst characteristics, including: Construct a cumulative statistic for the time series data stream in the multi-source data, and mark it as a mutation point when the cumulative statistic exceeds the preset control limit, which is represented by formula (2), and formula (2) is as follows: ; In formula (2), is the cumulative statistic at time , is the cumulative statistic at time , is the observed value at time , is the reference mean value, is the allowable offset parameter; Classify and generate at least one of policy burst characteristics, promotion burst characteristics, or regional burst characteristics according to the data type corresponding to the mutation point; Convert the burst information into burst risk weights using the fuzzy logic algorithm, including: Establish a fuzzy rule base, where the input variables of the fuzzy rule base are the intensity of the burst characteristics and the duration of the burst characteristics, and the output variable of the fuzzy rule base is the burst risk weight; The fuzzy inference result is converted into a numerical sudden risk weight through defuzzification operation.

[0082] In this embodiment, when sorting the influence coefficients according to the AHP algorithm, the matrix elements of the judgment matrix are generated by the relative importance of the data types of multi-source data, where the relative importance is determined according to the business characteristics of the current link node. For example, the comparison scale of the logistics timeliness data and the supplier data in the procurement node is higher than that in the warehousing node.

[0083] The normalization process of the eigenvector is achieved by dividing each component by the sum of the vectors, making the sum of the influence coefficients a fixed value.

[0084] When calculating the risk probability of each data type, the parameters in the Poisson distribution model are determined by statistically averaging the number of occurrences of risk events within the unit time window in the historical data information. For example, the daily average value of the number of delivery delays in the logistics node in the past 30 days.

[0085] When extracting the sudden information in the multi-source data through the CUSUM algorithm, the allowable offset parameter in the CUSUM algorithm is preset according to the business scenario. For example, in the logistics timeliness fluctuation it is set to 1.5 times the daily average deviation. The reference mean is the moving average of the historical data sliding window. For example, the moving average of the observed values in the previous 24 hours is taken as the current value.

[0086] When converting the sudden information into a sudden risk weight using the fuzzy logic algorithm, the fuzzy rule base adopts the Mamdani type fuzzy inference system, and the membership function of the input variable is triangular. For example, the intensity of the policy sudden feature "high" corresponds to the vertex of the trigonometric function being updated 10 times a day on average, and the bottom span is 5 - 15 times. The defuzzification operation calculates the centroid position of the fuzzy output region through the centroid method and converts it into a sudden risk weight.

[0087] This embodiment improves the accuracy and scenario adaptability of the risk coefficient calculation through hierarchical impact assessment and dynamic sudden feature extraction. The AHP algorithm constructs a judgment matrix based on node characteristics to quantify the static influence weight of multi-source data; the Poisson distribution model statistically counts the historical risk occurrence frequency to strengthen the probability weight of high-frequency events; the CUSUM algorithm combines the sliding window mean and the offset parameter to accurately capture sudden features such as policy mutations; the Mamdani type fuzzy rule base realizes the reliable conversion of unstructured sudden information into risk weights through triangular membership functions and centroid method defuzzification.

[0088] Please refer to Figure 5, in some embodiments, generating supply product quality information based on consumption behavior data and public opinion data, and binding the supply product quality information to a supplier includes: S401. Process the return records in the consumption behavior data, and count the category return rates of each supplier; S402. Extract the text evaluations in the consumption behavior data for NLP sentiment analysis to generate the quality satisfaction score of the supplier; S403. Extract key fields from the public opinion data to identify quality events associated with the supplier; S404. Calculate the public opinion risk value based on the occurrence frequency and spread range of the quality events; S405. Weightedly fuse the category return rate, quality satisfaction score, and public opinion risk value to generate a comprehensive quality score; S406. Divide the quality levels according to the comprehensive quality score to obtain the supply product quality information; S407. Store the supply product quality information in association with the corresponding supplier; And, when the quality level is lower than the preset level threshold, trigger the screening mechanism for alternative suppliers; And, update the supply product quality information according to the newly added consumption behavior data and public opinion data at a preset cycle.

[0089] In step S401, the category return rate is calculated by counting the proportion of the return quantity of each supplier in a specific product category to the total sales volume, reflecting the problem rate in the actual use of the product. For example, the category return rate of the diaper category of a maternal and child supplier can be quantified by the ratio of the monthly return quantity to the total shipment quantity.

[0090] In step S402, the quality satisfaction score can be generated by performing sentiment analysis on the consumer review text through a pre-trained NLP model. After the model outputs the sentiment polarity score, it is linearly mapped to a preset score interval to quantify the subjective satisfaction of consumers with the product quality.

[0091] In step S403, key field extraction identifies the types and spread ranges of quality events associated with the supplier from the public opinion data. For example, the keyword "battery failure" is captured through text matching technology and associated with the corresponding supplier.

[0092] In step S404, the public opinion risk value is weighted and calculated based on the occurrence frequency, spread breadth, and severity of the quality event type within a unit time. The spread breadth is quantified by the logarithmic transformation of the repost quantity or comment quantity, and the event severity is adjusted by a preset coefficient according to the type of quality problem or service problem.

[0093] In step S405, the comprehensive quality score linearly fuses the category return rate, quality satisfaction score, and public opinion risk value through preset weights. Among them, the return rate and satisfaction score are used as negative indicators to deduct scores, and the public opinion risk value is used as a positive indicator to increase scores. Finally, a comprehensive quality score ranging from 0 to 100 is generated. The lower the comprehensive quality score, the worse the quality.

[0094] In step S406, the quality level can be divided according to the preset comprehensive quality score threshold. Suppliers with a comprehensive quality score lower than the lowest threshold are classified into the low-quality level, and the screening mechanism for alternative suppliers in the foregoing embodiments is triggered. Preferably, the preset comprehensive quality score threshold is determined comprehensively through industry standards, historical supplier score distributions, and expert experience. For example, critical values are divided by combining historical data quantiles to ensure that the level division matches the actual business risk tolerance.

[0095] In step S407, the supplier quality information is bound and stored with the supplier unique identifier and dynamically updated at a preset cycle. When updating, the return rate, satisfaction score, and public opinion risk value are recalculated to ensure data timeliness.

[0096] Optionally, the screening mechanism for alternative suppliers may include the following steps: When the supplier quality level is lower than the preset level threshold, a candidate set of alternative suppliers that meet the business requirements of the current link node is retrieved from the database. The screening conditions include: 1) The quality level is not lower than that of the current supplier and has passed the qualification review; 2) The geographical coverage matching degree is calculated as the overlap rate between the supplier service scope and the business area of the current node, and it is required that the matching degree ≥ the preset threshold; 3) The historical on-time delivery rate is calculated based on historical order fulfillment data, and suppliers with the top 20% on-time rates are preferentially selected; The preset quantity is dynamically adjusted according to the real-time volatility of the current node. For every 10% increase in the real-time volatility, the preset quantity increases by 1; Finally, a weighted scoring model is constructed based on the quality level, geographical matching degree, and on-time rate, and the top preset quantity of alternative suppliers is selected according to the total score ranking and updated to the link node information.

[0097] This embodiment realizes the accurate evaluation and risk response of supplier quality through a multi-dimensional index fusion and dynamic update mechanism. The category return rate quantifies the actual usage problem rate of the product, NLP sentiment analysis analyzes consumers' subjective evaluations, the public opinion risk value captures the impact of event dissemination, and a weighted fusion generates a comprehensive quality score. The preset threshold divides the quality level and triggers the screening of alternative suppliers, and the dynamic update mechanism ensures data timeliness. For example, when the score is lower than the threshold, an alternative plan matching is started, and combined with periodic re-evaluation, the quality status is ensured to be synchronized in real time, improving the supply chain risk response efficiency.

[0098] In some embodiments, a preset number of alternative suppliers are screened in a database according to risk information and link node information, and the alternative suppliers are updated to the link node information. The preset number is configured to be generated according to the real-time volatility of the current link node, including: Calculate the real-time volatility of the current link node, and the real-time volatility is obtained by converting the degree of change of key indicators in the link node information; Dynamically adjust the preset number according to the magnitude of the real-time volatility; Rank the suppliers to be replaced in the database according to preset screening conditions. The preset screening conditions include geographical coverage matching degree, supply quality grade, historical delivery on-time rate, and cost premium; Select the top-ranked preset number of suppliers to be replaced, denoted as alternative suppliers, and update them to the link node information.

[0099] In this embodiment, the real-time volatility is calculated by statistically analyzing the degree of temporal change of key indicators of the link node (such as order volume and inventory consumption rate), and specifically, the fluctuation amplitude can be quantified based on the ratio of the standard deviation to the mean of the indicators within a moving time window.

[0100] The preset number is dynamically adjusted according to the magnitude of the real-time volatility. The greater the real-time volatility, the larger the preset number; Rank the alternative suppliers in the database according to preset screening conditions. The screening conditions include: the geographical coverage matching degree with the current link node; the supply quality grade is not lower than the preset requirement; the historical delivery on-time rate reaches the preset standard; the cost premium does not exceed the preset ratio. Among them, the geographical coverage matching degree is calculated by the overlap rate between the supplier service area and the business coverage area of the current node. For example, if a logistics node covers the East China region, then suppliers whose service scope includes this region are screened; the cost premium refers to the excess ratio of the alternative supplier's quotation relative to the current supplier, which needs to be lower than the preset upper limit. The priority ranking is achieved through weighted comprehensive scoring. For example, the geographical matching degree weight is 40%, the quality grade weight is 30%, the delivery on-time rate weight is 20%, and the cost premium weight is 10%. Those with a higher total score are given priority.

[0101] Preferably, when the basic risk weight and the sudden risk weight in the risk information exceed the dynamic threshold, the alternative supplier switching mechanism is triggered. The dynamic threshold is adjusted inversely according to the real-time volatility, that is, when the real-time volatility increases, the dynamic threshold decreases accordingly. Specifically, when the weighted sum of the basic risk weight and the sudden risk weight exceeds this threshold, the alternative supplier switching mechanism is triggered.

[0102] In this embodiment, a dynamic regulation mechanism driven by real-time volatility is adopted to improve the accuracy and response efficiency of alternative supplier screening. The number of alternative suppliers is dynamically adjusted based on the fluctuation range of key indicators to ensure resource redundancy in high-volatility scenarios; multi-dimensional screening conditions are combined with weighted priority sorting to ensure the balanced optimization of geographical coverage matching, quality grade, and cost control; the dynamic threshold is adjusted inversely with the real-time volatility, and when the sum of risk weights exceeds the threshold, a quick switch is triggered to enhance the risk resistance ability of the supply chain.

[0103] In some embodiments, a risk conduction path is generated based on risk information and supply chain information, and the risk conduction path includes path nodes of multiple different risk levels, including: Generate an initial topology according to the node connection rules in the supply chain information; Calculate the hierarchical correlation degree between each link node, and the hierarchical correlation degree is configured to be generated by the physical connection strength, data interaction frequency, and business dependence degree between multiple link nodes; Construct multi-directional correlation lines on the initial topology according to the hierarchical correlation degree to generate a composite topology; Initialize the risk conduction coefficient for multiple link nodes in the composite topology, and the risk conduction coefficient is configured to be obtained through the following steps: Calculate the conduction influence weight of the previous link node and the subsequent link node of the current link node; And calculate the collaborative risk weight of other link nodes at the same level as the current link node; And calculate the indirect conduction attenuation weight of multiple link nodes across levels and the current link node; Fuse the conduction influence weight, collaborative risk weight, and indirect conduction attenuation weight to obtain the risk conduction coefficient; Quantify the risks of multiple link nodes according to the risk information and the composite topology to obtain the risk increment of each link node, and generate the final risk weight of each link node, which is represented by formula (3), and formula (3) is as follows: ; In formula (3), is the final risk weight of the current link node, is the initial risk value of the current link node, is the risk conduction coefficient, is the risk increment; Generate a risk conduction path with different risk levels according to the final risk weight and the composite topology.

[0104] In this embodiment, the hierarchical correlation degree refers to the closeness of the business relationship between link nodes, which is obtained by weighted calculation through the physical connection strength (such as the number of transportation lines between logistics nodes and warehousing nodes), data interaction frequency (such as the number of inventory synchronization times between order nodes and warehousing nodes), and business dependence degree (such as the proportion of raw material supply from supplier nodes to production nodes).

[0105] Constructing multi-directional associated lines can be understood as superimposing functional connection paths across levels and across businesses (such as the reverse logistics channel where after-sales nodes are directly associated with supplier nodes) on the basis of the initial topological structure, and finally forming a composite topological structure that supports multi-dimensional risk conduction.

[0106] In the steps of obtaining the risk conduction coefficient, the conduction influence weight represents the risk transfer ability of the previous node to the current node. For example, the probability of stockout in the warehouse due to procurement delay; the collaborative risk weight quantifies the superimposed effect of risks occurring simultaneously among nodes at the same level, such as the chain effect of delays of multiple logistics nodes; the indirect conduction attenuation weight measures the risk attenuation rate of cross-level nodes. For example, the intensity of the indirect impact of sudden changes in customs clearance policies on overseas warehouse nodes. The risk conduction coefficient is obtained by weighted fusion of the conduction influence weight, the collaborative risk weight, and the indirect conduction attenuation weight, and is used to quantify the overall intensity of risk conduction between link nodes.

[0107] Risk increment Based on the risk information (such as the probability of supplier delivery delay) of the current link node and the risk conduction coefficient The product calculation reflects the risk diffusion value to the associated link nodes. The final risk weight Is generated by weighted summation of the initial risk value Of the current link node and the risk increment

[0108] When it is detected that the risk information or supply chain information has changed, the risk conduction path is updated.

[0109] In this embodiment, through the composite topology modeling and dynamic conduction calculation mechanism, the accurate identification and real-time update of the supply chain risk conduction path are realized. The hierarchical correlation degree is calculated based on the physical connection strength, data interaction frequency, and business dependence degree, and multi-directional associated lines are constructed to form a composite topological structure; the risk conduction coefficient is generated by fusing the conduction influence weight, the collaborative risk weight, and the indirect conduction attenuation weight to quantify the risk diffusion intensity between nodes; the final risk weight is calculated by combining the initial risk value and the risk increment to dynamically generate multi-level conduction paths. Through the dynamic update of the topological structure and the conduction coefficient, the timeliness of cascading risk warning and the accuracy of risk pre-schemes are improved.

[0110] By adopting the above technical solutions, the present invention is different from the prior art and has the following beneficial effects: ​The present invention divides supply chain information according to link nodes and conducts multi-source data collection, achieving all-dimensional monitoring of supply chain risks; combines supplier data, consumer behavior data, and public opinion data to generate dynamically bound supply product quality information, enhancing the objectivity and real-time nature of supplier evaluation; dynamically adjusts the number of alternative suppliers based on real-time volatility, enhancing the flexible response ability of supply chain nodes; accurately captures the conduction path and cascade effect of risks among supply chain nodes by constructing a composite topology structure and a risk conduction coefficient calculation mechanism; uses a risk quantification model to fuse basic risk weights and sudden risk weights, ensuring the comprehensiveness of risk assessment and the response sensitivity to emergencies; finally, generates a risk pre-plan adapted to the link scenario through multi-dimensional data fusion, realizing a full-process closed-loop management from risk identification, conduction analysis to pre-plan formulation, effectively improving the timeliness of supply chain risk early warning and the accuracy of disposal strategies.

[0111] Finally, it should be noted that although the above embodiments have been described in the text of the specification and the drawings of the present application, the patent protection scope of the present application cannot be limited thereby. Any technical solutions generated by equivalent structure or equivalent process substitution or modification using the content recorded in the text of the specification and the drawings of the present application based on the essential concept of the present application, as well as those directly or indirectly implementing the technical solutions of the above embodiments in other related technical fields, are all included in the patent protection scope of the present application.

Claims

1. A supply chain full-link risk early warning platform based on big data, characterized in that, Including: A supply chain division module, configured to obtain supply chain information and divide the supply chain information according to link nodes to obtain a plurality of link node information; A data collection module, configured to obtain multi-source data of each piece of the link node information, where the multi-source data includes supplier data, consumer behavior data, logistics timeliness data, regional data, and public opinion data; A risk assessment module, configured to generate risk information for each piece of the link node information according to the multi-source data, and generate supply quality information according to the consumer behavior data and public opinion data, and bind the supply quality information to the supplier; An alternative supplier module, configured to screen a preset number of alternative suppliers in a database according to the risk information and the link node information, and update the alternative suppliers to the link node information, where the preset number is configured to be generated according to the real-time volatility of the current link node; A risk path conduction module, configured to generate a risk conduction path according to the risk information and the supply chain information, where the risk conduction path includes a plurality of path nodes with different risk levels; A risk preplan module, configured to generate a risk preplan according to the risk information, the risk conduction path, and the updated link node information.

2. The supply chain full-link risk early warning platform based on big data according to claim 1, wherein The data collection module includes: A data standardization unit, configured to perform standardization processing on the multi-source data to obtain standardized data, including: Performing first keyword field extraction on the supplier data and converting the first keyword field into a first structured data entry; Performing data processing on the return records in the consumer behavior data to obtain a category return rate, and performing NLP sentiment analysis on the text evaluations in the consumer behavior data to generate sentiment polarity scores; Constructing a regional supplier mapping table with the regional data, the supplier data, and regional policies; Performing second keyword field extraction on the public opinion data and converting the second keyword field into a second structured data entry; Converting the logistics timeliness data into a third structured data entry; A spatio-temporal association unit, configured to associate the standardized data with link nodes; A real-time update unit, configured to implement dynamic update of multi-source data through a stream processing framework.

3. The supply chain full-link risk early warning platform based on big data according to claim 1, characterized in that The risk path conduction module includes: A node association unit, configured to construct a logical dependency relationship of a plurality of link nodes to obtain a composite topology structure, including: Generating an initial topology structure according to the node connection rules in the supply chain information; Calculating the hierarchical association between each of the link nodes, constructing multi-directional association lines on the initial topology structure, and generating a composite topology structure; Initializing a risk conduction coefficient for a plurality of link nodes in the composite topology structure; A risk quantification unit, configured to perform risk quantification on a plurality of link nodes according to the risk information and the composite topology structure to obtain a risk increment for each link node, and generate a final risk weight for each link node; A path adjustment unit, configured to generate a risk conduction path with different risk levels according to the final risk weight and the composite topology structure.

4. A method for early warning of risks in the entire supply chain link based on big data, characterized in that, Applicable to the risk warning platform according to any one of claims 1-3, the method includes: Obtain supply chain information, divide the supply chain information according to link nodes to obtain multiple link node information; Regularly obtain multi-source data of each link node information according to a preset collection frequency and perform preprocessing, and map and store the preprocessed multi-source data with the link nodes; Generate risk information for each link node information based on the preprocessed multi-source data, and generate supply quality information based on the consumption behavior data and public opinion data, and bind the supply quality information to the supplier; Screen a preset number of alternative suppliers in the database according to the risk information and the link node information, and update the alternative suppliers to the link node information, and the preset number is configured to be generated according to the real-time volatility of the current link node; Generate a risk conduction path according to the risk information and the supply chain information, and the risk conduction path includes multiple path nodes with different risk levels; Generate a risk plan according to the risk information, the risk conduction path and the updated link node information.

5. The method for early warning of risks in the entire supply chain link based on big data according to claim 4, characterized in that, Obtain multi-source data of each link node information and perform preprocessing. Mapping and storing the preprocessed multi-source data with the link node includes: Extract the first key field from the supplier data and convert the first key field into the first structured data entry; Perform data processing on the return records in the consumption behavior data to obtain the category return rate, and perform NLP sentiment analysis on the text evaluation in the consumption behavior data to generate sentiment polarity scores; Construct a regional supplier mapping table with the regional data, supplier data and regional policies; Extract the second key field from the public opinion data and convert the second key field into the second structured data entry; Convert the logistics timeliness data into the third structured data entry; Generate node data information in a preset format with the first structured data entry, the second structured data entry, the third structured data entry, the regional supplier mapping table and the sentiment polarity score; Map and store the node data information with the link node information; And, evaluate whether the node data information meets the preset fluctuation condition. If not, adjust the preset collection frequency of the multi-source data.

6. The method for early warning of risks in the entire supply chain based on big data according to claim 4, wherein, Generating risk information for each link node information based on the preprocessed multi-source data includes: Sort the influence coefficients of the preprocessed multi-source data associated with it according to the AHP algorithm according to the link node information to obtain the influence coefficient of each data type in the multi-source data; Convert the risk weight of each data type through a linear normalization function according to the influence coefficient, denoted as the initial risk weight; Calculate the risk probability of each data type, and the risk probability is configured to be calculated using the Poisson distribution model by the number of risk occurrences in the historical data information of the same data type of the same link node; Fuse the risk probability and the initial risk weight to form a basic risk weight; Extract the sudden information in the multi-source data through the CUSUM algorithm, and the sudden information includes policy sudden characteristics, promotion sudden characteristics and regional sudden characteristics; Convert the burst information into a burst risk weight using a fuzzy logic algorithm; Fuse the basic risk weight and the burst risk weight to obtain the risk information.

7. The method for early warning of risks in the entire supply chain based on big data according to claim 6, characterized in that Sort the influence coefficients of the preprocessed multi-source data associated with the link node according to the AHP algorithm based on the link node information, and the influence coefficients of each data type in the multi-source data include: Construct a judgment matrix corresponding to the node type of the current link node, and the matrix elements of the judgment matrix are generated by the relative importance of the data types of the multi-source data; Calculate the maximum eigenvalue of the judgment matrix and its associated eigenvector, and each component of the eigenvector corresponds to a data type; Normalize the eigenvector to obtain the influence coefficients of multiple data types; Calculate the risk probability of each data type, and the risk probability is configured to be calculated using the Poisson distribution model through the number of risk occurrences in the historical data information of the same data type of the same link node, including: Statistically calculate the average number of risk events of the current data type per unit time according to the historical data information; Calculate the probability of at least one risk event occurring in the current time window as the risk probability through the Poisson distribution probability function, which is represented by formula (1), and formula (1) is as follows: ； In formula (1), is the risk probability, is the number of occurrences of the risk event, is the base of the natural logarithm, is the average number of occurrences of the risk event per unit time; Extract the burst information from the multi-source data through the CUSUM algorithm, and the burst information includes policy burst characteristics, promotion burst characteristics, and regional burst characteristics, including: Construct a cumulative statistic for the time series data stream in the multi-source data, and mark it as a mutation point when the cumulative statistic exceeds a preset control limit, which is represented by formula (2), and formula (2) is as follows: ； In formula (2), is the cumulative statistic at moment, is the cumulative statistic at moment, is the observed value at moment, is the reference mean value, is the allowable offset parameter; Classify and generate at least one of the policy burst characteristics, promotion burst characteristics, or regional burst characteristics according to the data type corresponding to the mutation point; Convert the burst information into a burst risk weight using a fuzzy logic algorithm, including: Establish a fuzzy rule base, where the input variables of the fuzzy rule base are the intensity of the burst characteristics and the duration of the burst characteristics, and the output variable of the fuzzy rule base is the burst risk weight; Convert the fuzzy inference result into a numerical burst risk weight through defuzzification operation.

8. The method for early warning of risks in the entire supply chain based on big data according to claim 4, characterized in that, Generate supply product quality information according to the consumption behavior data and public opinion data, and associate the supply product quality information with the supplier, including: Perform data processing on the return records in the consumption behavior data, and statistically calculate the category return rates of each supplier; Extract the text evaluations in the consumption behavior data for NLP sentiment analysis to generate the quality satisfaction scores of the suppliers; Extract keyword fields from the public opinion data to identify quality events associated with the suppliers; Calculate the public opinion risk value according to the occurrence frequency and spread range of the quality events; Fuse the category return rate, quality satisfaction score, and public opinion risk value with weights to generate a comprehensive quality score; Divide the quality level according to the comprehensive quality score to obtain the supply product quality information; Associate and store the supply product quality information with the corresponding supplier; And, when the quality level is lower than the preset level threshold, trigger the screening mechanism for alternative suppliers; And, update the supply quality information according to the newly added consumer behavior data and public opinion data at a preset cycle.

9. The supply chain full-link risk early warning method based on big data according to claim 4, characterized in that, Screen a preset number of alternative suppliers in the database according to the risk information and the link node information, and update the alternative suppliers to the link node information. The preset number is configured to be generated according to the real-time volatility of the current link node, including: Calculate the real-time volatility of the current link node, and the real-time volatility is obtained by converting the change degree of key indicators in the link node information; Dynamically adjust the preset number according to the magnitude of the real-time volatility; Rank the suppliers to be replaced in the database according to preset screening conditions, and the preset screening conditions include geographical coverage matching degree, supply quality level, historical delivery on-time rate, and cost premium; Select a preset number of suppliers to be replaced with the top rankings, denoted as alternative suppliers, and update them to the link node information.

10. The method for early warning of risks in the entire supply chain based on big data according to claim 4, characterized in that Generate a risk conduction path according to the risk information and the supply chain information. The risk conduction path includes multiple path nodes with different risk levels, including: Generate an initial topological structure according to the node connection rules in the supply chain information; Calculate the hierarchical correlation degree between each link node. The hierarchical correlation degree is configured to be generated by the physical connection strength, data interaction frequency, and business dependence degree between multiple link nodes; Construct multi-directional correlation lines on the initial topological structure according to the hierarchical correlation degree to generate a composite topological structure; Initialize the risk conduction coefficient for multiple link nodes in the composite topological structure. The risk conduction coefficient is configured to be obtained through the following steps: Calculate the conduction influence weight of the previous link node and the subsequent link node of the current link node; And calculate the collaborative risk weight of other link nodes at the same level as the current link node; And calculate the indirect conduction attenuation weight of multiple link nodes across levels and the current link node; Fuse the conduction influence weight, collaborative risk weight, and indirect conduction attenuation weight to obtain the risk conduction coefficient; Perform risk quantification on multiple link nodes according to the risk information and the composite topological structure to obtain the risk increment of each link node, and generate the final risk weight of each link node, which is represented by formula (3). The formula (3) is as follows: ； In formula (3), is the final risk weight of the current link node, is the initial risk value of the current link node, is the risk conduction coefficient, is the risk increment; Generate a risk conduction path with different risk levels according to the final risk weight and the composite topological structure.

Images