Model-enabled supply chain risk early warning method and system

Abstract

The invention relates to a model-enabled supply chain risk early warning method and system, and the method comprises the following steps: detecting whether an abnormal order block exists in collected supply chain data or not, and carrying out the restoration of a supply chain path according to the abnormal order block if the abnormal order block exists, and obtaining a risk link diagram; inputting the risk link diagram into a preset time sequence diagram neural network model for risk conduction analysis to obtain risk probability distribution and a prediction influence range; and generating a risk early warning report based on the risk probability distribution and the prediction influence range, and pushing the risk early warning report to a related decision maker to support risk prevention and control and emergency response. And the risk early warning often lags behind the actual risk occurrence process.

Description

Technical Field

[0001] This invention relates to the field of supply chain technology, and in particular to a model-enabled supply chain risk early warning method and system. Background Technology

[0002] With the deepening of globalization and the rapid changes in market demand, modern supply chains have evolved from simple linear structures into complex networks with multiple nodes and links. In such networks, problems in any link can trigger a chain reaction, such as fluctuations in raw material supplier capacity, sudden delays in logistics, or abnormal orders from downstream distributors. These problems often spread throughout the entire supply chain in a short period, causing unpredictable losses to companies. However, many companies still rely on manual processing and analysis of scattered data when dealing with supply chain risks. Faced with massive amounts of order data, logistics information, and inventory records, they not only struggle to quickly identify potential risks but also cannot timely determine the potential path of risk spread, resulting in risk warnings often lagging behind the actual occurrence of risks. Summary of the Invention

[0003] The main technical problem addressed in this application is to provide a model-enabled supply chain risk early warning method and system, which solves the technical problem that traditional methods cannot timely determine the path of abnormal order spread, resulting in risk warnings often lagging behind the actual occurrence of risks.

[0004] To address the aforementioned technical issues, this application employs a model-enabled supply chain risk early warning method, comprising the following steps: The system checks the collected supply chain data for any abnormal order blocks. If any are found, the supply chain path is reconstructed based on these abnormal order blocks to obtain a risk link diagram. The risk link diagram is input into a preset time-series neural network model for risk transmission analysis to obtain the risk probability distribution and predicted impact range. A risk warning report is generated based on the risk probability distribution and the predicted impact range, and the risk warning report is pushed to relevant decision-makers to support risk prevention and emergency response.

[0005] Furthermore, the detection of whether there are abnormal order blocks in the collected supply chain data includes: Feature extraction is performed on the collected supply chain data to obtain order volume fluctuation values ​​and transaction frequency maps, and a visual feature vector is constructed based on the order volume fluctuation values ​​and transaction frequency maps; Based on the visualized feature vector, it is determined whether there are abnormal order blocks in the supply chain data.

[0006] Furthermore, the process of reconstructing the supply chain path based on abnormal order blocks to obtain a risk link diagram includes: The abnormal order block is subjected to order association analysis. By using key fields such as supplier information, customer information, and product information in the order, upstream and downstream orders that have a direct transaction relationship with the orders in the abnormal order block are matched to obtain a set of associated orders. The orders in the set of associated orders are then sorted by time to obtain an order time series. Based on the order time series, the path of the associated order set is sorted out. Starting from the orders in the abnormal order block, the upstream and downstream related orders of the abnormal orders are connected in chronological order to construct an initial supply chain path diagram. The nodes and edges in the initial supply chain path diagram are visualized and labeled to obtain a risk link diagram.

[0007] Furthermore, the step of inputting the risk link diagram into a preset time-series neural network model for risk transmission analysis to obtain the risk probability distribution and predicted impact range includes: The risk link graph is input into a preset temporal graph neural network model; the temporal graph neural network model includes a graph convolutional embedding module, a temporal recursive aggregation module, and a risk diffusion prediction module; The risk link graph is analyzed based on the graph convolutional embedding module to obtain a node topology matrix including node adjacency weights and edge connection strengths. Based on the temporal recursive aggregation module, the node topology matrix is ​​subjected to temporal dependency extraction to obtain a temporal propagation vector that includes the risk transmission delay and temporal correlation degree between nodes. Based on the risk diffusion prediction module, multi-step risk transmission calculations are performed on the time-series propagation vector to obtain the risk probability distribution and predicted impact range.

[0008] Furthermore, the generation of a risk warning report based on the risk probability distribution and the predicted impact range includes: The risk probability distribution is visualized using color mapping rules to obtain risk heat distribution information, and the predicted impact range is bounded to obtain the risk diffusion boundary. Based on the risk heat distribution information and risk diffusion boundary, the report template is filled to obtain an initial early warning report framework. Then, based on the preset decision rule base, the content of the initial early warning report framework is supplemented to obtain a risk early warning report.

[0009] Furthermore, the initial early warning report framework is supplemented with content based on a preset decision rule base to obtain a risk early warning report, including: The risk heat map distribution information in the initial early warning report framework is processed to classify risk levels. Different color areas are classified and judged by preset risk level thresholds to obtain high, medium and low risk area identifiers. Key nodes are extracted from each of the high, medium and low risk area identifiers. Based on the key node extraction results, rule matching processing is performed on the preset decision rule base. Through dual matching of node type and risk level, the corresponding prevention and control suggestion template is selected from the decision rule base. Based on the prevention and control suggestion template, the content of the initial early warning report framework is filled to obtain a risk early warning report containing specific prevention and control measures.

[0010] Furthermore, the step of pushing the risk warning report to relevant decision-makers to support risk prevention and emergency response includes: The risk warning report is processed for decision-making role matching. By extracting the risk level, business modules involved and key node information from the report, it is compared with a preset decision-maker role database to obtain a list of target decision-makers. The decision-makers in the target decision-maker list are prioritized, and priority scores are calculated based on the decision-maker's scope of responsibility and risk correlation to obtain the decision-maker push results. Based on the decision-maker push results, the risk warning report will be pushed to relevant decision-makers to support risk prevention and emergency response.

[0011] This invention also provides a model-enabled supply chain risk early warning system, comprising: The detection unit is used to detect whether there are abnormal order blocks in the collected supply chain data. If they are found, the supply chain path is reconstructed based on the abnormal order blocks to obtain a risk link diagram. The analysis unit is used to input the risk link diagram into a preset time-series neural network model to perform risk transmission analysis, and obtain the risk probability distribution and predicted impact range. The push unit is used to generate a risk warning report based on the risk probability distribution and the predicted impact range, and push the risk warning report to relevant decision-makers to support risk prevention and emergency response.

[0012] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0013] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods described above.

[0014] The above solution detects whether there are abnormal order blocks in the collected supply chain data. If so, it reconstructs the supply chain path based on the abnormal order blocks to obtain a risk link diagram. The risk link diagram is then input into a preset time-series neural network model for risk transmission analysis to obtain the risk probability distribution and predicted impact range. Based on the risk probability distribution and predicted impact range, a risk warning report is generated and pushed to relevant decision-makers to support risk prevention and emergency response. This solution solves the technical problem that traditional methods cannot promptly determine the path of abnormal order spread, causing risk warnings to often lag behind the actual risk occurrence. It helps enterprises overcome the previous predicament of not being able to accurately identify the source of risk or understand the impact link when facing massive amounts of data. It makes risk correlation information scattered in procurement, production, logistics and other links clear and traceable, greatly improving the accuracy and efficiency of risk identification and buying valuable time for subsequent risk prevention and control. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of a model-enabled supply chain risk early warning method in one embodiment of the present invention; Figure 2 This is a structural block diagram of a model-enabled supply chain risk early warning system in one embodiment of the present invention; Figure 3 This is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0017] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0019] Specifically, the model-enabled supply chain risk early warning method in this embodiment includes the following steps: like Figure 1 As shown, Figure 1This invention provides a model-enabled supply chain risk early warning method, comprising the following steps: Step S1: Detect whether there are abnormal order blocks in the collected supply chain data. If so, restore the supply chain path based on the abnormal order blocks to obtain the risk link diagram.

[0020] Specifically, the first step is to preprocess the collected supply chain data, which includes order information, node interaction records, and logistics flow data. First, invalid values ​​and redundant items are removed through data cleaning. Then, algorithms based on threshold judgment and feature comparison are used to detect abnormal order blocks. For example, if the order volume of a batch of orders far exceeds the historical average for the same period of the same period by 30% and the delivery cycle is abnormally shortened, it is identified as an abnormal order block. If an abnormal order block is detected, information related to the upstream suppliers, midstream warehousing nodes, and downstream distributors associated with that order block is retrieved. Order flow records and data interaction logs between each node are retrieved, and the order transmission path is traced according to time sequence and business logic. For example, the ordering merchant of the abnormal order is traced back to the raw material supplier, and then to the logistics company responsible for transportation. Finally, the complete supply chain path is reconstructed, and a risk link diagram containing each risk node, the relationship between nodes, and the direction of risk propagation is generated.

[0021] Step S2: Input the risk link diagram into a preset time-series neural network model for risk transmission analysis to obtain the risk probability distribution and predicted impact range.

[0022] Specifically, the first step is to confirm that the preset temporal graph neural network model has been trained and the model parameters have been debugged and optimized through multiple sets of historical supply chain risk data to adapt to the link structure under different scenarios. Then, the node attributes (such as node type and historical risk records) and edge weights (such as the strength of business association between nodes) in the risk link graph are converted into a vector format that the model can recognize. At the same time, the interaction data of each node in the past three months are supplemented according to the time series, such as the order response delay frequency of a certain production node in the past 90 days. The model will first capture the spatial association features of the nodes in the risk link graph through graph convolutional layers, and then use temporal convolutional layers to analyze the dynamic changes of risk under different time slices. For example, when the upstream supplier node has a raw material shortage risk, the model will calculate the probability of the risk being transmitted to the midstream processing node and the downstream distribution node in the next week and two weeks. Finally, the model outputs the risk probability distribution of each node, as well as the predicted impact range such as the number of nodes that the risk may cover and the business links involved.

[0023] Step S3: Generate a risk warning report based on the risk probability distribution and predicted impact range, and push the risk warning report to relevant decision-makers to support risk prevention and emergency response.

[0024] Specifically, the risk probability value of each node is first extracted from the risk probability distribution and divided into three levels: high (above 80%), medium (50%-80%), and low (below 50%). These levels are then marked on the corresponding link nodes within the predicted impact range. For example, upstream raw material supplier A (risk probability 92%), midstream processing plant B (risk probability 68%), and the logistics route connecting them (risk probability 75%) are highlighted. At the same time, data on the response effects of the last three similar risks are supplemented as a reference. This is used to integrate a clearly structured risk warning report. Then, through the company's internal decision support system, the report is pushed to relevant decision-makers based on their responsibilities and authority. For example, supplier risk details are pushed to the purchasing manager, and transportation route risk response suggestions are pushed to the logistics director, ensuring that the report accurately reaches the relevant decision-makers.

[0025] In a specific embodiment, the detection of whether there are abnormal order blocks in the collected supply chain data includes: Feature extraction is performed on the collected supply chain data to obtain order volume fluctuation values ​​and transaction frequency maps, and a visual feature vector is constructed based on the order volume fluctuation values ​​and transaction frequency maps; Based on the visualized feature vector, it is determined whether there are abnormal order blocks in the supply chain data.

[0026] Specifically, the process begins by filtering out fields directly related to orders from the collected supply chain data, including core information such as order placement time, quantity of goods, trading partners, and delivery cycle for each order. Then, a sliding window method is used to calculate order volume fluctuations—for example, using a 7-day window, the deviation rate of the average daily order volume in the current window is compared with the average daily order volume of the past 6 consecutive windows. If the deviation rate of a certain window exceeds a pre-set threshold of 25%, that window is marked as a potential fluctuation range. At the same time, the number of order transactions per hour is counted at a 24-hour time granularity, and the trend of transaction frequency changes over time is presented in the form of a line graph, forming a transaction frequency graph. For example, on the 3rd day after a major e-commerce promotion, if the transaction frequency suddenly drops to 30% of the daily average during a certain period, such abnormal fluctuations will be clearly shown in the graph. Next, the specific values ​​of order volume fluctuations (such as a deviation rate of 32% for a certain window or a fluctuation range of 18% for a certain period) and the key features of the trading frequency chart (such as the time of peak occurrence, the duration of troughs, and abnormal gap intervals) are transformed into multi-dimensional data arrays. For example, the information is integrated using a structure like [32%, 18%, "14:00-16:00 trough", "2-hour gap"] to construct a visual feature vector containing numerical and descriptive features. The numerical features reflect the degree of quantification of fluctuations, while the descriptive features correspond to abnormal patterns in the frequency chart. When determining the existence of abnormal order blocks, the constructed visual feature vector is compared with the feature vector library of historical normal order data. The similarity between the two is calculated. If the similarity is less than 60%, and the vector contains two or more abnormal features (such as a 45% order volume deviation rate and a 3-hour transaction gap), then an abnormal order block is determined to exist in the supply chain data. For example, if an appliance company's visual feature vector shows features such as "52% order volume deviation rate" and "high-frequency transactions from 2:00 AM to 5:00 AM" during a non-promotional period, and the similarity with historical normal vectors is only 48%, then an abnormal order block can be identified. Throughout the process, feature extraction avoids data noise interference, such as removing duplicate order records caused by system failures, ensuring that order volume fluctuation values ​​and transaction frequency graphs can truly reflect the actual state of order data. The construction of visual feature vectors takes into account both the quantitative attributes of data and the intuitive attributes of graphics, so that anomaly detection is supported by accurate data and can capture abnormal patterns that are not easily reflected by numerical values ​​alone through graphical features.

[0027] In a specific embodiment, the step of reconstructing the supply chain path based on abnormal order blocks to obtain a risk link diagram includes: The abnormal order block is subjected to order association analysis. By using key fields such as supplier information, customer information, and product information in the order, upstream and downstream orders that have a direct transaction relationship with the orders in the abnormal order block are matched to obtain a set of associated orders. The orders in the set of associated orders are then sorted by time to obtain an order time series. Based on the order time series, the path of the associated order set is sorted out. Starting from the orders in the abnormal order block, the upstream and downstream related orders of the abnormal orders are connected in chronological order to construct an initial supply chain path diagram. The nodes and edges in the initial supply chain path diagram are visualized and labeled to obtain a risk link diagram.

[0028] Specifically, for the identified abnormal order blocks, key fields such as supplier information (e.g., supplier number, name, product category), customer information (e.g., customer company code, region, procurement demand type), and product information (e.g., product model, specifications, material code) are extracted for each order. Then, using the database association function of the enterprise supply chain management system, these fields are used as matching criteria to filter out upstream and downstream orders that have a direct transaction relationship with the abnormal order. For example, if the supplier number of an abnormal order is "SUPP-2025-089" and the product model is "MP-032", this can be used as a clue to find the previous stocking order from which the supplier supplied the same model of product to the customer of this order, as well as the subsequent distribution order from which the customer resold this product to downstream distributors. These filtered orders together constitute a set of related orders. Then, the actual order placement time or confirmation time of each order is extracted from the set of related orders and sorted according to the time format of "year-month-day-hour-minute". For example, the upstream stocking order at 9:30 on October 5, 2025, the abnormal order at 14:15 on October 8, and the downstream distribution order at 11:20 on October 12 are arranged in sequence to form an order time sequence. If multiple related orders at the same time point are encountered, they are supplemented and sorted in descending order of order amount to ensure that the sequence logic is clear.

[0029] Next, starting with the orders in the abnormal order block, and referring to the order time series, we first take the nodes corresponding to the abnormal orders (including information such as the business department to which the order belongs and the warehouse location involved) as core nodes. Then, according to the chronological order, we connect upstream related order nodes that precede the abnormal orders in the time series—for example, we map the stock preparation order on October 5th to the node “Production Workshop A of Supplier SUPP-2025-089”. We then connect downstream related order nodes that follow the abnormal orders in the time series—for example, we map the distribution order on October 12th to the node “Regional Warehouse B of Distributor DIST-2025-156”. At the same time, we use arrowed lines to indicate the direction of order flow, with the arrows pointing from upstream nodes to downstream nodes. In this way, we construct an initial supply chain path diagram that includes core nodes, upstream and downstream nodes, and flow directions. In the visualization annotation stage, nodes are first annotated: Core nodes corresponding to abnormal orders are marked with solid red circles, and the annotation includes the order number and the type of abnormality (e.g., "abnormal surge in order volume"); first-level upstream and downstream nodes directly connected to the core node are marked with hollow orange circles, and the annotation includes the node name and the number of associated orders; second-level and higher-level related nodes are marked with hollow blue squares, and the annotation only retains the node name and key product information. Next, edges are annotated: On the arrows connecting core nodes and first-level nodes, the actual number of days the order has been processed (e.g., "3 days") and whether there is a delay (e.g., "1 day delay") are annotated; on the arrows connecting first-level nodes and second-level nodes, only the number of days the order has been processed is annotated. Through such node and edge annotations, a risk link diagram is ultimately formed that intuitively presents the risk propagation path, node relationships, and key anomaly information. For example, in the risk link diagram of an electronics company, the core node is labeled "Order ORD-2025-678 (order volume surged abnormally by 200%)", and its upstream first-level node is labeled "Supplier SUPP-2025-089 (2 related orders)". The arrow connecting the two is marked "5 days (delayed by 2 days)", clearly reflecting the connection and flow issues between the abnormal order and the upstream supplier. Throughout the process, the path analysis avoids indirect orders that are not directly related to transactions, retaining only directly related orders with clear supply and demand relationships to ensure the accuracy of the initial supply chain path diagram. The visual annotations, through the differentiation of colors and shapes, make the node information of different levels and types clear at a glance, providing a clear graphical basis for subsequent risk analysis.

[0030] In a specific embodiment, the step of inputting the risk link diagram into a preset time-series neural network model for risk transmission analysis to obtain the risk probability distribution and predicted impact range includes: The risk link graph is input into a preset temporal graph neural network model; the temporal graph neural network model includes a graph convolutional embedding module, a temporal recursive aggregation module, and a risk diffusion prediction module; The risk link graph is analyzed based on the graph convolutional embedding module to obtain a node topology matrix including node adjacency weights and edge connection strengths. Based on the temporal recursive aggregation module, the node topology matrix is ​​subjected to temporal dependency extraction to obtain a temporal propagation vector that includes the risk transmission delay and temporal correlation degree between nodes. Based on the risk diffusion prediction module, multi-step risk transmission calculations are performed on the time-series propagation vector to obtain the risk probability distribution and predicted impact range.

[0031] Specifically, the pre-constructed risk link graph is first fully input into a pre-defined time-series graph neural network model. This model has been trained and its parameters optimized using over 500 sets of historical supply chain risk event data (covering typical scenarios such as raw material shortages and logistics disruptions) to ensure adaptability to the link structures of different industries. The graph convolutional embedding module within the model first analyzes the topology of the risk link graph, calculating the frequency of business interactions between each node and its neighboring nodes (e.g., monthly order collaborations) and data transmission density (e.g., daily average information exchange volume) to generate node adjacency weights. For example, if an upstream supplier node and a midstream production node collaborate on 20 orders per month and exchange an average of 15 data points daily, their adjacency weights can be increased. The adjacency weight is set to 0.85 for nodes with only 5 collaborations and 3 data exchanges, while the adjacency weight is set to 0.32 for nodes with only 5 collaborations and 3 data exchanges. At the same time, the edge connection strength is determined based on indicators such as order fulfillment rate and logistics on-time rate between nodes. For example, the edge connection strength is marked as 0.91 for a fulfillment rate of 98% and an on-time rate of 95%, and is adjusted to 0.45 for the opposite. Finally, these data are integrated to form a node topology matrix that includes node adjacency weights and edge connection strengths. This process can transform the graphical link structure into a numerical matrix that the model can calculate. For example, in the node topology matrix of a certain automotive parts supply chain, the adjacency weight of "engine supplier A - assembly plant B" is 0.88 and the edge connection strength is 0.92, which clearly reflects the close business relationship between the two.

[0032] Next, the temporal recursive aggregation module extracts temporal dependencies based on the node topology matrix. This module uses an LSTM (Long Short-Term Memory) structure to capture the dynamic relationships between nodes in different time slices. For example, taking 72 hours as a time slice, it analyzes the risk transmission between "engine supplier A and assembly plant B" in the past 5 slices: there is no risk transmission in the first slice; in the second slice, there is a 24-hour transmission delay due to supplier capacity issues, with a temporal correlation of 0.7; in the third to fifth slices, the delay is shortened to 12 hours, and the correlation increases to 0.85. The module integrates these data into a temporal propagation vector that includes the risk transmission delay and temporal correlation between nodes. The vector dimension is consistent with the number of nodes in the node topology matrix. For example, in the temporal propagation vector of the above automotive supply chain, the dimension value corresponding to "supplier A - assembly plant B" is [24h, 0.7; 12h, 0.85], which intuitively presents the temporal change pattern of risk transmission.

[0033] Finally, the risk diffusion prediction module performs multi-step risk transmission calculations on the time-series propagation vector. Using a Monte Carlo simulation algorithm combined with the risk resistance coefficients of supply chain nodes (such as the number of backup suppliers and safety stock levels), it first calculates the probability of risk occurrence at a single node—for example, the probability of production interruption at assembly plant B due to the risk from supplier A is 78%. Then, it extrapolates the probability of risk diffusion upstream and downstream through the time-series propagation vector between nodes: the probability of upstream transmission to other customers of supplier A (such as car manufacturer C) is 62%, and the probability of downstream transmission to dealer D is 85%. Simultaneously, it determines the predicted impact range by combining the number of associated nodes and business coverage. For example, the risk may cover 3 upstream suppliers, 2 midstream manufacturers, and 5 downstream dealers. The final output includes a risk probability distribution table containing the risk probability of each node, as well as a report indicating the number of affected nodes and the predicted impact range of business segments. In the aforementioned automotive supply chain risk probability distribution table, supplier A has a risk probability of 91%, assembly plant B 78%, and dealer D 85%. The predicted impact range covers the entire chain from "engine supply to vehicle assembly to terminal sales," providing accurate data support for subsequent risk prevention and control. Throughout the process, the model automatically eliminates abnormal data interference (such as temporarily adjusted order data) to ensure that the output of each module can truly reflect the actual situation of supply chain risk transmission. Furthermore, the data transmission between modules adopts a standardized format to avoid analytical bias caused by data incompatibility.

[0034] In a specific embodiment, generating a risk warning report based on the risk probability distribution and the predicted impact range includes: The risk probability distribution is visualized using color mapping rules to obtain risk heat distribution information, and the predicted impact range is bounded to obtain the risk diffusion boundary. Based on the risk heat distribution information and risk diffusion boundary, the report template is filled to obtain an initial early warning report framework. Then, based on the preset decision rule base, the content of the initial early warning report framework is supplemented to obtain a risk early warning report.

[0035] Specifically, the risk probability distribution is first visualized according to a preset color mapping rule. This rule divides the risk probability into three intervals: low risk (probability ≤ 40%) corresponds to blue, medium risk (40% < probability ≤ 70%) corresponds to yellow, and high risk (probability > 70%) corresponds to red. Within each interval, the color tone is further subdivided according to the probability value—for example, nodes with a probability of 32% are marked with light blue, nodes with a probability of 58% are marked with medium yellow, and nodes with a probability of 89% are marked with dark red. Simultaneously, during the mapping process... The risk probability value of each node will be directly marked next to the corresponding color block, forming risk heat map information. For example, in the risk heat map distribution of a certain fast-moving consumer goods supply chain, raw material supplier X (probability 92%) is marked with a dark red color block, regional warehouse Y (probability 65%) is marked with a medium yellow color block, and terminal store Z (probability 28%) is marked with a light blue color block. The size of the color block will also be adjusted according to the business importance of the node (such as the proportion of annual transaction volume). The color block area of ​​important nodes is 1.5 times larger than that of ordinary nodes, making the risk level and node importance clear at a glance. Next, the predicted impact range is delineated. First, all nodes that may be affected by the risk are marked in the supply chain network topology diagram. Then, these nodes are enclosed by dashed boxes. The shape of the dashed boxes is adjusted according to the actual distribution of the nodes. If the affected nodes are concentrated in the upstream raw material supply link of the supply chain, the dashed box will be "elongated" and cover the area. If the impact range spans multiple links in the upstream and downstream, the dashed box will be "irregular polygon". At the same time, the number of affected nodes (such as "12 nodes in total") and the core affected business links (such as "raw material procurement - finished product warehousing") are marked on the outside of the dashed box to form a clear risk diffusion boundary. For example, the risk diffusion boundary of an electronics supply chain includes 3 chip suppliers, 2 motherboard manufacturers, and 4 finished product assembly plants. The outside is marked "9 nodes, core impact: chip supply - motherboard production".

[0036] Subsequently, based on the risk heat distribution information and the risk diffusion boundary, report template filling processing is carried out. The report template used is designed in advance according to the structure of "Risk Profile - Details of Heat Distribution - Explanation of Diffusion Boundary - Preliminary Suggestions". When filling, first insert the visualization chart corresponding to the risk heat distribution information into the "Details of Heat Distribution" section, and supplement a text description below the chart to explain the risk probability intervals corresponding to each color block and the key high-risk nodes (such as "The dark red oil paint block corresponds to a risk probability > 80%, where Supplier X and Supplier Y are the core high-risk nodes"); then insert the risk diffusion boundary map into the "Explanation of Diffusion Boundary" section, and synchronously supplement the business types and association relationships of each node within the boundary (such as "There are 5 upstream supply nodes and 4 midstream production nodes within the boundary, and the nodes are associated through monthly purchase orders"), thus building the initial early warning report framework. Finally, based on the preset decision rule library, content supplementation processing is carried out on the initial early warning report framework. The decision rule library stores response strategies and reference data for different risk scenarios - for example, when the high-risk node is a raw material supplier, the rule library will provide response solutions such as "Enable alternative suppliers" and "Adjust the procurement cycle", as well as the success rates of responses in past similar scenarios (such as "In cases where alternative suppliers were enabled in 2024, 90% achieved risk mitigation"); when the risk diffusion boundary covers the warehousing link, suggestions such as "Temporarily allocate inventory" and "Optimize the warehousing layout" and the corresponding cost estimates will be supplemented (such as "Temporarily allocating inventory is expected to increase logistics costs by 12%"). These contents are supplemented to the "Preliminary Suggestions" section of the report, and at the same time, speculation on the potential causes of the risk occurrence is supplemented to the "Risk Profile" section (such as "Combined with historical data, the high risk of Supplier X may be related to recent fluctuations in raw material prices"), and finally a risk early warning report with complete content and sufficient data support is formed. For example, in the risk early warning report of a certain machinery manufacturing enterprise, the "Preliminary Suggestions" section not only includes specific solutions such as "Enable alternative steel supplier A, and it is expected to resume supply within 1 week", but also marks "The success rate of this solution in similar risks in March 2024 was 85%, and it is expected to increase procurement costs by 8%", providing comprehensive reference information for decision-makers. Throughout the process, template filling ensures the correspondence between the chart and the text description, avoiding inconsistent situations between the chart and the text description, while content supplementation adjusts the priority of suggestions according to the actual severity of the risk. In high-risk scenarios, emergency response measures are supplemented first, and in medium- and low-risk scenarios, long-term optimization suggestions are emphasized to ensure the practicality and pertinence of the report content.

[0037] In a specific embodiment, the content supplementation processing of the initial early warning report framework based on the preset decision rule library to obtain a risk early warning report includes: The risk heat map distribution information in the initial early warning report framework is processed to classify risk levels. Different color areas are classified and judged by preset risk level thresholds to obtain high, medium and low risk area identifiers. Key nodes are extracted from each of the high, medium and low risk area identifiers. Based on the key node extraction results, rule matching processing is performed on the preset decision rule base. Through dual matching of node type and risk level, the corresponding prevention and control suggestion template is selected from the decision rule base. Based on the prevention and control suggestion template, the content of the initial early warning report framework is filled to obtain a risk early warning report containing specific prevention and control measures.

[0038] Specifically, the risk level classification process is first carried out on the risk heat map distribution information in the initial early warning report framework. The preset risk level thresholds are clearly defined: color areas with a risk probability > 70% are identified as high-risk areas, those with a risk probability ≤ 70% (40% < risk probability) are identified as medium-risk areas, and those with a risk probability ≤ 40% are identified as low-risk areas. During the classification, the probability values ​​of each color block in the heat map are verified. For example, in a food supply chain heat map, the dark red block marking flour supplier A has a risk probability of 91%, which is identified as a high-risk area according to the threshold; the medium yellow block marking baking workshop B has a risk probability of 62%, which is identified as a medium-risk area; and the light blue block marking distribution center C has a risk probability of 35%, which is identified as a low-risk area. Subsequently, corresponding high, medium, and low risk area labels are added to each identified area. The labels are presented in the format of "risk level + area name", such as "high risk - flour supplier A area" and "medium risk - baking workshop B area". After the risk level classification is completed, key nodes are extracted for each region. The extraction is based on the business weight of the node in the supply chain (such as the percentage of annual transaction volume and the centrality of the supply chain network). Typically, the top 3 nodes with the highest business weight in the region are selected as key nodes. For example, in the high-risk region of flour supplier A, this region includes three production bases of supplier A. Base 1 accounts for 65% of the annual transaction volume and has a network centrality of 0.82, Base 2 accounts for 25% and has a centrality of 0.61, and Base 3 accounts for 10% and has a centrality of 0.35. Ultimately, Base 1 and Base 2 are extracted as key nodes for this high-risk region. In the medium-risk region of baking workshop B, which includes two production lines, production line 1 accounts for 70% of the daily output and is extracted as a key node. In the low-risk region of distribution center C, due to the small differences in the business weight of each storage point, only storage point 1 with the largest throughput is extracted as a key node.

[0039] Next, based on the key node extraction results, the preset decision rule base is processed for rule matching. The decision rule base stores prevention and control suggestion templates in a two-dimensional dimension of "node type-risk level". The node types cover 12 categories, including suppliers, production workshops, warehousing centers, and distribution nodes. Each type of node has a unique template corresponding to different risk levels. During matching, the type of key node is first determined, and then a double matching is performed based on the level of the risk area to which it belongs. For example, the key node "Flour Supplier A Base 1" extracted from the high-risk area belongs to the "Raw Material Supplier" type and corresponds to the high-risk level. In the rule base, the "Raw Material Supplier - High Risk" prevention and control suggestion template is matched, which includes three core suggestions and operation steps: "Activate backup supplier for emergency procurement", "Negotiation with existing supplier for expedited production", and "Adjust downstream production plan". The key node "Baking Workshop B Production Line 1" in the medium-risk area belongs to the "Production Workshop" type and is matched with the "Production Workshop - Medium Risk" template, which includes suggestions such as "Increase equipment inspection frequency", "Adjust backup production line", and "Optimize production schedule". The key node "Distribution Center C Warehouse Point 1" in the low-risk area belongs to the "Warehouse Center" type and is matched with the "Warehouse Center - Low Risk" template, which includes suggestions such as "Strengthen dynamic inventory monitoring" and "Plan replenishment cycle in advance". Subsequently, the initial early warning report framework was populated with content based on the matched prevention and control suggestion templates. During the population process, specific parameters were added based on the actual situation of key nodes. For example, under the suggestion of "Activate backup supplier for emergency procurement" in the "Raw Material Supplier - High Risk" template, the detail that "Backup supplier D can supply 50 tons of flour per day, with a procurement cycle of 3 days, and the cost is 8% higher than that of the existing supplier A" was added. Under the suggestion of "Deploy backup production line" in the "Production Workshop - Medium Risk" template, the information that "Backup production line 2 can produce 20,000 boxes of bread per day, and equipment debugging needs to be completed 4 hours in advance" was added. At the same time, the prevention and control suggestions for each key node were sorted according to the risk level. High-risk nodes were suggested to be placed in the "Emergency Response Measures" section of the report, and medium- and low-risk nodes were suggested to be placed in the "Routine Prevention and Control Suggestions" section. Finally, a risk early warning report containing specific prevention and control measures, operating parameters, and implementation priorities was formed. Throughout the process, rule matching automatically excludes outdated suggestion templates from the library (such as schemes that have not been updated for more than 2 years) to ensure the timeliness of the prevention and control suggestions. When filling in the content, the basic information of key nodes (such as production capacity and contact information) will be checked to avoid parameter errors and make the prevention and control measures in the report practically operable. For example, for the backup supplier D information supplemented for "flour supplier A base 1", its current production capacity and cooperation qualifications will be verified in advance to ensure that the suggestions are feasible.

[0040] In a specific embodiment, pushing the risk warning report to relevant decision-makers to support risk prevention and emergency response includes: The risk warning report is processed for decision-making role matching. By extracting the risk level, business modules involved and key node information from the report, it is compared with a preset decision-maker role database to obtain a list of target decision-makers. The decision-makers in the target decision-maker list are prioritized, and priority scores are calculated based on the decision-maker's scope of responsibility and risk correlation to obtain the decision-maker push results. Based on the decision-maker push results, the risk warning report will be pushed to relevant decision-makers to support risk prevention and emergency response.

[0041] Specifically, the risk warning report first undergoes decision-making role matching processing. Core information is extracted from the report: risk level (e.g., high risk), involved business modules (e.g., raw material procurement, manufacturing), and key nodes (e.g., steel supplier A, stamping workshop A). ​​This information is then compared item by item with a pre-defined database of decision-making roles. The database establishes a mapping relationship based on "scope of responsibility - associated business module - corresponding risk level." For example, the procurement director's scope of responsibility includes raw material supplier management, and the associated business module is "raw material procurement," corresponding to all risk levels; the production manager's scope of responsibility covers production workshop operations, and the associated business module is "manufacturing," also corresponding to all risk levels; while the warehouse manager is only associated with the "warehousing and logistics" module, and will not match if the report does not involve this module. Taking a risk warning report from an auto parts company as an example, the extracted information is "high risk, involving raw material procurement (steel supplier A) and production manufacturing (stamping workshop A), with key nodes being steel supplier A's factory 1 and stamping workshop A's production line 2". After comparison with the database, the purchasing director (related to raw material procurement), the production manager (related to production manufacturing), and the supply chain vice president (whose responsibilities cover all business modules) all meet the matching conditions, thus generating a list of target decision-makers including these three individuals.

[0042] Next, the decision-makers in the target decision-maker list are prioritized, with priority scores calculated based on a weighted average of "responsibility scope matching degree (40%) + risk relevance degree (60%)". The responsibility scope matching degree is scored as follows: "complete match = 100 points, partial match = 60 points, indirect relevance = 30 points". For example, the Purchasing Director's responsibility scope completely matches the "raw material procurement" module, so he gets 40 points (100 points × 40%); the Production Manager completely matches the "production and manufacturing" module, so he gets 40 points; although the Supply Chain Vice President covers all modules, the matching degree for a single module is calculated as partial match, so he gets 24 points (60 points × 40%). Risk relevance is calculated based on the decision-maker's control authority over key nodes. Direct control of key nodes earns 100 points, indirect control earns 60 points, and no control authority earns 0 points. Each point is then multiplied by a 60% weight. For example, a purchasing director directly controls steel supplier Factory A (100 points × 60%), earning 60 points; a production manager directly controls stamping workshop production line A (2), earning 60 points; and a supply chain vice president indirectly controls all key nodes, earning 36 points (60 points × 60%). The final total score is: Purchasing Director 100 points (40 + 60), Production Manager 100 points (40 + 60), and Supply Chain Vice President 60 points (24 + 36). If total scores are the same, ranking is based on job level (e.g., if the Supply Chain Vice President's job level is higher than the Purchasing Director's, they take priority if total scores are the same). This results in a decision-maker recommendation order of "Supply Chain Vice President → Purchasing Director → Production Manager" (Note: Here, because the Supply Chain Vice President has a higher job level, they are prioritized despite a lower total score; the actual ranking needs to be adjusted according to the company's job level system).

[0043] Finally, based on the decision-maker push results, the risk warning report is pushed to relevant decision-makers. The push method is selected according to the company's existing system. If there is an internal collaboration platform, it is sent through the platform with a "high priority" and "return within 24 hours" label; otherwise, it is pushed via encrypted email with the subject line marked "[Urgent] Supply Chain Risk Warning Report - High Risk - Involving Raw Material Procurement and Manufacturing," ensuring that decision-makers can quickly identify its importance. Personalized viewing permissions are assigned to different decision-makers during the push. For example, the purchasing director can view detailed risk data for steel supplier A (such as historical cooperation issues and a list of backup suppliers), the production manager can view production adjustment suggestions for stamping workshop A, and the supply chain vice president can view the complete report and all node data. This supports different decision-makers in carrying out targeted risk prevention and emergency response—the purchasing director can initiate emergency procurement based on the backup supplier information in the report, the production manager can arrange backup capacity based on production line adjustment suggestions, and the supply chain vice president can coordinate the overall prevention and control plan to ensure that risks are handled efficiently. Throughout the process, the push system automatically records the delivery status. If it is not read within 2 hours, an SMS reminder is triggered to avoid affecting the efficiency of emergency response due to information delays, and push logs are retained for subsequent traceability.

[0044] Please see Figure 2 , Figure 2 This is a schematic diagram of the framework of an embodiment of the supply chain risk early warning system enabled by the model of this application. Figure 2 As shown, the model-enabled supply chain risk early warning system includes a detection unit 1, which detects whether there are abnormal order blocks in the collected supply chain data. If so, it reconstructs the supply chain path based on the abnormal order blocks to obtain a risk link diagram. An analysis unit 2 is used to input the risk link diagram into a preset time-series neural network model for risk transmission analysis to obtain the risk probability distribution and predicted impact range. A push unit 3 is used to generate a risk early warning report based on the risk probability distribution and predicted impact range, and push the risk early warning report to relevant decision-makers to support risk prevention and emergency response.

[0045] Reference Figure 3 This invention also provides a computer device whose internal structure can be as follows: Figure 3 As shown, the computer device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores the data corresponding to this embodiment. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0046] Those skilled in the art will understand that Figure 3 The structures shown are merely block diagrams of some structures related to the present invention and do not constitute a limitation on the computer devices on which the present invention is applied.

[0047] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0048] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the present invention and embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0049] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0050] The description of the various embodiments above tends to emphasize the differences between the various embodiments. The similarities or similarities between them can be referred to, and for the sake of brevity, they will not be repeated here.

[0051] In the several embodiments provided in this application, it should be understood that the disclosed methods and apparatus can be implemented in other ways. For example, the apparatus implementations described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms.

[0052] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0053] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0054] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0055] If the technical solution of this application involves personal information, the product using this technical solution has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using this technical solution has obtained the user's separate consent before processing the sensitive personal information, and also meets the requirement of "express consent". For example, at personal information collection devices such as cameras, clear and prominent signs are set up to inform users that they have entered the scope of personal information collection and that personal information will be collected. If an individual voluntarily enters the collection scope, it is deemed that they have agreed to the collection of their personal information; or on the personal information processing device, with clear signs / information informing users of the personal information processing rules, authorization is obtained from the individual through pop-up information or by asking the individual to upload their personal information; wherein, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

Claims

1. A model-enabled supply chain risk early warning method, characterized in that, Includes the following steps: The system detects whether there are abnormal order blocks in the collected supply chain data. If so, it reconstructs the supply chain path based on the abnormal order blocks to obtain a risk link diagram. The risk link diagram is input into a preset time-series neural network model for risk transmission analysis to obtain the risk probability distribution and predicted impact range. A risk warning report is generated based on the risk probability distribution and the predicted impact range, and the risk warning report is pushed to relevant decision-makers to support risk prevention and emergency response.

2. The model-enabled supply chain risk early warning method according to claim 1, characterized in that, The detection of whether there are abnormal order blocks in the collected supply chain data includes: Feature extraction is performed on the collected supply chain data to obtain order volume fluctuation values ​​and transaction frequency maps, and a visual feature vector is constructed based on the order volume fluctuation values ​​and transaction frequency maps; Based on the visualized feature vector, it is determined whether there are abnormal order blocks in the supply chain data.

3. The model-enabled supply chain risk early warning method according to claim 1, characterized in that, The process of reconstructing the supply chain path based on abnormal order blocks to obtain a risk link diagram includes: The abnormal order block is subjected to order association analysis. By using key fields such as supplier information, customer information, and product information in the order, upstream and downstream orders that have a direct transaction relationship with the orders in the abnormal order block are matched to obtain a set of associated orders. The orders in the set of associated orders are then sorted by time to obtain an order time series. Based on the order time series, the path of the associated order set is sorted out. Starting from the orders in the abnormal order block, the upstream and downstream related orders of the abnormal orders are connected in chronological order to construct an initial supply chain path diagram. The nodes and edges in the initial supply chain path diagram are visualized and labeled to obtain a risk link diagram.

4. The model-enabled supply chain risk early warning method according to claim 1, characterized in that, The step of inputting the risk link diagram into a preset time-series neural network model for risk transmission analysis to obtain the risk probability distribution and predicted impact range includes: The risk link graph is input into a preset temporal graph neural network model; the temporal graph neural network model includes a graph convolutional embedding module, a temporal recursive aggregation module, and a risk diffusion prediction module; The risk link graph is analyzed based on the graph convolutional embedding module to obtain a node topology matrix including node adjacency weights and edge connection strengths. Based on the temporal recursive aggregation module, the node topology matrix is ​​subjected to temporal dependency extraction to obtain a temporal propagation vector that includes the risk transmission delay and temporal correlation degree between nodes. Based on the risk diffusion prediction module, multi-step risk transmission calculations are performed on the time-series propagation vector to obtain the risk probability distribution and predicted impact range.

5. The model-enabled supply chain risk early warning method according to claim 1, characterized in that, The generation of a risk warning report based on the risk probability distribution and the predicted impact range includes: The risk probability distribution is visualized using color mapping rules to obtain risk heat distribution information, and the predicted impact range is bounded to obtain the risk diffusion boundary. Based on the risk heat distribution information and risk diffusion boundary, the report template is filled to obtain an initial early warning report framework. Then, based on the preset decision rule base, the content of the initial early warning report framework is supplemented to obtain a risk early warning report.

6. The model-enabled supply chain risk early warning method according to claim 5, characterized in that, The initial early warning report framework is supplemented with content based on a preset decision rule base to obtain a risk early warning report, including: The risk heat map distribution information in the initial early warning report framework is processed to classify risk levels. Different color areas are classified and judged by preset risk level thresholds to obtain high, medium and low risk area identifiers. Key nodes are extracted from each of the high, medium and low risk area identifiers. Based on the key node extraction results, rule matching processing is performed on the preset decision rule base. Through dual matching of node type and risk level, the corresponding prevention and control suggestion template is selected from the decision rule base. Based on the prevention and control suggestion template, the content of the initial early warning report framework is filled to obtain a risk early warning report containing specific prevention and control measures.

7. The model-enabled supply chain risk early warning method according to claim 1, characterized in that, The step of pushing the risk warning report to relevant decision-makers to support risk prevention and emergency response includes: The risk warning report is processed for decision-making role matching. By extracting the risk level, business modules involved and key node information from the report, it is compared with a preset decision-maker role database to obtain a list of target decision-makers. The decision-makers in the target decision-maker list are prioritized, and priority scores are calculated based on the decision-maker's scope of responsibility and risk correlation to obtain the decision-maker push results. Based on the decision-maker push results, the risk warning report will be pushed to relevant decision-makers to support risk prevention and emergency response.

8. A model-enabled supply chain risk early warning system, characterized in that, include: The detection unit is used to detect whether there are abnormal order blocks in the collected supply chain data. If they are found, the supply chain path is reconstructed based on the abnormal order blocks to obtain a risk link diagram. The analysis unit is used to input the risk link diagram into a preset time-series neural network model to perform risk transmission analysis, and obtain the risk probability distribution and predicted impact range. The push unit is used to generate a risk warning report based on the risk probability distribution and the predicted impact range, and push the risk warning report to relevant decision-makers to support risk prevention and emergency response.

9. A computer device, characterized in that, The method includes a memory and a processor that are coupled to each other, wherein the memory stores program instructions and the processor executes the program instructions to implement the model-enabled supply chain risk warning method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The system stores program instructions that can be executed by a processor, the program instructions being used to implement the model-enabled supply chain risk warning method according to any one of claims 1 to 7.

Images