Supply chain risk quantitative evaluation method and system based on dynamic matter spectrum

Abstract

The present application relates to the technical field of risk analysis, in particular to a supply chain risk quantitative evaluation method and system based on a dynamic matter graph, comprising collecting multi-source heterogeneous data, constructing a four-dimensional space-time model containing a time dimension, a geographical space dimension, a supply chain network space dimension and a risk influence space dimension, and representing a supply chain event as four-dimensional space-time data; identifying a supply chain entity from the four-dimensional space-time data, extracting a risk event and constructing a matter graph, the matter graph taking a risk event as a node and an evolution relationship between events as an edge, and calculating a relationship weight between events; calculating a probability quantitative value of a risk transmission path based on the matter graph, and obtaining a comprehensive risk score of a target entity; generating a risk mitigation strategy based on the comprehensive risk score; monitoring a deviation between an actual risk occurrence and a prediction result, and updating the matter graph and the risk mitigation strategy through an adaptive parameter optimization and an incremental learning mechanism, thereby forming a self-evolving risk evaluation system.

Description

Technical Field

[0001] This invention relates to the field of risk analysis technology, specifically to a method and system for quantitative assessment of supply chain risks based on dynamic reasoning graphs. Background Technology

[0002] With the increasing complexity of global supply chain networks, supply chain risk assessment has become a crucial aspect of corporate management and financial supervision. Current technologies for supply chain risk assessment employ methods such as historical data analysis, scorecard models, or construct risk assessment models based on corporate financial indicators, credit records, and industry data. However, these methods all have significant limitations: first, they lack analysis of the evolutionary patterns of risk events, failing to capture the dynamic process of risk transmission within the supply chain; second, risk quantification is primarily based on historical statistical data, lacking sufficient ability to identify new data; and third, they lack effective representation of the causal and temporal relationships among multi-dimensional risk factors. Event graphs, as a knowledge representation method, construct directed graph structures with events as nodes and evolutionary relationships between events as edges, capable of describing the logical relationships of events, such as sequence, causality, and conditions. However, existing event graph technologies lack event abstraction modeling methods and risk quantification assessment mechanisms specific to supply chain scenarios, failing to meet the needs of accurate supply chain risk assessment and early warning in actual business operations.

[0003] To address this, a method and system for quantitative assessment of supply chain risks based on dynamic event graphs are proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for quantitative assessment of supply chain risks based on dynamic event graphs. This includes collecting multi-source heterogeneous data, constructing a four-dimensional spatiotemporal model encompassing time, geospatial, supply chain network, and risk impact dimensions, representing supply chain events as four-dimensional spatiotemporal data; identifying supply chain entities from the four-dimensional spatiotemporal data using entity recognition algorithms, extracting risk events, and constructing an event graph. The event graph uses risk events as nodes and evolutionary relationships between events as edges, calculating the weights of these relationships; calculating the probability quantification of risk transmission paths based on the event graph to obtain a comprehensive risk score for the target entity; generating risk mitigation strategies based on the comprehensive risk score; monitoring the deviation between actual risk occurrence and predicted results, and updating the event graph and risk mitigation strategies through adaptive parameter optimization and incremental learning mechanisms to form a self-evolving risk assessment system.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A supply chain risk quantification assessment method and system based on dynamic event graphs includes:

[0007] Collect multi-source heterogeneous data, label the multi-source heterogeneous data with spatiotemporal coordinates, and construct a four-dimensional spatiotemporal model that includes time dimension, geographic space dimension, supply chain network space dimension and risk impact space dimension, and represent supply chain events as four-dimensional spatiotemporal data;

[0008] Based on the entity recognition algorithm, supply chain entities are identified from the four-dimensional spatiotemporal data, risk events are extracted and a process graph is constructed. The process graph uses risk events as nodes and the evolutionary relationship between events as edges, and a dynamic weight update mechanism is used to calculate the relationship weight between events.

[0009] Based on the aforementioned event graph, the probability of risk transmission paths is calculated to obtain the comprehensive risk score of the target entity;

[0010] Based on the comprehensive risk score, risk mitigation strategies are generated, including supplier replacement strategies, inventory adjustment strategies, and transportation route optimization strategies.

[0011] By monitoring the deviation between the actual occurrence of risks and the predicted results, and updating the underlying logic graph and risk mitigation strategies through adaptive parameter optimization and incremental learning mechanisms, a self-evolving risk assessment system is formed.

[0012] Preferably, constructing the four-dimensional spatiotemporal model includes: using standardization to convert multi-source heterogeneous data into a unified RDF triple format; the multi-source heterogeneous data includes business registration data, judicial litigation data, tax invoice data, and logistics and transportation data; extracting timestamp information for each type of data and standardizing it into time coordinates; converting address information into latitude and longitude coordinates through geocoding services; mapping the upstream and downstream positions of entities in the supply chain to network topology coordinates; determining the risk impact radius based on the risk event type and impact range; and establishing a four-dimensional R-tree index structure to support spatiotemporal range queries to obtain four-dimensional spatiotemporal data.

[0013] Preferably, the construction of the event graph includes:

[0014] Entity recognition: Named entity recognition is performed on text content in four-dimensional spatiotemporal data to identify entity types such as suppliers, manufacturers, products, raw materials, transportation routes, and warehouse supply chains;

[0015] Risk event extraction: Based on dependency parsing and semantic role labeling technology, risk events such as supply disruption, quality problems, price fluctuations and logistics delays are extracted from four-dimensional spatiotemporal data. Each risk event is represented as a four-tuple structure containing risk source entity, risk type, affected object and severity.

[0016] Relationship extraction: Identify sequential, causal, and conditional relationships among risk events through text matching and rule-based reasoning, and construct directed edges to connect related event nodes;

[0017] Dynamic weight calculation: The time decay weight is calculated by combining the time interval between events, the spatial influence weight is calculated based on the four-dimensional spatial distance, and the structural weight is calculated based on the centrality of the entity network. The dynamic weight values ​​of the edges between events are obtained by combining the results.

[0018] Preferably, the step of obtaining the comprehensive risk score includes:

[0019] Search the event graph for all transmission paths from the risk source event to the target entity, and record the intermediate nodes and edge weights of each path. Calculate the transmission probability of a single path by applying the conditional probability formula to the dynamic weights of all edges on the path. Calculate the joint impact probability of multiple paths on the target entity, and handle overlapping impacts between paths using the conditional independence assumption. Combining five dimensions—time urgency, spatial coverage, network criticality, impact severity, and propagation speed—calculate the comprehensive risk score of the target entity by weighted summation of the transmission probability and joint impact probability, and map it to four risk levels: low, medium, high, and extremely high.

[0020] Preferably, the risk mitigation strategy includes: modeling suppliers, manufacturers, and distributors in the supply chain as independent intelligent agents with state space, action space, and reward function; each intelligent agent determines its own state risk level based on the current comprehensive risk score and selects the optimal action from a predefined set of actions;

[0021] Supplier replacement strategy generation steps: Set screening criteria based on the current supplier risk level, conduct multi-objective optimization evaluation of candidate suppliers in terms of cost, quality and reliability, and output the optimal replacement plan and switchover timeline;

[0022] Inventory adjustment strategy generation steps: Calculate the inventory safety factor based on the risk level, combine historical demand data and supply lead time, and use the economic order quantity model to calculate the optimal inventory level;

[0023] The steps for generating a transportation route optimization strategy are as follows: Analyze the risk nodes of the current transportation route and use the shortest path algorithm to replan alternative transportation routes that avoid high-risk areas.

[0024] Preferably, the self-evolutionary risk assessment system update includes:

[0025] A sliding time window is set to collect real-time data on actual risk events, and the absolute deviation between the predicted probability and the actual probability, the deviation between the predicted time and the actual time, and the deviation between the predicted severity and the actual severity are calculated. When the overall deviation exceeds the first threshold, the time decay coefficient, spatial influence coefficient, and network weight coefficient are adjusted. When the overall deviation exceeds the second threshold, the event graph structure is updated, new event nodes and relationship edges are added, and the probability of risk transmission paths is recalculated. Performance verification is performed after each model update, and if the verification score is lower than the performance threshold, the system is rolled back to the previous version.

[0026] A supply chain risk quantification assessment system based on dynamic event graphs includes:

[0027] The system comprises the following components: a spatiotemporal data acquisition unit, which collects multi-source heterogeneous data, annotates it with spatiotemporal coordinates, constructs a four-dimensional spatiotemporal model, and represents supply chain events as four-dimensional spatiotemporal data; a causal graph construction unit, which identifies supply chain entities from the four-dimensional spatiotemporal data, extracts risk events, and constructs a causal graph; a risk assessment unit, which calculates the probability of risk transmission paths based on the causal graph and obtains a comprehensive risk score for the target entity; a risk strategy generation unit, which generates risk mitigation strategies based on the comprehensive risk score; and a risk system update unit, which monitors the deviation between the actual risk occurrence and the predicted results, updates the causal graph and risk mitigation strategies, and forms a self-evolving risk assessment system.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] 1. This invention introduces a four-dimensional spatiotemporal model that includes time, geospatial, supply chain network, and risk impact dimensions. By standardizing multi-source heterogeneous data, data from multiple fields such as industry and commerce, judiciary, taxation, and logistics can be fused and modeled within the same spatiotemporal semantic framework, forming structured four-dimensional spatiotemporal data. This model can dynamically and in real-time represent the spatiotemporal distribution and network location relationships of risk events, providing a high-precision and queryable data foundation for subsequent risk propagation analysis and response strategy formulation, thus improving modeling efficiency and risk data processing capabilities in large-scale data environments.

[0030] 2. This invention proposes a technical approach based on constructing a dynamic event-based causal graph. Through steps such as entity identification, risk event extraction, relationship identification, and dynamic edge weight updating, a causal evolution network among risk events is established. The transmission impact between events is dynamically quantified, making the event-based graph realistically interpretable and logically feasible in representing risk propagation paths. Combining five indicators—time urgency, spatial coverage, network criticality, propagation speed, and impact severity—the risk exposure degree of target entities is comprehensively scored and graded, adapting to the trend of risk dynamic evolution over time and space. This significantly improves the accuracy of risk quantification and the reconstruction of propagation mechanisms.

[0031] 3. This invention introduces multi-agent modeling, treating suppliers, manufacturers, and distributors in the supply chain as independent agents with state spaces, action spaces, and reward functions. Each agent selects its optimal risk response behavior based on its own risk status, generating personalized and executable mitigation strategies. Strategies such as supplier replacement, inventory adjustment, and transportation route optimization, through the interconnected analysis of current risk levels, historical data, and business objectives, employ multi-objective optimization and shortest path algorithms to form highly operational dynamic response solutions, enhancing the supply chain's resilience in the face of risks. This invention possesses the capabilities of continuous model learning, self-correction, and real-time strategy iteration, improving the intelligence level of risk control and its ability to adapt to complex environmental changes. Attached Figure Description

[0032] Figure 1 A schematic diagram of the supply chain risk quantification assessment method based on dynamic event graph provided by the present invention;

[0033] Figure 2 A schematic diagram of the supply chain risk quantification assessment system based on dynamic event graph provided by the present invention;

[0034] Figure 3 A schematic diagram illustrating the steps for obtaining the comprehensive risk score provided by this invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] This invention provides a supply chain risk quantification assessment method based on dynamic event graphs, applicable to a supply chain risk quantification assessment system based on dynamic event graphs. The method flowchart is shown below. Figure 1The system architecture diagram includes a spatiotemporal data acquisition unit, a causal graph construction unit, a risk assessment unit, a risk strategy generation unit, and a risk system update unit. See the attached diagram for details. Figure 2 The specific technical solution is as follows:

[0037] Collect multi-source heterogeneous data, label the multi-source heterogeneous data with spatiotemporal coordinates, and construct a four-dimensional spatiotemporal model that includes time dimension, geographic space dimension, supply chain network space dimension and risk impact space dimension, and represent supply chain events as four-dimensional spatiotemporal data;

[0038] The construction of the four-dimensional spatiotemporal model includes: standardizing multi-source heterogeneous data into a unified RDF triple format; the multi-source heterogeneous data includes business registration data, judicial litigation data, tax invoice data, and logistics and transportation data; extracting timestamp information for each type of data and standardizing it into time coordinates T; converting address information into latitude and longitude coordinates (X,Y) through geocoding services; mapping the entity's upstream and downstream positions in the supply chain to network topology coordinates N; determining the risk impact radius R based on the risk event type and impact range; combining the above four dimensions to form (T,(X,Y),N,R) coordinates; establishing a four-dimensional R-tree index structure to support spatiotemporal range queries; and obtaining four-dimensional spatiotemporal data.

[0039] In this embodiment, a four-dimensional spatiotemporal model incorporating time, geospatial dimensions, supply chain network, and risk impact is constructed to effectively integrate and analyze heterogeneous data such as business registration, legal proceedings, tax invoices, and logistics transportation. By employing RDF triple-standardized data and combining it with a four-dimensional R-tree index structure, efficient spatiotemporal range queries are supported, significantly improving the accuracy and timeliness of supply chain risk assessment and providing strong support for enterprise decision-making.

[0040] The four-dimensional R-tree index structure is optimized using a distributed edge computing architecture, including: designing a geographic location-based sharding strategy to divide global supply chain data into multiple geographic partitions according to latitude and longitude, with edge computing nodes deployed in each partition; constructing a hierarchical index structure, including a global spatiotemporal index layer, a regional aggregation index layer, and a local real-time index layer, supporting second-level risk event retrieval; implementing a dynamic load balancing mechanism to adaptively adjust the allocation of computing resources according to the risk event density and query frequency in each region; ensuring data consistency among distributed nodes by using the PBFT algorithm to handle node failures and network partitioning issues; and designing a streaming data processing engine to support incremental index updates and rapid propagation of real-time risk event streams.

[0041] Based on the entity recognition algorithm, supply chain entities are identified from the four-dimensional spatiotemporal data, risk events are extracted and a process graph is constructed. The process graph uses risk events as nodes and the evolutionary relationship between events as edges, and a dynamic weight update mechanism is used to calculate the relationship weight between events.

[0042] The construction of the principle graph includes:

[0043] Entity recognition: Named entity recognition is performed on text content in four-dimensional spatiotemporal data to identify entity types such as suppliers, manufacturers, products, raw materials, transportation routes, and warehouse supply chains;

[0044] Risk event extraction: Based on dependency parsing and semantic role labeling technology, risk events such as supply disruption, quality problems, price fluctuations and logistics delays are extracted from four-dimensional spatiotemporal data. Each risk event is represented as a four-tuple structure containing risk source entity, risk type, affected object and severity.

[0045] Relationship extraction: Identify sequential, causal, and conditional relationships among risk events through text matching and rule-based reasoning, and construct directed edges to connect related event nodes;

[0046] Dynamic weight calculation: The time decay weight is calculated by combining the time interval between events, the spatial influence weight is calculated based on the four-dimensional spatial distance, and the structural weight is calculated based on the centrality of the entity network. The dynamic weight values ​​of the edges between events are obtained by combining the results.

[0047] In this embodiment, by accurately identifying supply chain entities and extracting risk events from four-dimensional spatiotemporal data, a contextual graph is constructed with risk events as nodes and the evolutionary relationships between events as edges, enabling a structured representation and dynamic analysis of supply chain risks. A dynamic weight update mechanism is employed, comprehensively considering time, space, and structural factors to accurately calculate the weights of relationships between events, improving the timeliness and accuracy of risk assessment. This provides enterprises with efficient risk management support, helping to promptly identify and respond to potential risks and ensure the stable operation of the supply chain.

[0048] Based on the aforementioned event graph, the probability of risk transmission paths is calculated to obtain the comprehensive risk score of the target entity;

[0049] The steps for obtaining the comprehensive risk score are as follows: Figure 3 ,include:

[0050] In this embodiment, all transmission paths from the risk source event to the target entity are searched in the event graph, and the intermediate nodes and edge weights of each path are recorded. The transmission probability of a single path is obtained by applying a conditional probability formula to the dynamic weights of all edges on the path. The joint impact probability of multiple paths on the target entity is calculated, and the overlapping impact between paths is handled by the assumption of conditional independence. Combining five dimensions of indicators—time urgency, spatial coverage, network criticality, impact severity, and propagation speed—the transmission probability and joint impact probability are weighted and summed to calculate the comprehensive risk score of the target entity, which is then mapped to four risk levels: low, medium, high, and extremely high. In this embodiment, the risk transmission path probability is calculated using the event graph and combined with multi-dimensional indicators to assess the comprehensive risk of the target entity. By searching all paths, calculating the single-path probability, and handling the joint impact of multiple paths, comprehensiveness and accuracy are ensured. The comprehensive risk score is mapped to four risk levels, improving risk management efficiency and supply chain stability.

[0051] Based on the comprehensive risk score, risk mitigation strategies are generated, including supplier replacement strategies, inventory adjustment strategies, and transportation route optimization strategies.

[0052] The risk mitigation strategy includes: modeling suppliers, manufacturers, and distributors in the supply chain as independent intelligent agents with state space, action space, and reward function; each intelligent agent determines its own state risk level based on the current comprehensive risk score and selects the optimal action from a predefined set of actions;

[0053] Supplier replacement strategy generation steps: Set screening criteria based on the current supplier risk level, conduct multi-objective optimization evaluation of candidate suppliers in terms of cost, quality and reliability, and output the optimal replacement plan and switchover timeline;

[0054] Inventory adjustment strategy generation steps: Calculate the inventory safety factor based on the risk level, combine historical demand data and supply lead time, and use the economic order quantity model to calculate the optimal inventory level;

[0055] The steps for generating a transportation route optimization strategy are as follows: Analyze the risk nodes of the current transportation route and use the shortest path algorithm to replan alternative transportation routes that avoid high-risk areas.

[0056] In this embodiment, a smart agent model is constructed and risk mitigation strategies are generated by combining comprehensive risk scores to improve supply chain management efficiency. The supplier replacement strategy selects candidate suppliers based on risk levels, performs multi-objective optimization of cost, quality, and reliability, and quickly outputs the optimal replacement solution to reduce the risk of disruption. The inventory adjustment strategy uses risk levels and historical data to calculate a safety factor and optimize inventory levels to balance sudden demand and cost control. The transportation route optimization strategy analyzes risk nodes and uses the shortest path algorithm to plan alternative routes to enhance transportation stability.

[0057] The intelligent agent employs digital twin technology for enhanced modeling, including: creating a digital twin model for the supply chain, synchronizing the state, behavior, and environmental parameters of physical entities in real time; establishing a real-time data acquisition system based on an IoT sensor network, including RFID tags, GPS locators, temperature and humidity sensors, and vibration sensors, forming a two-way mapping between physical and digital spaces; designing predictive digital twin algorithms that combine physical modeling, data-driven modeling, and hybrid modeling to predict the future state evolution trajectory of entities; constructing a virtual interaction protocol between twins to simulate complex interaction relationships between supply chain entities and support risk propagation simulation in multiple scenarios; and realizing an autonomous decision-making mechanism based on digital twins, allowing the intelligent agent to pre-simulate the execution effect of risk mitigation strategies in a virtual environment and optimize decision parameters before applying them to the supply chain.

[0058] By monitoring the deviation between the actual occurrence of risks and the predicted results, and updating the underlying logic graph and risk mitigation strategies through adaptive parameter optimization and incremental learning mechanisms, a self-evolving risk assessment system is formed.

[0059] The updated risk assessment system for self-evolution includes:

[0060] A sliding time window is set to collect real-time data on actual risk events, and the absolute deviation between the predicted probability and the actual probability, the deviation between the predicted time and the actual time, and the deviation between the predicted severity and the actual severity are calculated. The absolute deviation between the predicted probability and the actual probability, the deviation between the predicted time and the actual time, and the deviation between the predicted severity and the actual severity are weighted and summed to obtain the comprehensive deviation.

[0061] When the overall deviation exceeds the first threshold, adjust the time decay coefficient, spatial influence coefficient, and network weight coefficient; when the overall deviation exceeds the second threshold, update the event graph structure, add new event nodes and relationship edges, and recalculate the probability of risk transmission paths; after each model update, perform performance verification, and if the verification score is lower than the performance threshold, roll back to the previous version.

[0062] In this embodiment, a self-evolving risk assessment system enables dynamic monitoring and prediction of supply chain risks. A sliding time window is used to collect data in real time, calculate the deviation between predictions and actual results, and automatically adjust model parameters and update the event graph to ensure the timeliness and accuracy of risk assessment. An incremental learning mechanism reduces computational overhead, while performance verification and rollback mechanisms ensure system stability. This system proactively identifies emerging risks, optimizes risk mitigation strategies, and enhances supply chain resilience and decision-making efficiency.

[0063] The intelligent decision-making mechanism employing graph neural networks includes: constructing a graph-based event graph embedding model to automatically learn the latent feature representations of event nodes and relationship edges; designing a multi-agent deep reinforcement learning framework to model each node in the supply chain as a collaborative agent and optimize the global risk control strategy through a shared reward mechanism; establishing a risk scenario generator based on generative adversarial networks to simulate the supply chain response capability under extreme risk conditions and generate risk propagation paths for strategy robustness verification; and achieving cross-domain risk knowledge transfer, enabling the model to quickly adapt to emerging risk types and industry specificities.

[0064] This invention improves the accuracy and timeliness of supply chain risk assessment by constructing a four-dimensional spatiotemporal model to integrate multi-source heterogeneous data and combining event graphs and graph neural networks to learn risk characteristics. Multi-agent deep reinforcement learning optimizes the global risk control strategy, while generative adversarial networks and digital twins enhance strategy robustness and predictive ability. Meta-learning enables cross-domain knowledge transfer. A distributed four-dimensional R-tree index and blockchain consensus mechanism support millisecond-level querying and data consistency. Adaptive parameter optimization and incremental learning form a self-evolving system that quickly adapts to emerging risks.

[0065] Example 2:

[0066] The system collects heterogeneous data from multiple sources, including business registration data, legal litigation data, tax invoice data, and logistics and transportation data. Based on the data standardization engine technology of DataFun.com, this heterogeneous data is converted into a unified RDF triple format.

[0067] In practice, business registration data includes fields such as basic enterprise information, equity structure, and business scope; judicial litigation data covers case type, party information, and judgment results; tax invoice data records information such as transaction amount, commodity category, and transaction time; and logistics and transportation data includes elements such as transportation route, cargo type, and transportation timeliness. OCR recognition technology and data cleaning and conversion algorithms are used to ensure data quality and format consistency.

[0068] Timestamp information is extracted from each type of data and standardized into time coordinates T, achieving hourly accuracy to accurately reflect the temporal characteristics of risk events. Geocoding services convert information such as company registration addresses, court jurisdiction addresses, and logistics node addresses into latitude and longitude coordinates (X,Y), covering geographical location information within the target area.

[0069] Based on the upstream and downstream positions of enterprises in the industry chain, and leveraging the operational data indicator system accumulated by DataLink, enterprises are mapped to network topology coordinates N. These coordinates reflect the hierarchical position and importance of the enterprise within the entire industry network. Based on risk event types and historical impact data, the risk impact radius R is calculated, forming a complete coordinate system (T,(X,Y),N,R).

[0070] A four-dimensional R-tree index structure is established to support spatiotemporal range queries within the target region. The index structure is hierarchically structured according to administrative divisions, with corresponding index nodes established for provincial, municipal, and county levels. Query response time is controlled within seconds to meet the business needs of real-time risk monitoring.

[0071] Based on multi-dimensional evaluation technology, named entity recognition is performed on text content in four-dimensional spatiotemporal data. In the industrial intelligent computing space environment, the types of supply chain entities that the system can identify include: upstream raw material suppliers, midstream manufacturing and processing enterprises, downstream sales and distribution companies, key product categories, core raw material categories, major transportation routes, and important warehousing facilities.

[0072] The entity recognition algorithm, combined with the application scenario indicator evaluation model accumulated by DataFinance Intelligence, can accurately distinguish the functional positioning and importance of different types of enterprises in the industrial chain. The accuracy of entity recognition is further verified through correlation analysis with business registration information.

[0073] Based on dependency parsing and semantic role labeling technologies, risk events related to the industrial chain are extracted from four-dimensional spatiotemporal data. The types of risk events cover multiple dimensions, including supply disruptions, quality issues, price fluctuations, logistics delays, cash flow disruptions, and policy changes.

[0074] Each risk event is represented as a four-tuple structure containing the risk source entity, risk type, affected entities, and severity level. For example, an event where a raw material supplier suspends production due to environmental issues would have the following four-tuple structure: (Supplier A, supply disruption; downstream manufacturing companies B, C, and D, high-risk level). The severity level is quantitatively assessed based on historical data and an expert decision-making system.

[0075] By employing text matching and rule-based reasoning techniques, the algorithm identifies sequential, causal, and conditional relationships among risk events. Within the data-driven intelligent computing space of DataFinance, the relation extraction algorithm can identify complex cross-regional and cross-industry connections and construct directed edges to link related event nodes.

[0076] Dynamic weight calculation comprehensively considers three dimensions: time decay weight is calculated based on the time interval between events, with shorter time intervals resulting in higher weights; spatial influence weight is calculated based on four-dimensional spatial distance, with events that are geographically closer or at similar levels in the industry chain having higher weights; and structural weight is calculated based on the centrality of entities in the network, with events related to core enterprises having higher propagation weights.

[0077] In the constructed event graph, the system employs a depth-first search algorithm to find all possible transmission paths from the risk source event to the target entity. During the search process, the intermediate nodes traversed by each path and their corresponding edge weights are recorded to ensure the completeness and accuracy of the path search.

[0078] Targeting the regional characteristics of data-driven intelligent connectivity services, the system can identify risk transmission paths across provinces and cities, providing decision support for coordinated regional industrial development. The search depth is set to six degrees of separation, ensuring both comprehensiveness and keeping computational complexity within an acceptable range.

[0079] The propagation probability of each risk propagation path is obtained by multiplying the dynamic weights of all edges along that path. The calculation considers the attenuation effect of path length on the propagation probability; the longer the path, the lower the probability. Probability normalization ensures that the probability values ​​of all paths are between 0 and 1.

[0080] When multiple transmission paths exist from the same risk source to the same target entity, the conditional independence assumption is used to handle the overlapping effects between paths. The joint probability of the multiple paths is calculated using the union formula in probability theory to avoid the problem of overestimating the probability value due to repeated calculations.

[0081] The comprehensive risk score of the target entity is calculated using a weighted summation method, combining five dimensions: time urgency, spatial coverage, network criticality, impact severity, and propagation speed. The weighting coefficients are adjusted and optimized based on DataLink's practical experience in different regions and its expert decision-making system.

[0082] The comprehensive risk score is mapped to four risk levels: low, medium, high, and extremely high, each corresponding to different risk response strategies and resource allocation plans. Low-risk levels are monitored routinely, medium-risk levels have increased monitoring frequency, high-risk levels activate early warning mechanisms, and extremely high-risk levels immediately implement emergency response.

[0083] Suppliers, manufacturers, and distributors in the industrial intelligent computing space are modeled as independent intelligent agents with state space, action space, and reward function. The state space includes key indicators such as the enterprise's financial status, production capacity, inventory level, and order status. The action space covers executable operations such as adjusting production plans, changing procurement strategies, optimizing inventory allocation, and adjusting sales strategies.

[0084] The reward function design comprehensively considers multiple objectives such as cost control, service quality, risk reduction, and profit maximization, and is quantitatively calculated using multi-dimensional evaluation technology from DataFinance. Each agent determines its own state risk level based on its current comprehensive risk score and selects the optimal action combination from a predefined action set.

[0085] Based on the risk level of the target suppliers, the system automatically sets screening criteria for candidate suppliers. These criteria include multiple dimensions such as geographical location restrictions, production capacity requirements, quality standards, and price range. For candidate suppliers that pass the initial screening, a multi-objective optimization evaluation is conducted, considering cost, quality, and reliability. Cost evaluation includes factors such as procurement costs, transportation costs, and switchover costs; quality evaluation is based on indicators such as historical supply records, certifications, and customer reviews; and reliability evaluation considers factors such as financial stability, production stability, and timely delivery. Through a multi-objective optimization algorithm, the system outputs the optimal replacement solution and a detailed switchover timeline to ensure supply chain continuity.

[0086] Based on the risk level of the target entity, the corresponding inventory safety factor is calculated. A standard safety factor is used for low-risk levels, the safety factor is increased for medium-risk levels, and the safety factor is significantly increased for high-risk and very high-risk levels. Combining historical demand data and supply lead time, the optimal inventory level is calculated using an economic order quantity (EOQ) model.

[0087] The inventory adjustment strategy considers the balance between storage costs, stockout costs, and ordering costs, and uses mathematical optimization methods to determine the optimal order quantity and reorder point. The system dynamically adjusts inventory strategy parameters based on real-time monitoring of risk changes, achieving intelligent and adaptive inventory management.

[0088] The analysis identifies risk nodes in the current transportation routes, including areas prone to traffic congestion, areas frequently affected by natural disasters, and areas with strict policy regulations. A shortest path algorithm is then used to replan alternative transportation routes that avoid high-risk areas, while considering constraints such as transportation cost, transportation time, and transportation reliability.

[0089] The transportation route optimization strategy offers multiple alternatives, including the shortest time route, the lowest cost route, and the highest reliability route, allowing decision-makers to choose according to actual conditions. The system monitors changes in the risk status of each transportation route in real time and adjusts route planning and vehicle scheduling arrangements accordingly.

[0090] A 30-day sliding time window is set to collect real-time data on actual risk events occurring within the service area of ​​the Data Finance and Intelligent Connection. First-hand risk event information is obtained through data sharing mechanisms established with government departments, industry associations, and key enterprises.

[0091] The system calculates the absolute deviation between the predicted probability and the actual probability of occurrence, the deviation between the predicted time and the actual time, and the deviation between the predicted severity and the actual severity. The deviation calculation uses a weighted average method, giving higher weight to recent events to ensure the timeliness of the assessment results.

[0092] When the overall deviation exceeds a preset first threshold, the system automatically initiates a parameter optimization program. The optimization includes adjusting the time decay coefficient, spatial influence coefficient, and network weight coefficient. Parameter adjustment employs optimization algorithms such as gradient descent to determine the optimal parameter values ​​by minimizing the prediction deviation.

[0093] During parameter optimization, the system retains multiple historical parameter configurations, forming a parameter evolution trajectory. By comparing and analyzing the prediction results under different parameter configurations, the optimal parameter combination is identified, improving the accuracy of risk assessment.

[0094] If the overall deviation exceeds the second threshold, the system performs a graph structure update operation. The update includes adding newly discovered event nodes, establishing newly identified relationship edges, and adjusting the weights of existing edges. The structure update uses an incremental learning approach to avoid the computational overhead of rebuilding the entire graph.

[0095] New event nodes are automatically generated using entity recognition and event extraction algorithms, while new relationship edges are automatically established using relationship extraction algorithms. Simultaneously, the system retrains the probabilistic prediction model, using updated event graph data to train new model parameters, thereby improving the model's ability to identify emerging risk patterns.

[0096] After each model update, the system automatically performs performance validation tests. These tests use a separate test dataset to evaluate the updated model's performance in metrics such as accuracy, recall, and precision. If the validation score falls below a preset performance threshold, the system automatically rolls back to the previous stable version. All thresholds in this solution are derived from historical data and expert experience.

[0097] The version management mechanism records detailed information for each update, including update time, update content, and performance metrics. Through version comparison analysis, patterns and trends in model evolution are identified, providing a basis for subsequent optimization and improvement.

[0098] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A supply chain risk quantification assessment method based on dynamic event graphs, characterized in that, include: Collect multi-source heterogeneous data, label the multi-source heterogeneous data with spatiotemporal coordinates, and construct a four-dimensional spatiotemporal model that includes time dimension, geographic space dimension, supply chain network space dimension and risk impact space dimension, and represent supply chain events as four-dimensional spatiotemporal data; The construction of the four-dimensional spatiotemporal model includes: standardizing multi-source heterogeneous data into a unified RDF triple format; the multi-source heterogeneous data includes business registration data, judicial litigation data, tax invoice data, and logistics and transportation data; extracting timestamp information for each type of data and standardizing it into time coordinates; converting address information into latitude and longitude coordinates through geocoding services; mapping the upstream and downstream positions of entities in the supply chain to network topology coordinates; determining the risk impact radius based on the risk event type and impact range; and establishing a four-dimensional R-tree index structure to support spatiotemporal range queries to obtain four-dimensional spatiotemporal data. Based on the entity recognition algorithm, supply chain entities are identified from the four-dimensional spatiotemporal data, risk events are extracted and a process graph is constructed. The process graph uses risk events as nodes and the evolutionary relationship between events as edges, and a dynamic weight update mechanism is used to calculate the relationship weight between events. The construction of the principle graph includes: Entity recognition: Named entity recognition is performed on text content in four-dimensional spatiotemporal data to identify entity types such as suppliers, manufacturers, products, raw materials, transportation routes, and warehouse supply chains; Risk event extraction: Based on dependency parsing and semantic role labeling technology, risk events such as supply disruption, quality problems, price fluctuations and logistics delays are extracted from four-dimensional spatiotemporal data. Each risk event is represented as a four-tuple structure containing the risk source entity, risk type, affected object and severity. Relationship extraction: Identify sequential, causal, or conditional relationships between risk events through text matching and rule-based reasoning, and construct directed edges to connect related event entities; Dynamic weight calculation: The time decay weight is calculated by combining the time interval between events, the spatial influence weight is calculated based on the four-dimensional spatial distance, the structural weight is calculated based on the centrality of the entity network, and the dynamic weight value of the edge between events is obtained by combining the results. Based on the aforementioned event graph, the probability of risk transmission paths is calculated to obtain the comprehensive risk score of the target entity; Based on the comprehensive risk score, risk mitigation strategies are generated, including supplier replacement strategies, inventory adjustment strategies, and transportation route optimization strategies. By monitoring the deviation between the actual occurrence of risks and the predicted results, and updating the underlying logic graph and risk mitigation strategies through adaptive parameter optimization and incremental learning mechanisms, a self-evolving risk assessment system is formed.

2. The supply chain risk quantification assessment method based on dynamic event graphs according to claim 1, characterized in that: The steps for obtaining the comprehensive risk score include: Search the event graph for all transmission paths from the risk source event to the target entity, and record the intermediate nodes and edge weights of each path. Calculate the transmission probability of a single path by applying the conditional probability formula to the dynamic weights of all edges on the path. Handle the overlapping effects between paths by assuming conditional independence, and calculate the joint impact probability of multiple paths on the target entity. Combine five dimensions of indicators—time urgency, spatial coverage, network criticality, impact severity, and propagation speed—to calculate the comprehensive risk score of the target entity by weighted summation of the transmission probability and joint impact probability, and map it to four risk levels: low, medium, high, or extremely high.

3. The supply chain risk quantification assessment method based on dynamic event graphs according to claim 1, characterized in that: The risk mitigation strategy generation steps include: modeling suppliers, manufacturers, and distributors in the supply chain as independent agents with state space, action space, and reward function; each agent determines its own state risk level based on the current comprehensive risk score and selects the optimal action from a predefined set of actions; the supplier replacement strategy generation steps include: setting screening conditions based on the current supplier risk level, performing multi-objective optimization evaluation of candidate suppliers in terms of cost, quality, and reliability, and outputting the optimal replacement scheme and switching schedule; and the inventory adjustment strategy generation steps include: calculating the inventory safety factor based on the risk level, and using the economic order quantity model to calculate the optimal inventory level in conjunction with historical demand data and supply lead time. The steps for generating a transportation route optimization strategy are as follows: Analyze the risk nodes of the current transportation route and use the shortest path algorithm to replan alternative transportation routes that avoid high-risk areas.

4. The supply chain risk quantification assessment method based on dynamic event graphs according to claim 1, characterized in that: The updated risk assessment system for self-evolution includes: A sliding time window is set to collect real-time data on actual risk events, and the absolute deviation between the predicted probability and the actual probability, the deviation between the predicted time and the actual time, and the deviation between the predicted severity and the actual severity are calculated. When the overall deviation exceeds the first threshold, the time decay coefficient, spatial influence coefficient, and network weight coefficient are adjusted. When the overall deviation exceeds the second threshold, the event graph structure is updated, new event nodes and relationship edges are added, and the probability of risk transmission paths is recalculated. Performance verification is performed after each model update, and if the verification score is lower than the performance threshold, the system is rolled back to the previous version.

5. A supply chain risk quantification assessment system based on dynamic event graphs, characterized in that, Implementing the supply chain risk quantification assessment method based on dynamic event graph as described in claim 1 includes: The system comprises the following components: a spatiotemporal data acquisition unit, which collects multi-source heterogeneous data, annotates the multi-source heterogeneous data with spatiotemporal coordinates, constructs a four-dimensional spatiotemporal model, and represents supply chain events as four-dimensional spatiotemporal data; a causal graph construction unit, which identifies supply chain entities from the four-dimensional spatiotemporal data, extracts risk events, and constructs a causal graph; a risk assessment unit, which calculates the probability of risk transmission paths based on the causal graph and obtains a comprehensive risk score for the target entity; a risk strategy generation unit, which generates risk mitigation strategies based on the comprehensive risk score; and a risk system update unit, which monitors the deviation between the actual risk occurrence and the predicted results, updates the causal graph and risk mitigation strategies, and forms a self-evolving risk assessment system.

Images