A knowledge graph-based supply chain risk transmission identification and early warning system and method
By constructing a supply chain risk knowledge graph and combining it with algorithms such as Bayesian networks and Monte Carlo simulations, the problem of identifying and providing early warnings of multi-level transmission paths in risk management under complex supply chain scenarios was solved, achieving unified linkage and improved accuracy in risk management.
Patent Information
- Application Number
- CN202611445610.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-17
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies struggle to achieve unified representation of multi-source heterogeneous data, automatic identification of multi-level risk transmission paths, continuous analysis of risk probability and loss, and unified linkage of graded early warning in complex supply chain scenarios, resulting in poor risk management effectiveness.
A knowledge graph-based approach is adopted to construct a supply chain risk knowledge graph, perform multi-hop path queries and rule reasoning, and combine Bayesian networks, Monte Carlo simulations and Shapley value analysis to calculate risk quantification results and generate a comprehensive risk index to achieve graded early warning.
It enables automatic identification of multi-level transmission paths of supply chain risks, joint quantification of risk probability and loss, and coordinated monitoring of graded early warning, thereby improving the accuracy and efficiency of risk management.
Smart Images

Figure CN123117828A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of supply chain risk management and artificial intelligence technology, specifically to a knowledge graph-based supply chain risk transmission identification and early warning system and method. Background Technology
[0002] As supply chain networks continue to expand, multi-level and cross-regional supply and dependence relationships are formed among suppliers, mines, ports, shipping routes, warehouses, transportation lines, and production enterprises.
[0003] Following geopolitical events, natural disasters, port congestion, shipping route disruptions, and price fluctuations, their impacts can spread step by step along relationships such as supply, transportation, dependence, cost transmission, and substitution. The object of supply chain risk management has evolved into a complex network that includes multiple entities, relationships, and transmission processes.
[0004] Currently, supply chain risk management mainly relies on scattered business data, manual rules, and single-point thresholds for monitoring. While these methods can determine whether a certain indicator is abnormal, they are difficult to uniformly represent supply chain entities and their relationships in different business systems, and it is also difficult to identify the affected objects and transmission processes after a risk occurs.
[0005] Existing technologies have begun to use knowledge graphs to model the relationships between supply chain entities, facilities and business operations, and to use graph queries or graph reasoning to analyze the scope of risk propagation. This can improve the analysis of risk entity association and multi-level propagation. However, the relevant solutions mainly focus on displaying risk relationships, identifying the scope of impact or calculating the probability of transmission, and have not yet fully solved the problem of connecting the results of various risk analyses in complex supply chain scenarios.
[0006] Specifically, existing technologies still have the following shortcomings:
[0007] Multi-source heterogeneous data makes it difficult to continuously form a unified and queryable graph of supply chain entity relationships;
[0008] Multi-hop transmission path identification and risk probability and loss range calculation are usually performed separately, making it difficult to form a continuous analysis from risk source location to impact quantification;
[0009] The comprehensive risk index lacks explanations of key contributing factors, making it difficult to determine the main reasons for the increased risk.
[0010] There is a lack of unified linkage between the risk calculation results and the four-level early warning, regional risk display, and key node health monitoring. Summary of the Invention
[0011] The technical problem to be solved by the present invention is to overcome the above-mentioned technical defects and provide a knowledge graph-based supply chain risk transmission identification and early warning system and method that can realize multi-source heterogeneous data graph fusion, automatic identification of multi-level risk transmission paths, joint quantification of risk probability and loss, and hierarchical early warning linkage monitoring.
[0012] To address the aforementioned technical problems, the present invention provides the following technical solution: a knowledge graph-based method for identifying and warning about supply chain risk transmission, comprising the following steps:
[0013] S1: Extract risk entities and their relationships from supply chain business data according to the preset schema, and write them into a graph database to build a supply chain risk knowledge graph;
[0014] S2: Receive the risk source node and query conditions, start from the risk source node, perform multi-hop path query and rule reasoning in the knowledge graph to obtain the risk transmission path and affected nodes;
[0015] S3: Based on the risk transmission path and historical risk data, perform risk quantification calculations to obtain risk quantification results;
[0016] S4: Calculate the comprehensive risk index based on the risk quantification results, and generate early warning information by comparing it with the preset early warning threshold.
[0017] Preferably, the construction of the supply chain risk knowledge graph in S1 includes configuring entity types, relationship types, and attributes;
[0018] The corresponding validator is invoked according to the schema type to perform syntax validity and entity relationship consistency checks on the configuration changes, and the valid schema is applied to the risk knowledge graph.
[0019] Preferably, before writing to the graph database in step S1, incremental update processing of risk entities and their relationships is performed, including:
[0020] The extracted risk entities are compared with the entities in the existing map to determine whether they are new entities or entities to be updated.
[0021] Perform attribute fusion on the entity to be updated;
[0022] Add, modify, or delete corresponding relationships based on changes in entities;
[0023] Write the incrementally processed entities and relationships into the graph database.
[0024] Preferably, the rule reasoning in S2 includes calculation rules, verification rules, reasoning rules, and filtering rules;
[0025] Among them, the calculation rules dynamically calculate the target attributes based on existing attributes, the verification rules determine whether the data meets business constraints, the inference rules derive new association results, and the filtering rules exclude entities or relationships that do not meet the requirements based on risk level or transmission conditions.
[0026] Preferably, the risk quantification calculation in S3 includes risk occurrence probability calculation, loss impact range simulation, and key factor contribution analysis, wherein:
[0027] Risk occurrence probability calculation: The probability is calculated using a Bayesian network algorithm based on historical data of similar events and current environmental data.
[0028] Loss impact simulation: The Monte Carlo simulation algorithm is used to simulate the impact based on the probability of risk occurrence, the strength of entity association, and historical loss data.
[0029] Key factor contribution analysis: The Shapley value analysis algorithm is used to decompose the comprehensive risk index and calculate the contribution ratio of each risk factor.
[0030] Preferably, the risk quantification calculation in S3 is divided into parallel subtasks based on a parallel computing framework;
[0031] It also includes splitting model training and accuracy evaluation of different parameter combinations into parallel tasks based on newly added historical risk case data, comparing the risk identification accuracy of each parameter combination, and selecting model parameters that meet the evaluation conditions.
[0032] Preferably, the calculation of the comprehensive risk index in S4 includes calculating the comprehensive risk index by weighting the risk probability, the degree of impact, and the diffusion speed according to the configuration weights;
[0033] The warning thresholds include four levels: red, orange, yellow, and blue. The warning level is determined based on the comparison between the comprehensive risk index and the four levels of thresholds.
[0034] Preferred examples also include knowledge loops and regional risk heat maps;
[0035] The regional risk heat map is generated based on the geographical location of risk nodes and their warning levels;
[0036] The knowledge loop involves generating early warning information, querying the historical risk case database based on the nodes associated with the risk transmission path, generating candidate response strategies, and comparing the cost and timeliness of the candidate strategies based on the substitution relationships in the risk knowledge graph.
[0037] Data from risk identification, quantification, early warning, and handling processes will be used to generate risk audit reports and written into a historical risk case database.
[0038] Another aspect of this invention discloses a knowledge graph-based supply chain risk transmission identification and early warning system, comprising:
[0039] The data interface layer includes a data acquisition and management module, which interfaces with external acquisition sources and completes unified access to raw data;
[0040] The capability support layer includes a risk rule engine, risk algorithm components, and indicator calculation and visualization modules, among which:
[0041] The risk rule engine performs the configuration, modeling, execution, monitoring, and auditing of risk rules;
[0042] The risk algorithm component invokes the risk quantification algorithm and executes algorithm iterations;
[0043] The indicator calculation and visualization module performs indicator-based processing and visualization of data and processing results;
[0044] The business application layer includes a knowledge graph application module and a risk scenario application module, wherein:
[0045] The knowledge graph application module constructs a supply chain risk knowledge graph and performs multi-hop path queries and rule reasoning based on the knowledge graph to obtain the risk transmission path and affected nodes;
[0046] The risk scenario application module performs risk identification, early warning generation, response strategy recommendation and monitoring analysis based on the risk transmission path and risk quantification results, forming a business closed loop.
[0047] The system management layer provides identity authentication, access control, multi-tenant isolation, and standardized interface management.
[0048] Another aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the knowledge graph-based supply chain risk transmission identification and early warning method as described above.
[0049] The advantages of this invention compared to the prior art are:
[0050] This invention describes suppliers, warehouses, transportation lines and their relationships through a unified supply chain knowledge graph, enabling dispersed structured, semi-structured and document data to be used in the same risk analysis process;
[0051] By performing multi-hop graph queries and rule reasoning starting from the risk source node, the transmission hierarchy, transmission path and impact description from the risk source to the affected business node can be output, reducing the problem of not being able to identify cascading effects when relying solely on single-point indicators;
[0052] By combining Bayesian networks, Monte Carlo simulations, and Shapley value analysis, the probability of risk occurrence, loss range, and contribution of key factors are obtained in the same business process, so that the risk quantification results simultaneously have probability, scope of impact, and explanation of causes.
[0053] By weighting risk probability, impact degree, and diffusion speed to form a comprehensive risk index, and linking this index with four-level thresholds, risk heat maps, and key node health monitoring, the risk calculation results can be directly transformed into graded early warning information. Attached Figure Description
[0054] Figure 1 This is a schematic diagram of the overall system architecture of the present invention.
[0055] Figure 2 This is a schematic diagram of the risk assessment and early warning business operation process.
[0056] Figure 3 This is a screenshot of the knowledge base module page.
[0057] Figure 4 This is a screenshot of the early warning center page. Detailed Implementation
[0058] The following is in conjunction with the appendix Figure 1-4 The present invention will be described in further detail below.
[0059] like Figure 2 As shown, this demonstrates a real-world business scenario based on risk assessment and early warning, showcasing the complete business process from modules such as building a historical risk case library, building a trade compliance rule library, iterative optimization of risk models, building an industry entity relationship knowledge graph, displaying a global risk heatmap, monitoring the health of key industry nodes, identifying multi-hop risk transmission paths, dynamically calculating the probability of risk events, simulating and predicting the scope of risk losses, calculating multi-dimensional risk indices, automatically generating four-level (red, orange, yellow, blue) early warning levels, tracing the root causes of risks and quantifying factor contributions, recommending intelligent response strategies, stress testing for extreme events, pre-generating flexible adjustment plans, reusing and matching case strategies, automatically verifying document compliance, to generating risk audit reports.
[0060] This invention is a knowledge graph-based system for identifying and dynamically warning about the transmission of risks in complex supply chains. The system consists of four main parts: a data interface layer, a capability support layer, a business application layer, and a system management layer. The following provides a complete description of the composition of each part and their interrelationships.
[0061] Data Interface Layer: This is the system's business data entry point, with the data acquisition and management module at its core. This module interfaces with various external data sources, completing the unified access of raw data and providing basic data input for subsequent layered processing. It is the "business data input end" of the entire architecture.
[0062] Capability Support Layer: This is the core processing layer of the architecture, focusing on the risk rule base, integrating risk algorithm components and indicator calculation and visualization, supporting the entire risk processing and data output chain, and is divided into three main sub-modules:
[0063] The risk rule engine includes rule configuration, rule modeling, rule execution, rule monitoring and auditing, and also connects to the rule library and model library to realize the definition, execution, iteration and control of risk rules. It is the core execution carrier of risk business logic.
[0064] The risk algorithm component undertakes the processing needs of the rule engine, providing algorithm-level capability support, covering aspects such as algorithm invocation and algorithm iteration.
[0065] As the front-end of the data output and display layer, the indicator calculation and visualization is responsible for processing the incoming data, rule processing results, and algorithm calculation results into indicators to support subsequent visualization display. It also includes capabilities such as calculation task scheduling and indicator management, building a bridge connecting "data-computation-display".
[0066] Business Application Layer: This layer forms the core of the architecture for business implementation and scenario application, and is divided into two main modules: knowledge graph application and risk scenario application.
[0067] Knowledge graph applications take knowledge graph construction as the core, connect upwards to algorithm plugins and indicator calculation capabilities, and support risk scenario applications downwards, realizing graph-based modeling of risk-related entities and relationships, and providing associated dimensions and visualization views for risk identification and analysis;
[0068] Risk scenario applications encompass specific business scenarios and serve as the ultimate implementation vehicle for risk-related business. Through collaboration with knowledge graphs and capability support layers, a complete business loop of "risk identification - early warning - handling - monitoring - analysis" is formed.
[0069] System Management Layer: Provides identity authentication, access control, and multi-tenant isolation to ensure system security, auditability, and maintainability. It centrally manages the configuration of all system resources and supports functional isolation of knowledge bases, large models, and data sources based on user permissions. Following the OpenAPI 3.0 specification, it provides standardized RESTful API interfaces to facilitate integration with the tenderer's existing platform and third-party business systems. It is responsible for the entire lifecycle management of interface authentication, rate limiting, monitoring, and fault tolerance.
[0070] The functional design of this invention includes four parts: risk graph management, risk quantification and early warning visualization, risk algorithm plugin, and knowledge closed-loop application. Risk graph management is used to construct and maintain supply chain risk entities, relationships, and reasoning rules; the risk algorithm plugin is used to calculate risk probabilities, simulate loss ranges, analyze key risk factors, and optimize model parameters; risk quantification and early warning visualization is used to generate risk indices, early warning levels, and risk monitoring results; and the knowledge closed-loop application is used to generate response plans, perform stress tests, reuse historical cases, and generate audit records based on the risk graph and risk calculation results.
[0071] The risk graph management includes three functions: schema editing, incremental updates of the risk graph, and risk graph inference rules.
[0072] Schema editing
[0073] The schema editor is used to define the knowledge structure of the risk graph and configure entity types, concept types, event types, and standard types visually. It further configures the attributes of each type and the relationships between different types. Risk control experts or business personnel can define the entities, relationships, and business levels to be extracted based on specific risk types. This allows the large model to extract risk knowledge from business data and documents according to the schema, avoiding overly macro or micro extraction results and maintaining consistency in relationships between risk types.
[0074] The system sets corresponding validators for different types of schemas. When a schema is added or modified, the system identifies the type of schema to be changed, calls the corresponding validator to check the syntax validity, type definition completeness, and entity relationship consistency; only schema changes that pass the validation are applied to the risk graph. By encapsulating the validation logic for different types separately, it is possible to configure the corresponding validator when adding a new schema type without changing the validation process for other types.
[0075] Incremental update of risk map
[0076] Incremental updates to the risk graph are used to add new knowledge, modify existing knowledge, or delete outdated knowledge based on an existing risk graph. The process includes: detecting the timestamp, version number, or change log of the data source to identify newly added or modified records; extracting the changed data to a processing queue; converting the raw data into object records according to the knowledge graph construction process; linking entities in the changed data with entities in the existing risk graph to determine whether the corresponding entity is a newly added entity or an entity to be updated; performing a fusion operation on the entity to be updated; adding, modifying, or deleting relationships based on entity changes; and writing the processing results to the graph database.
[0077] The aforementioned incremental updates only process data that has changed relative to the existing risk graph, avoiding redundant extraction and reconstruction of all data. This reduces data processing volume and maintains the continuity of historical knowledge. Risk graph incremental updates are part of the knowledge graph maintenance function, and their processing results serve as the data foundation for subsequent risk graph queries and inferences.
[0078] Risk Graph Reasoning Rules
[0079] Risk graph inference rules are used to control queries, calculations, verifications, inferences, and filtering across entity relationships, reducing the expansion of irrelevant relationships caused by unlimited increases in propagation hops. The system builds a knowledge graph domain rule language based on the QL-Express rule engine, enabling business users to configure calculation rules, verification rules, inference rules, and filtering rules according to risk scenarios.
[0080] The system comprises several components: calculation rules for dynamically calculating target attributes based on existing attributes; verification rules for determining whether data conforms to business constraints; inference rules for deriving new association results based on known entity relationships; and filtering rules for determining whether to process corresponding records based on risk levels or other conditions. During rule execution, the system reads the risk source, calculation scope, maximum propagation hop count, and configured rules. Starting from the risk source node, it queries related entities and relationships, determines the allowed path for further expansion based on inference rules, and excludes entities or relationships that do not meet the conditions based on filtering rules, thus forming a targeted multi-hop risk propagation path.
[0081] Risk quantification and early warning visualization:
[0082] Risk quantification and early warning visualization includes multi-dimensional risk index calculation, four-level early warning (red, orange, yellow, blue), global risk heatmap, and health monitoring of key industry nodes. This part receives risk probability, loss or impact level, risk spread speed, and contribution of key factors from the risk algorithm plugin, and completes indicator calculation, early warning judgment, and visualization display.
[0083] Multi-dimensional risk index calculation
[0084] The system connects to the database containing risk indicators and establishes a data view. Within this view, three types of indicators—risk probability, impact level, and diffusion speed—and their weights are configured. The system supports setting default weights and custom weights, saving the weight configurations to the corresponding data view, and recalculating the comprehensive risk index based on the updated indicator data. The calculation scope can be switched between a single node, a specified region, and the entire supply chain. The system can also generate risk index trend data according to configured periods such as the past 7 days, 30 days, or 90 days.
[0085] The four-level warning system (red, orange, yellow, blue) is automatically generated.
[0086] The system compares the comprehensive risk index with four threshold levels—red, orange, yellow, and blue—to automatically determine the warning level. Each threshold level can be adjusted by the user, and the system can verify the warning results after adjustment using historical risk index data. When the corresponding threshold is reached, the system generates a warning message containing the risk level, scope, risk index, and trigger time, and pushes it via email, SMS, system pop-ups, or WeChat Work, with a push delay of no more than one minute. The warning message supports closed-loop management of reception, processing, and feedback.
[0087] Global risk heat map display
[0088] The system generates regional risk heatmaps based on the geographical location of risk nodes and the warning level. It supports switching between four geographical levels: global, national, city, and county. Users can drill down from a region to view the corresponding risk index, risk nodes, warning level, and trends. The heatmap updates every 10 minutes and supports filtering the displayed region by warning level. It can also overlay external environmental data layers such as weather, traffic, or policy data.
[0089] Industry key node health monitoring
[0090] The system uses suppliers, warehouses, and transportation lines as key nodes. It determines the health status of nodes based on node risk indices and preset health thresholds, and displays the node health, current status, region, and importance through an industry resilience dashboard. The system supports displaying the health change curves of nodes over the past 7 or 30 days; when the health level falls below the corresponding grading threshold, a node alarm is triggered, and a node risk profile containing historical risk records is generated.
[0091] Risk Algorithm Plugin:
[0092] The risk algorithm plugin employs the Dask parallel computing framework, breaking down risk calculation tasks into subtasks that can be executed in parallel. It processes historical risk data and real-time environmental data through task parallelism and data parallelism. The plugin includes Bayesian network algorithms, Monte Carlo simulation algorithms, Shapley value analysis algorithms, and reinforcement learning parameter optimization functions.
[0093] Bayesian network algorithm
[0094] The system encapsulates the probability updates of the Bayesian network into schedulable tasks, configuring the execution cycle according to minute-level, hour-level, or day-level time granularity. For the probability calculation of a single risk event and the calculation of multiple risk events superimposed, the multi-node probability update and the joint probability calculation of multiple events are broken down into parallel subtasks, which are allocated by Dask to multi-threaded, multi-processed, or multi-computing nodes for execution. Historical risk data and real-time environmental data are read in parallel and preprocessed before being input into the Bayesian network, which outputs the probability of occurrence of risk events, and the result error is controlled to not exceed 5% through calculation precision control.
[0095] Monte Carlo simulation algorithm
[0096] The system uses risk probability, entity association strength, historical loss data, and risk diffusion rate as inputs for Monte Carlo simulations. It breaks down the more than 10,000 random sampling tasks into multiple subtasks, which are then distributed across multiple processing cores or computing nodes. Each subtask invokes a random number generation component to complete the simulation. The system aggregates the calculation results from each subtask and uses mean and quantile statistics to obtain the loss interval and impact range. The system can also re-execute the simulation after adjusting individual variables such as risk probability and association strength, outputting sensitivity analysis results for the corresponding variables on the loss outcome.
[0097] Shapley value analysis algorithm
[0098] The system breaks down the task of traversing the marginal contribution of feature subsets of risk factors into multiple parallel subtasks, and calculates the Shapley value and contribution percentage of each risk factor using a Shapley value analysis component. The system identifies key risk factors that lead to an increase in the overall risk index based on their contribution, and can encapsulate the contribution thresholds of these key factors into monitoring tasks, triggering alerts when the contribution reaches the corresponding threshold.
[0099] Reinforcement learning parameter optimization
[0100] Based on newly added historical risk case data, the system breaks down model training and accuracy evaluation for different parameter combinations into parallel tasks. It compares the risk identification accuracy or strategy recommendation hit rate of each parameter combination and selects the model parameters that meet the evaluation criteria. The system configures model iteration tasks on a monthly or quarterly basis, saves model parameters and evaluation results before and after iteration, and supports historical model parameter rollback and iteration effect comparison.
[0101] Knowledge closed-loop application:
[0102] The knowledge-closed-loop application encapsulates risk graph management and risk algorithm plugins into tool capabilities that can be invoked by the intelligent agent platform. These capabilities include intelligent response strategy recommendation, extreme event stress testing, pre-generation of flexible adjustment plans, case strategy reuse and matching, automatic document compliance verification, and risk audit report generation. Each functional intelligent agent shares the risk graph, historical risk case library, trade compliance rule library, and risk algorithm output results through data exchange and command invocation.
[0103] Recommended intelligent response strategies
[0104] The intelligent response strategy recommendation function analyzes the risk event input by the user, calls the risk graph to query multi-hop risk entity nodes associated with the event, and queries the historical risk case library based on the scenario characteristics of the nodes. Based on the matched historical risk handling cases, candidate response strategies are generated, and the cost, timeliness, and applicability of the candidate strategies are compared using the substitution relationships between different nodes in the risk graph. After the user modifies the strategy variables, the system re-executes the graph query, case matching, and strategy comparison, and outputs the updated results.
[0105] The extreme event stress test receives the event type, intensity, and duration configured by the user, performs a full-chain supply chain query and risk simulation according to the configured parameters, summarizes the risk transmission path, risk probability, and loss simulation results, and generates a stress test report according to the impact assessment report template. Test scenarios, process data, and reports can be saved, and risk maps can be imported as simulation cases.
[0106] The system pre-generates flexible adjustment plans by matching historical risk cases with multi-hop risk entity nodes obtained from the risk graph query. Based on historical handling strategies and the relationships between alternative nodes in the risk graph, it generates multiple flexible adjustment plans and compares the cost, timeliness, and feasibility of each plan. After the user changes the plan variables, the system recalculates and outputs the adjusted plan results.
[0107] Case strategy reuse matching extracts the scenario features of the current risk event and calculates the similarity with the case features in the historical risk case library. The top 5 cases with the highest matching degree and their handling strategies are output according to the matching degree, providing case data for intelligent response strategy recommendation and flexible adjustment plan pre-generation.
[0108] Automatic document compliance verification parses user-uploaded PDF, Word, Excel, or XML files, extracts business information from the files, queries the trade compliance rule database, compares the extracted results with applicable rules, marks content that does not comply with the rules, and outputs the reasons for the violation and rectification suggestions.
[0109] Risk audit reports are generated by acquiring data from risk identification, risk calculation, early warning, strategy recommendation, stress testing, and handling processes, and generating reports according to user-configured risk audit report templates. The generated reports can be signed and archived through the intelligent agent platform by connecting to electronic signature and document management systems, and can also be written into the case library as historical risk cases.
[0110] The system of this invention first constructs a risk map through schema editing and knowledge extraction, and continuously supplements it with risk entities, relationships, historical risk cases, and trade compliance rules through incremental updates. When a risk event is triggered, it queries multi-hop risk transmission paths based on the risk source and inference rules selected by the user. Then, it calls the Bayesian network algorithm to calculate the risk probability of related nodes, the Monte Carlo simulation algorithm to calculate the loss range and impact scope, and the Shapley value analysis algorithm to determine key risk factors. Based on the risk probability, impact degree, and diffusion speed, it calculates a comprehensive risk index, generates a four-level early warning, and displays it through a risk heatmap and the health status of key nodes. In the risk disposal phase, the knowledge closed-loop application generates response strategies and flexible adjustment plans based on the risk map, algorithm results, and historical cases. After disposal, the entire process data generates a risk audit report and writes it into the historical risk case library, providing data for subsequent case matching, incremental updates of the risk map, and optimization of model parameters.
[0111] In specific implementation of the present invention,
[0112] Implementation Method 1: Supply Chain Risk Transmission Identification and Early Warning
[0113] First, business personnel define entity types such as suppliers, warehouses, transportation lines, and ports through a graphical schema configuration interface, define relationship types such as supply, transportation, dependence, impact, and substitution, and set entity fields, relationship attributes, and relationship weights. The system reads data from the business database, forms, and documents, performs text segmentation, entity relationship extraction, semantic alignment, and knowledge fusion; the automatically extracted results are manually verified and then written into the graph database, thus forming a supply chain knowledge graph.
[0114] When a route disruption risk occurs, the user selects the corresponding route or port entity as the risk source in the knowledge graph canvas, sets the risk label, risk level, maximum propagation hops, and inference rules, and selects the entire link as the calculation scope. The system parses the query conditions, calls the path query capability of the graph database, and searches layer by layer from the risk source for nodes with transportation, supply, or dependency relationships, obtaining multi-hop candidate paths from the risk source to warehouses, suppliers, or production enterprises, and sorts them according to relationship weight, event time, and impact scope.
[0115] The system inputs historical data on similar events and current environmental data into the Bayesian network algorithm plugin to obtain the probability of risk occurrence; this probability, entity association strength, historical loss data, and diffusion speed are then input into the Monte Carlo simulation algorithm plugin to obtain the loss range, impact range, and transmission level; finally, the Shapley value analysis algorithm plugin is used to decompose the comprehensive risk index to obtain the contribution ratio of each risk factor.
[0116] The risk quantification and early warning module calculates a comprehensive risk index from 0 to 100 based on user-configured weights, assessing risk probability, impact level, and spread rate. This index is then compared to four threshold levels: red, orange, yellow, and blue. When the index reaches a given threshold, the system generates an early warning message that includes the warning level, affected area, risk index, and trigger time. The system updates the area color on the risk heatmap and simultaneously updates the health status of relevant suppliers, warehouses, and transportation lines.
[0117] Implementation Method 2: Extreme Event Stress Test
[0118] Users input hypothetical events and parameters through the extreme event stress test function, such as a 30% increase in oil prices, and select the entire supply chain as the simulation scope. The system uses these event parameters as input for risk calculation, identifies potentially affected nodes and paths based on the transportation, supply, and cost relationships in the supply chain knowledge graph, and calls Bayesian network and Monte Carlo simulation algorithms to calculate the risk probability, cost impact, and loss range of each node, outputting an assessment report that includes the risk transmission path and loss estimates.
[0119] The knowledge loop application module generates multiple flexible adjustment plans based on stress test results and alternative node information in the knowledge graph, displaying the cost, timeliness, and applicable conditions of each plan. Users can modify plan parameters and re-execute the simulation. Events, analysis results, handling processes, and handling effects that have been manually confirmed are written into the historical risk case library. Subsequent risk events can call up similar cases and corresponding handling strategies through similarity matching.
[0120] The contents not described in detail in this specification are existing technologies known to those skilled in the art.
[0121] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.
[0122] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.
Claims
1. A knowledge graph-based supply chain risk transmission identification and early warning method, characterized in that: Includes the following steps: S1: Extract risk entities and their relationships from supply chain business data according to the preset schema, and write them into a graph database to build a supply chain risk knowledge graph; S2: Receive the risk source node and query conditions, start from the risk source node, perform multi-hop path query and rule reasoning in the knowledge graph to obtain the risk transmission path and affected nodes; S3: Based on the risk transmission path and historical risk data, perform risk quantification calculations to obtain risk quantification results; S4: Calculate the comprehensive risk index based on the risk quantification results, and generate early warning information by comparing it with the preset early warning threshold.
2. The knowledge graph-based supply chain risk transmission identification and early warning method according to claim 1, characterized in that: The construction of the supply chain risk knowledge graph in S1 includes configuring entity types, relationship types, and attributes; The corresponding validator is invoked according to the schema type to perform syntax validity and entity relationship consistency checks on the configuration changes, and the valid schema is applied to the risk knowledge graph. 3.The knowledge graph-based supply chain risk transmission identification and early warning method according to claim 1, characterized in that: Before writing to the graph database in step S1, incremental update processing of risk entities and their relationships is performed, including: The extracted risk entities are compared with the entities in the existing map to determine whether they are new entities or entities to be updated. Perform attribute fusion on the entity to be updated; Add, modify, or delete corresponding relationships based on changes in entities; Write the incrementally processed entities and relationships into the graph database.
4. The knowledge graph-based supply chain risk transmission identification and early warning method according to claim 1, characterized in that: The rule-based reasoning in S2 includes calculation rules, verification rules, reasoning rules, and filtering rules; Among them, the calculation rules dynamically calculate the target attributes based on existing attributes, the verification rules determine whether the data meets business constraints, the inference rules derive new association results, and the filtering rules exclude entities or relationships that do not meet the requirements based on risk level or transmission conditions.
5. The knowledge graph-based supply chain risk transmission identification and early warning method according to claim 1, characterized in that: The risk quantification calculation in S3 includes risk occurrence probability calculation, loss impact range simulation, and key factor contribution analysis, wherein: Risk occurrence probability calculation: The probability is calculated using a Bayesian network algorithm based on historical data of similar events and current environmental data. Loss impact simulation: The Monte Carlo simulation algorithm is used to simulate the impact based on the probability of risk occurrence, the strength of entity association, and historical loss data. Key factor contribution analysis: The Shapley value analysis algorithm is used to decompose the comprehensive risk index and calculate the contribution ratio of each risk factor.
6. The knowledge graph-based supply chain risk transmission identification and early warning method according to claim 5, characterized in that: In S3, the risk quantification calculation is broken down into parallel subtasks based on a parallel computing framework. It also includes splitting model training and accuracy evaluation of different parameter combinations into parallel tasks based on newly added historical risk case data, comparing the risk identification accuracy of each parameter combination, and selecting model parameters that meet the evaluation conditions.
7. The knowledge graph-based supply chain risk transmission identification and early warning method according to claim 1, characterized in that: The calculation of the comprehensive risk index in S4 includes weighting the risk probability, impact degree and diffusion speed according to the configuration weights to obtain the comprehensive risk index; The warning thresholds include four levels: red, orange, yellow, and blue. The warning level is determined based on the comparison between the comprehensive risk index and the four levels of thresholds.
8. The knowledge graph-based supply chain risk transmission identification and early warning method according to claim 7, characterized in that: This also includes knowledge loops and regional risk heat maps; The regional risk heat map is generated based on the geographical location of risk nodes and their warning levels; The knowledge loop involves generating early warning information, querying the historical risk case database based on the nodes associated with the risk transmission path, generating candidate response strategies, and comparing the cost and timeliness of the candidate strategies based on the substitution relationships in the risk knowledge graph. Data from risk identification, quantification, early warning, and handling processes will be used to generate risk audit reports and written into a historical risk case database.
9. A knowledge graph-based supply chain risk transmission identification and early warning system, using the supply chain risk transmission identification and early warning method as described in any one of claims 1-8, characterized in that: include: The data interface layer includes a data acquisition and management module, which interfaces with external acquisition sources and completes unified access to raw data; The capability support layer includes a risk rule engine, risk algorithm components, and indicator calculation and visualization modules, among which: The risk rule engine performs the configuration, modeling, execution, monitoring, and auditing of risk rules; The risk algorithm component invokes the risk quantification algorithm and executes algorithm iterations; The indicator calculation and visualization module performs indicator-based processing and visualization of data and processing results; The business application layer includes a knowledge graph application module and a risk scenario application module, wherein: The knowledge graph application module constructs a supply chain risk knowledge graph and performs multi-hop path queries and rule reasoning based on the knowledge graph to obtain the risk transmission path and affected nodes; The risk scenario application module performs risk identification, early warning generation, response strategy recommendation and monitoring analysis based on the risk transmission path and risk quantification results, forming a business closed loop. The system management layer provides identity authentication, access control, multi-tenant isolation, and standardized interface management.
10. A computer readable storage medium having stored thereon a computer program, characterized in that: When the computer program is executed by the processor, it implements the knowledge graph-based supply chain risk transmission identification and early warning method as described in any one of claims 1 to 8.
