Supply chain health assessment and risk early warning method based on graph database

Abstract

The invention discloses a supply chain health assessment and risk early warning method based on a graph database. Firstly, supply chain historical data and emergency data are integrated to construct a standardized database, and a three-layer super network model containing enterprise finance, supply relation and national trade relation is established based on the standardized database and stored in a graph database. Multi-source emergency data are collected in real time, an AIGC large model is used for predicting data changes and dynamically updating network nodes and edge scores, enterprise supply relations, financial conditions and health indexes of the national environment where the enterprise is located are calculated, and enterprise health scores and network overall scores are comprehensively generated. And performing dynamic early warning based on a preset threshold value, and displaying the network state in real time in combination with a visualization module. According to the method, the complex relation is efficiently processed through the graph database, the AIGC model dynamic prediction technology is combined, real-time monitoring, multi-dimensional risk analysis and health assessment of the supply chain are achieved, and the toughness and the anti-risk capacity of the supply chain are remarkably improved.

Description

Technical Field

[0001] The field of the present invention is the technical field of supply chain risk prediction, and specifically relates to a method for supply chain health assessment and risk early warning based on a graph database. Background Art

[0002] In the VUCA (volatility, uncertainty, complexity, and ambiguity) era, the global supply chain faces many complex challenges, including uncertain factors such as market environment changes, geopolitical conflicts, natural disasters, and the epidemic. In such a highly complex environment, supply chain risk assessment and early warning are particularly important. First, with the globalization and increasing dependence of the supply chain, the speed of risk transmission has increased significantly. The interruption of any link may have a chain reaction on the entire supply chain, resulting in the obstruction of the whole-chain operation. Second, in the fierce market competition, the stability and resilience of the supply chain not only determine the survival and development of enterprises, but also may have a profound impact on the overall national security. Especially in key areas, the supply chain crisis of core enterprises may threaten the stability of the national strategic industries. Therefore, building a sound supply chain risk management system is of great strategic significance for enterprises and even countries.

[0003] Existing patented technologies, such as "A Method, Device, Equipment, and Storage Medium for Supply Chain Risk Prediction" (Patent No.: 202311126116.5) and "An Intelligent Supply Chain Management Method Based on a Knowledge Graph" (Patent No.: 202011171965.9), mainly construct a knowledge graph of the supply chain by obtaining the historical data of supplier material supply, and mine the graph to predict risk supply chains and risk scopes. Although these technologies have made certain progress in the field of supply chain risk management, there are still the following deficiencies:

[0004] 1. Unsuitable for supply chain risk assessment and early warning in the VUCA era: The VUCA era is full of uncertainties and frequent emergencies. The knowledge graph constructed through historical data only reflects the past situation of the enterprise and cannot capture the current situation in real time. The risks calculated based on this are very inaccurate and difficult to support decision-making.

[0005] 2. Limited ability to handle complex relationships and dynamic data: Supply chain security is a multi-dimensional challenge, and its stability is deeply affected by multiple factors, such as the close cooperation between enterprises, the continuous growth of suppliers, and the trade interactions between countries. When dealing with these intertwined, complex, and evolving relationships and supply chain data, existing technologies are difficult to fully meet the stringent requirements for supply chain security management.

[0006] 3. Excessive focus on risk factors in the existing supply chain technologies: The above-mentioned patents mainly identify and predict supply chain risk factors based on knowledge graphs, but do not involve a comprehensive assessment of the overall health status of the supply chain and the supply network. The assessment of the health status of the supply network is the key foundation for preventing supply chain crises, which can help enterprises timely discover potential hidden dangers and enhance the resilience and stability of the supply chain.

[0007] 4. The existing technologies are difficult to meet the needs of the times: There are certain limitations in knowledge graphs and data mining technologies in data processing and knowledge extraction, and there is a significant gap between the overall effect and the current advanced AIGC models. Summary of the Invention

[0008] The purpose of the present invention is to provide a supply chain health assessment and risk warning method based on a graph database. The present invention can realize real-time dynamic monitoring of the supply chain, multi-dimensional risk analysis, and three-level health assessment of enterprises-networks-countries, thereby enhancing the resilience and risk resistance ability of the supply chain.

[0009] The technical solution of the present invention: A supply chain health assessment and risk warning method based on a graph database includes the following steps:

[0010] S1. Extract and integrate supply chain historical data and emergency historical data through the historical data ETL module to construct a standardized database;

[0011] S2. Construct a three-layer hypernetwork model including an enterprise financial relationship sub-network, an enterprise supply relationship sub-network, and a national trade relationship sub-network according to the standardized database, and store the structured data in the graph database;

[0012] S3. Real-time collect multi-source emergency data, extract event types and attributes through natural language processing, and generate a real-time emergency data set;

[0013] S4. Input the real-time emergency data set into the AIGC large model to predict data changes, dynamically update the node and edge scores of the hypernetwork model based on the prediction results, calculate the enterprise supply chain health indicators, and then obtain the evaluation results of the enterprise comprehensive health score and the supply chain network score based on the enterprise supply chain health indicators;

[0014] S5. Dynamically warn the evaluation results according to the preset risk threshold, and combine the visualization module to display the supply chain network status in real time.

[0015] For the above-mentioned supply chain health assessment and risk warning method based on a graph database, the step S1 specifically includes:

[0016] S11. The historical data ETL module connects to multi-source heterogeneous data through the data source interface layer, and extracts supply chain historical data and emergency historical data;

[0017] S12. Perform deduplication, missing value filling, outlier detection, and format standardization on the data;

[0018] S13. Merge the cleaned data by the primary key, align it by the time dimension, and load it into the standardized database.

[0019] In the foregoing supply chain health assessment and risk warning method based on the graph database, in step 2, the nodes of the enterprise financial relationship sub-network are enterprises and financial institutions, and the edges represent shareholding, loan, and holding relationships;

[0020] The nodes of the enterprise supply relationship sub-network are supply chain enterprises, and the edges represent product supply chains;

[0021] The nodes of the national trade relationship sub-network are national entities, and the edges represent tariff policies and trade agreements;

[0022] Among them, the hyper-edges of the enterprise financial relationship sub-network and the enterprise supply relationship sub-network are established on the same enterprises in the two-layer network, and the hyper-edges of the enterprise supply relationship sub-network and the national trade relationship sub-network are established on the enterprises and their affiliated countries.

[0023] In the foregoing supply chain health assessment and risk warning method based on the graph database, in step 2, structure the data of the nodes, edges, and hyper-edges of the hyper-network model and store it using the Neo4j graph database.

[0024] In the foregoing supply chain health assessment and risk warning method based on the graph database, the event extraction in step S3 includes:

[0025] Identify the types of emergencies based on the preset keyword combinations and syntactic pattern rules;

[0026] Obtain the event timestamp, scope of influence, and associated enterprise entities through named entity extraction technology;

[0027] Classify the emergencies into first-time data and second-time data. The former contains real-time monitoring attributes, and the latter contains the statistical results of economic impacts.

[0028] In the foregoing supply chain health assessment and risk warning method based on the graph database, the AIGC large model adopts a segmented training strategy:

[0029] In the first stage, use the first-time data to train the LSTM network to predict the immediate impact of emergencies on the supply chain;

[0030] In the second stage, fuse the second-time data, fine-tune the model parameters through incremental learning, and output the calibrated risk assessment results.

[0031] In the aforementioned supply chain health assessment and risk warning method based on a graph database, in step S4, the calculation method for the node scores of the enterprise financial relationship sub-network is as follows:

[0032] Enterprise financial health score = ω1 × stock price volatility + ω2 × market capitalization rate + ω3 × cash flow stability + ω4 × credit rating change;

[0033] Financial entity health score = ω5 × change in capital adequacy ratio + ω6 × change in non-performing loan ratio + ω7 × change in provision coverage ratio;

[0034] The calculation method for the edge scores of the enterprise financial relationship sub-network is as follows:

[0035] Financial relationship strength score = ω8 × (financial relationship amount / total enterprise assets × 100) + ω9 × (financial relationship amount / total financial entity assets × 100) + ω10 × recent financial relationship activity

[0036] The calculation method for the node scores of the enterprise supply relationship sub-network is as follows:

[0037] Enterprise supply health score = ω11 × on-time delivery rate + ω12 × change in inventory level + ω13 × change in transportation time;

[0038] The calculation method for the edge scores of the enterprise supply relationship sub-network is as follows:

[0039] Supply relationship strength score = ω14 × (recent supply volume / total enterprise demand × 100) + ω15 × (recent supply volume / total supplier supply × 100) + ω16 × recent supply relationship stability

[0040] The calculation method for the node scores of the national trade relationship sub-network is as follows:

[0041] National trade health score = ω17 × exchange rate volatility + ω18 × frequency of trade policy changes + ω19 × change in logistics time + ω20 × change in regulatory compliance.

[0042] The calculation method for the edge scores includes:

[0043] Financial relationship strength score = ω8 × (financial amount / total enterprise assets) + ω9 × (financial amount / total financial entity assets) + ω10 × trading activity;

[0044] Supply relationship strength score = ω14 × (supply volume / total enterprise demand) + ω15 × (supply volume / total supplier volume) + ω16 × supply stability.

[0045] The calculation method for the edge scores of the national trade relationship sub-network is as follows:

[0046] Strength score of trade relationship = ω21 × (Recent trade volume / Total trade volume of Country A × 100) + ω22 × (Recent trade volume / Total trade volume of Country B × 100) + ω23 × Recent stability of trade relationship

[0047] Among them, ω1 to ω23 are the weights of each dimension respectively.

[0048] For the aforementioned supply chain health assessment and risk warning method based on a graph database, the enterprise supply chain health indicators include the health score of supply relationship, the health score of the financial status of suppliers, and the health score of the status of the countries where the suppliers are located.

[0049] The health score of the supply relationship uses the average value of the scores of the edges in the enterprise supply relationship sub-network:

[0050] The health score of the financial status of the suppliers uses the average value of the scores of the nodes in the enterprise financial relationship sub-network.

[0051] The health score of the status of the countries where the suppliers are located uses the average value of the scores of the nodes in the national trade relationship sub-network.

[0052] For the aforementioned supply chain health assessment and risk warning method based on a graph database, the enterprise comprehensive health score is calculated as follows:

[0053] Enterprise comprehensive health score = ω24 × Health score of supply relationship + ω25 × Health score of the financial status of suppliers + ω26 × Health score of the status of the countries where the suppliers are located;

[0054] Among them, ω24 to ω26 are the weights of each dimension respectively.

[0055] For the aforementioned supply chain health assessment and risk warning method based on a graph database, the supply chain network score is obtained by averaging the enterprise comprehensive health scores of all enterprises.

[0056] Compared with the prior art, the present invention realizes the real-time analysis of the multi-dimensional dynamic relationships of the supply chain through the efficient relationship processing ability of the three-layer hyper-network model and the graph database; the present invention combines the phased training and multi-event serialization prediction technology of the AIGC large model, significantly improving the timeliness and accuracy of the impact assessment of emergencies. The present invention breaks through the limitations of traditional risk identification, constructs a quantitative assessment model from three levels of enterprise, network, and country, providing a scientific basis for optimizing the resilience of the supply chain. In addition, the present invention also supports multi-source data fusion and dynamic update, effectively solving the problem of data islands, strengthening the real-time monitoring and intelligent decision-making capabilities of the supply chain, and finally realizing the accurate early warning of supply chain risks and the full-link health management. Brief Description of the Drawings

[0057] Figure 1Schematic diagram of the historical data ETL module of the present invention;

[0058] Figure 2 Schematic diagram of the three - layer hyper - network model architecture of the present invention;

[0059] Figure 3 Schematic diagram of the process for obtaining and processing multi - source emergency event data of the present invention;

[0060] Figure 4 Schematic diagram of the AIGC large - model structure. Detailed implementation manners

[0061] The present invention will be further described below with reference to the drawings and embodiments, but it shall not be used as a basis for limiting the present invention.

[0062] Embodiment A supply - chain health assessment and risk warning method based on a graph database, comprising the following steps:

[0063] S1. Extract and integrate supply - chain historical data and emergency - event historical data through the historical data ETL module to construct a standardized database; in this step, the historical data ETL module connects to multi - source heterogeneous data through the data - source interface layer to extract supply - chain historical data and emergency - event historical data; among them, the supply - chain historical data includes procurement and supply data of each enterprise in the supply - chain network, as well as financial data of third parties regarding these enterprises (such as investment, shareholding, holding, and loans, etc.). The emergency - event historical data includes the type of emergency events that occurred, data information such as the time and location of occurrence at the first moment when the emergency event occurred, and data information at the second moment after the emergency event, such as the economic impact status caused by the emergency event, the affected status of relevant enterprises, and the affected status of relevant countries, etc.

[0064] S12. Perform deduplication, missing - value filling, outlier detection, and format standardization on the data; among them, the data cleaning includes:

[0065] Deduplication: Eliminate duplicate records in the data set to avoid analysis deviation. Its principle is to detect and delete duplicate entries by comparing primary keys or combined fields.

[0066] Missing - value handling: Includes substitution method and deletion method; the substitution method is to fill missing values with the mean, median, or specific default values. The deletion method is to delete data records with a relatively high proportion of missing values.

[0067] Outlier detection and handling: Use statistical methods (such as the 3 - standard - deviation rule) or machine - learning algorithms (such as Isolation Forest) to detect outliers, and decide whether to correct or eliminate them according to business logic.

[0068] Format consistency: Unify formats such as dates and currencies.

[0069] The data conversion includes:

[0070] Data type conversion: converting data fields into standard types that meet the analysis requirements (such as date type, integer type).

[0071] Data standardization: normalizing data from different sources according to a unified standard unit (such as currency unit).

[0072] Encoding and decoding: encoding (such as one-hot encoding) or decoding categorical variables to meet different analysis and storage requirements.

[0073] S13. Merge the cleaned data by the primary key, align it according to the time dimension, and then load it into the standardized database, which specifically includes:

[0074] Multi-source data merging: connecting data tables from different sources through the primary key or foreign key to achieve horizontal or vertical merging.

[0075] Data alignment: ensuring that time series data is aligned in the time dimension.

[0076] Conflict resolution: for the situation of inconsistent data sources, formulate priority rules (such as selecting the data source with the highest authority).

[0077] In this step, the structure of the historical data ETL module is as Figure 1 shown, and it includes:

[0078] 1) Data source interface layer: responsible for connecting and interacting with various data sources.

[0079] 2) Data extraction layer: extracting the required historical data from each data source.

[0080] 3) Data conversion layer: cleaning, converting, and integrating the extracted data.

[0081] 4) Data loading layer: loading the converted data into the pre-stored database and making a backup at the same time.

[0082] 5) Scheduling and monitoring layer: responsible for the scheduling, monitoring, and error handling of the ETL process.

[0083] S2. Build a three-layer hypernetwork model including the enterprise financial relationship sub-network, the enterprise supply relationship sub-network, and the national trade relationship sub-network based on the standardized database, and store the structured data in the graph database; as Figure 2 shown, the nodes of the enterprise financial relationship sub-network are enterprises and financial institutions, and the edges represent shareholding, loan, and holding relationships;

[0084] The nodes of the enterprise supply relationship sub-network are supply chain enterprises, and the edges represent product supply chains;

[0085] The nodes of the national trade relationship sub-network are national entities, and the edges represent tariff policies and trade agreements;

[0086] Among them, the hyper-edges of the enterprise financial relationship sub-network and the enterprise supply relationship sub-network are established on the same enterprises in the two-layer network, and the hyper-edges of the enterprise supply relationship sub-network and the national trade relationship sub-network are established on the enterprises and their affiliated countries.

[0087] Structurize the data of the nodes, edges and hyper-edges of the hyper-network model and store them using the Neo4j graph database for subsequent query and analysis. In addition, combine the real-time collected data to dynamically update the hyper-network to ensure that the network structure is consistent with the actual business scenario, providing a basis for real-time health assessment and risk warning.

[0088] S3. Collect multi-source emergency event data in real time, extract event types and attributes through natural language processing, and generate a real-time emergency event dataset; In this step, the collection of multi-source emergency event data uses web crawler technology to capture data from mainstream media and social networks in real time, and extract emergency events that potentially affect supply chain security. The preliminary event screening follows the principle of maximizing the recall rate to ensure that no important emergency events affecting supply chain security are missed.

[0089] As Figure 3 shown, it includes the following steps:

[0090] 3.1 Data collection;

[0091] Write a web crawler script to obtain real-time reports and event information from mainstream media websites (such as Xinhua News Agency, People's Daily Online, CNN, BBC, etc.) and social network platforms (such as Weibo, Twitter, Facebook, etc.).

[0092] Multi-source data fusion: Combine information from multiple data sources to supplement and verify event content and improve the recall rate.

[0093] Most of the existing data collection directions only focus on domestic data sources. The existing solution simultaneously monitors foreign official media (such as CNN, BBC, etc.) and social media (such as Twitter, Facebook, etc.) to ensure that domestic and foreign emergency events can be monitored in real time. At the same time, compare the reports of domestic and foreign official media and social media on the same emergency event, so as to ensure that more complete information about a certain emergency event can be obtained.

[0094] 3.2 Data preprocessing;

[0095] 3.2.1 Data cleaning;

[0096] Duplicate removal: Remove duplicate news reports and social media posts to avoid redundant data.

[0097] Denoising: Filter out content unrelated to supply chain security, such as advertisements, spam, HTML tags, special characters, stop words, etc.

[0098] 3.2.2 Data normalization;

[0099] Text normalization: Unify the encoding format of the text and perform processing such as word segmentation, part-of-speech tagging, named entity recognition (NER), etc.

[0100] Time standardization: Uniformly convert time information from different sources into a standard time format.

[0101] 3.3 Event extraction: Natural Language Processing (NLP)

[0102] Perform natural language processing on the preprocessed and normalized text data based on predefined rules (such as keyword combinations, sentence patterns, etc.), that is, extract information from the text to identify emergencies, and finally integrate the identified emergencies into a real-time emergency dataset for prediction by the AIGC large model vertical to the supply chain field.

[0103] Technical details:

[0104] 3.3.1 Rule definition;

[0105] 1. Keyword combination;

[0106] Keyword list: Define a keyword list related to emergencies, such as "accident", "interruption", "delay", "strike", "natural disaster", etc., which can be defined according to needs.

[0107] Combination rule: Combine multiple keywords to improve the accuracy of event recognition. For example, "transportation + interruption" may indicate transportation problems in the supply chain.

[0108] 2. Sentence pattern;

[0109] Syntactic pattern: Define specific syntactic structures to identify events. For example, a sentence pattern like "due to [reason], resulting in [result]" can be used to identify causal events.

[0110] Template matching: Use regular expressions or pattern matching techniques to identify sentences that match a specific pattern. For example, "[company name] announces [event type]" can be used to identify events announced by a company.

[0111] 3.3.2 Application of rules

[0112] Event Recognition: By applying keyword combinations and sentence patterns, rule definitions help identify events that may be contained in the text. For example, if the text contains "transportation route + interruption", it can be initially identified as a transportation accident.

[0113] Event Classification: Rule definitions also include the logic for event classification. For example, certain keyword combinations or patterns may correspond to specific event types, such as "weather + disaster" corresponding to natural disaster events.

[0114] 3.3.3 Information Extraction;

[0115] 1. Event Type Classification;

[0116] Rule - based Classification: After identifying an event, the first step in information extraction is to classify it according to the rule definition. For example, if the text contains "transportation route + interruption", it can be classified as a transportation accident.

[0117] Multi - label Classification: An event may belong to multiple categories, and information extraction needs to be processed according to the multi - label classification logic defined by the rules. For example, "transportation route interruption" may belong to both transportation accidents and supply chain interruption events.

[0118] 2. Event Attribute Extraction

[0119] First - time Data Information (Common Part of Real - time Monitoring and Historical Data Collection):

[0120] Time Information: Use time extraction tools (such as SUTime, HeidelTime, etc.) to extract the time when the event occurred, such as "January 15, 2025" or "next Monday".

[0121] Location Information: Use geocoding tools (such as GeoNames, OpenStreetMap, etc.) to extract the location where the event occurred, such as "a certain city", "a certain country".

[0122] Participant Information: Identify and extract the main participants in the event. For example, "a certain logistics company" can be identified as a participant in the event.

[0123] Scope of Impact: Extract words that include the scope of impact on the supply chain, such as "multiple supply chain links", "local impact", and "global scope", etc.

[0124] Second - time Data Information (Exclusive Part of Historical Data Collection):

[0125] Information data such as the economic impact status, the affected status of relevant enterprises, and the affected status of relevant countries, etc., obtained from the statistical reports of official media after the occurrence of emergencies, such as "global stock market crash", "bank failures", "rising unemployment rate", etc.

[0126] 3.3.4 Event Integration;

[0127] Real-time Emergency Incident Dataset:

[0128] For an emergency incident: Integrate the identified incident type and the first-time data information content above into a complete emergency incident data.

[0129] For an emergency incident group (multiple emergency incidents): Process each emergency incident in the emergency incident group according to the processing method of one emergency incident, and then number the obtained multiple emergency incident data to form a real-time emergency incident dataset for prediction by the AIGC large model perpendicular to the supply chain field.

[0130] Historical Emergency Incident Dataset: Composed of the incident type, the first-time data information + the second-time data information.

[0131] S4. Input the real-time emergency incident dataset into the AIGC large model to predict data changes, dynamically update the node and edge scores of the hypernetwork model based on the prediction results, calculate the enterprise supply chain health indicators, and then obtain the evaluation results of the enterprise comprehensive health score and the supply chain network score based on the enterprise supply chain health indicators;

[0132] In this step, as Figure 4 shown, the training of the AIGC large model is as follows:

[0133] (1) Training Data Preparation

[0134] Data Serialization: Serialize all the emergency incidents that occurred between two adjacent time nodes into an input data in a unified format according to the time sequence. Each incident is represented as a vector of a fixed length, containing the key information of the incident, such as the incident type, the occurrence time, the location, the impact scope, etc. If the number of incidents is insufficient, fill it with zero vectors or special markers; if the number of incidents exceeds the preset maximum length, truncate the extra incidents.

[0135] Timestamp Alignment: Ensure that the timestamps of all incidents are aligned with the time dimension of the supply chain data so that the model can better capture the relationship between the incidents and the supply chain state.

[0136] Training Dataset Construction: Combine the serialized incident data with the corresponding supply chain data to form training samples. Each sample contains an incident sequence and the supply chain state data of the corresponding time node.

[0137] (2) Phased Training

[0138] First-stage Training:

[0139] Conduct preliminary training using training samples containing first-time data. This data reflects the immediate information at the time of the emergency event.

[0140] Adopt sequence processing models (such as LSTM, Transformer) to model the event sequence, capturing the temporal relationships between events and the immediate impact of events on the supply chain.

[0141] After training is completed, save the weights and architecture of the model to provide a basis for subsequent fine-tuning or retraining.

[0142] Second-stage training:

[0143] After obtaining the second-time data (i.e., the detailed impact data caused by the emergency event), merge this data with the first-time data to form a complete training sample.

[0144] Select a fine-tuning or retraining strategy:

[0145] Fine-tuning: Load the model weights trained in the first stage, and then perform fine-tuning on the merged data. Since the emergency event data is dynamically changing and there are significant differences in the distribution between the first-stage and second-stage data, hybrid fine-tuning using Incremental Learning and Domain Adaptation is adopted. This method can retain the features learned in the first stage while further optimizing the model using the second-time data.

[0146] Retraining: Directly use the merged data to train the model from scratch. This method may achieve better performance but requires more computing resources and time.

[0147] Through segmented training, both timeliness and accuracy are taken into account, enabling preliminary prediction in the first time and more accurate calibration in the second time.

[0148] (3) Real-time prediction and dynamic update

[0149] Input data preparation: Serialize the current supply chain data and the monitored real-time emergency event dataset first-time data to form an input sequence.

[0150] Preliminary prediction: Import the serialized input data into the AIGC large model for the first-stage preliminary prediction, and give the possible impact on the supply chain network.

[0151] Risk assessment: According to the preliminary prediction results, recalculate the scores of the hypernetwork nodes and edges, and give the risk level of the preliminary prediction.

[0152] Dynamic update: After obtaining the data of the second time of the emergency event, a more accurate prediction of the second stage is carried out by combining the data of the first time and the second time. The scores of the hyper-network nodes and edges are recalculated, and a new predicted risk level is given; for the enterprise financial relationship sub-network, the nodes are enterprises and financial entities, the edges are the financial relationships between enterprises and financial entities, the node scores reflect the financial status of the enterprises, and the edge scores reflect the strength of the financial relationships. For the enterprise supply relationship sub-network, the nodes are enterprises, the edges are the supply relationships between enterprises, the node scores reflect the supply stability of the enterprises, and the edge scores reflect the strength of the supply relationships. For the national trade relationship sub-network, the nodes are countries, the edges are the trade relationships between countries, the node scores reflect the trade environment and macro stability of the countries; the edge scores reflect the strength of the trade relationships.

[0153] Among them, the calculation method of the node score of the enterprise financial relationship sub-network is:

[0154] Enterprise financial health score = ω1×Stock price volatility + ω2×Market capitalization rate + ω3×Cash flow stability + ω4×Credit rating change;

[0155] In the formula: The stock price volatility is the standard deviation of the enterprise's stock price in the past month; the market capitalization rate is the ratio of the enterprise's market value to the industry average market value; the cash flow stability is the standard deviation of the net cash flow from operating activities of the enterprise in the past three months; the credit rating change is the change value of the enterprise's credit rating in the past month; these data can all be obtained from the real-time data of the financial market and the reports of credit rating agencies.

[0156] Financial entity health score = ω5×Change in capital adequacy ratio + ω6×Change in non-performing loan ratio + ω7×Change in provision coverage ratio;

[0157] In the formula: The change in capital adequacy ratio is the change value of the financial entity's capital adequacy ratio in the past month; the change in non-performing loan ratio is the change value of the financial entity's non-performing loan ratio in the past month; the change in provision coverage ratio is the change value of the financial entity's provision coverage ratio in the past month; these data can also be obtained from the monthly reports of financial entities.

[0158] The calculation method of the edge score of the enterprise financial relationship sub-network is:

[0159] Financial relationship strength score = ω8×(Financial relationship amount / Enterprise total assets×100) + ω9×(Financial relationship amount / Financial entity total assets×100) + ω10×Financial relationship recent activity;

[0160] Wherein: The financial relationship amount is the amount of loans, investments, etc. between the enterprise and the financial entity; The recent activity of the financial relationship is the transaction frequency or amount change between the enterprise and the financial entity in the past month; In this way, the financial relationship strength score synthesizes the dynamic data of both the enterprise and the financial entity, and the calculation results are consistent.

[0161] The calculation method of the node score of the enterprise supply relationship sub-network is as follows:

[0162] Enterprise supply health score = ω11 × on-time delivery rate + ω12 × inventory level change + ω13 × transportation time change;

[0163] Wherein: The recent on-time delivery rate of the supplier is the on-time delivery rate of the supplier in the past month; The change in the supplier's inventory level is the change rate of the supplier's inventory level in the past month; The change in the supply chain transportation time is the change rate of the supply chain transportation time in the past month; These data can be obtained from the enterprise's procurement records and the supplier's monthly reports.

[0164] The calculation method of the edge score of the enterprise supply relationship sub-network is as follows:

[0165] Supply relationship strength score = ω14 × (recent supply volume / total enterprise demand × 100) + ω15 × (recent supply volume / total supplier supply volume × 100) + ω16 × recent stability of the supply relationship;

[0166] Wherein: The recent supply volume is the quantity of products or spare parts provided by the supplier to the enterprise in the past month; The total enterprise demand is the total demand for products or spare parts by the enterprise in the past month; The total supplier supply volume is the quantity of products or spare parts provided by the supplier to all enterprises in the past month; The recent stability of the supply relationship is the fluctuation of the supply volume in the past month; In this way, the supply relationship strength score synthesizes the dynamic data of both the enterprise and the supplier, and the calculation results are consistent.

[0167] The calculation method of the node score of the national trade relationship sub-network is as follows:

[0168] National trade health score = ω17 × exchange rate volatility + ω18 × frequency of trade policy changes + ω19 × change in logistics time + ω20 × change in regulatory compliance;

[0169] Wherein: The exchange rate volatility is the standard deviation of the national currency exchange rate in the past month; The frequency of trade policy changes is the number of trade policies issued by the country in the past month; The change in logistics transportation time is the change rate of the national logistics transportation time in the past month; The change in regulatory compliance is the change value of the national regulatory compliance score in the past month; These data can be obtained from the monthly reports of institutions such as the World Bank and the International Monetary Fund.

[0170] The edge score calculation method of the national trade relationship sub-network is as follows:

[0171] The trade relationship strength score = ω21 × (recent trade volume / total trade volume of country A × 100) + ω22 × (recent trade volume / total trade volume of country B × 100) + ω23 × recent stability of the trade relationship;

[0172] In the formula: The recent trade volume is the trade volume between country A and country B in the past month; the total trade volume of country A is the total trade volume of country A in the past month; the total trade volume of country B is the total trade volume of country B in the past month; the recent stability of the trade relationship is the fluctuation of the trade volume in the past month.

[0173] Among them, ω1 to ω23 are the weights of each dimension. Among them, the financial dimension (ω1 - ω10) is calculated by the EWM entropy weight method and dynamically adjusted quarterly; the supply dimension (ω11 - ω16) is determined by the AHP analytic hierarchy process and calibrated by experts annually; the country dimension (ω17 - ω23) is dynamically matched in combination with the IMF country risk index;

[0174] The enterprise supply chain health indicator includes the supply relationship health score, the supplier financial condition health score, and the condition health score of the country where the supplier is located;

[0175] The supply relationship health score uses the average value of the scores of the edges in the enterprise supply relationship sub-network:

[0176]

[0177] Among them, M is the number of edges between the enterprise and the supplier.

[0178] The supplier financial condition health score uses the average value of the scores of the nodes in the enterprise financial relationship sub-network:

[0179]

[0180] Among them, N is the number of financial entities that have financial relationships with the enterprise.

[0181] The condition health score of the country where the supplier is located uses the average value of the scores of the nodes in the national trade relationship sub-network:

[0182]

[0183] Among them, K is the number of countries where the enterprise is located.

[0184] The above three health scores are combined to obtain the comprehensive health score of the enterprise: comprehensive health score of the enterprise = ω24 × supply relationship health score + ω25 × supplier financial status health score + ω26 × supplier country status health score;

[0185] Among them, ω24 to ω26 are the weights of each dimension, which are dynamically allocated by the IMF Country Risk Index.

[0186] Averaging the overall health scores of all companies yields the supply chain network score:

[0187]

[0188] Among them, P is the number of enterprises in the supply chain network.

[0189] S5. Dynamically warn the supply chain health assessment results based on the preset risk threshold, and display the supply chain network status in real time in combination with the visualization module. In this step, for a specific enterprise, if its spare parts suppliers are numerous and spread across multiple countries, and these suppliers are financially sound, the spare parts they provide have high substitutability, and account for a low proportion of the enterprise's usage (below a certain set threshold), then the supply chain status of the enterprise is considered good. On the contrary, if the above conditions are not met, the enterprise faces certain supply chain risks. The health index of the entire supply chain network can be calculated by the health status of key enterprises in the network. When the health index of the enterprise's supply chain is lower than a certain threshold, the health index of the entire network is lower than a certain threshold, or an important node in the network has an extremely low score (damage to an important node may cause the entire network to be paralyzed), a risk warning is issued.

[0190] In summary, the solution described in the embodiment of the present invention can provide efficient management of complex supply chains and supply network security: In supply chain security management, enterprises need to deal with a series of complex and intertwined relationships, which not only cover financial cooperation between enterprises and close supply networks, but also involve a wide range of national trade dynamics. These multi-dimensional relationships have a direct impact on supply chain security, and their complexity often exceeds the processing capabilities of traditional relational databases. With its powerful capabilities, graph databases can naturally process and visualize these complex relationships, thereby significantly improving the efficiency and accuracy of data queries. This optimization not only speeds up the decision-making process, but also ensures the timeliness and effectiveness of supply chain security strategies.

[0191] Risk Warning of Empowering the Supply Chain with the AIGC Large Model of the Present Invention: To further enhance the foresight and accuracy of supply chain risk management, the AIGC large model technology is introduced to empower the risk warning system. Through deep learning algorithms, AIGC can mine and analyze potential patterns in a large amount of historical data, identifying key links that may lead to supply chain risks. The application of this technology enables the risk warning system to detect potential threats in advance and automatically generate analysis reports, providing a scientific basis for managers.

[0192] The AIGC Large Model of the Present Invention Adopts Segmented Training to Improve Real-time Performance: The present invention processes the data at the first time and the second time of emergencies separately. Preliminary predictions are made at the first time to quickly evaluate the immediate impact of the event; more accurate calibration is carried out at the second time by combining detailed data. This segmented training method not only takes into account timeliness and accuracy but also can quickly provide risk warnings at the initial stage of an emergency, buying precious time for decision-makers.

[0193] The present invention integrates multiple emergencies into input data in a unified format through a serialization method, which can effectively handle a dynamic number of events. This method enables the model to consider the combined effects of multiple emergencies simultaneously, rather than just the impact of a single event. By capturing the temporal relationships and interactions between events, the model can more accurately predict the impact of multiple emergencies on the supply chain together, enhancing the comprehensiveness and accuracy of risk assessment.

[0194] The present invention integrates the impact of emergencies to strengthen the resilience of the supply chain. In the context of globalization, the impact of emergencies (such as natural disasters, political unrest, etc.) on supply chain security cannot be ignored. Therefore, in risk warning, not only the risk management in daily operations should be concerned, but it is also particularly emphasized to take the impact of emergencies into consideration. By integrating multi-source information such as real-time news and social media and combining advanced prediction models, the specific impact of emergencies on the supply chain can be quickly evaluated, and strategies can be adjusted in a timely manner to ensure the flexibility and resilience of the supply chain.

[0195] To achieve continuous monitoring and immediate response to supply chain security, the present invention adopts an advanced stream processing framework. This framework can capture and process new information from the graph database and other data sources in real time, ensuring the real-time update and accuracy of supply chain data. Through rapid data processing and analysis, we can immediately evaluate the impact of external factors such as market changes and emergencies on the supply chain, providing managers with a near-real-time risk assessment report.

Claims

1. A supply chain health assessment and risk warning method based on graph database, characterized in that: The following steps are involved: S1. Extract and integrate supply chain historical data and emergency historical data through the historical data ETL module to build a standardized database; S2. Construct a three-layer super network model including the enterprise financial relationship sub-network, the enterprise supply relationship sub-network and the national trade relationship sub-network based on the standardized database, and store the structured data in the graph database; S3, collect multi-source emergency data in real time, extract event types and attributes through natural language processing, and generate real-time emergency data sets; S4. Input the real-time emergency data set into the AIGC big model to predict data changes, dynamically update the node and edge scores of the hypernetwork model based on the prediction results, calculate the enterprise supply chain health index, and then obtain the evaluation results of the enterprise comprehensive health score and supply chain network score based on the enterprise supply chain health index; S5. Provide dynamic warnings on the assessment results based on preset risk thresholds, and display the supply chain network status in real time in combination with the visualization module.

2. The supply chain health assessment and risk warning method based on graph database according to claim 1 is characterized in that: The step S1 specifically includes: S11, the historical data ETL module connects multi-source heterogeneous data through the data source interface layer to extract supply chain historical data and emergency historical data; S12, perform deduplication, missing value filling, outlier detection and format standardization on the data; S13. The cleaned data is merged by primary key, aligned by time dimension, and loaded into a standardized database.

3. The supply chain health assessment and risk warning method based on graph database according to claim 1 is characterized in that: In step 2, the nodes of the enterprise-finance relationship subnetwork are enterprises and financial institutions, and the edges represent shareholding, loan and controlling relationships; The nodes of the enterprise supply relationship sub-network are supply chain enterprises, and the edges represent product supply chains; The nodes of the national trade relations sub-network are national entities, and the edges represent tariff policies and trade agreements; Among them, the hyperedges of the enterprise financial relationship sub-network and the enterprise supply relationship sub-network are established on the same enterprise in the two-layer network, and the hyperedges of the enterprise supply relationship sub-network and the national trade relationship sub-network are established on the enterprise and its country.

4. The supply chain health assessment and risk warning method based on graph database according to claim 3 is characterized in that: In step 2, the data of nodes, edges and hyperedges of the hypernetwork model are structured and stored in the Neo4j graph database.

5. The supply chain health assessment and risk warning method based on graph database according to claim 1 is characterized in that: The event extraction in step S3 includes: Identify the type of emergency based on preset keyword combinations and syntactic pattern rules; Obtain event timestamps, impact scope, and associated enterprise entities through named entity extraction technology; Emergency events are divided into first-time data and second-time data, the former containing real-time monitoring attributes and the latter containing economic impact statistics.

6. The supply chain health assessment and risk warning method based on graph database according to claim 5 is characterized in that: The AIGC large model adopts a segmented training strategy: In the first phase, the LSTM network is trained using first-time data to predict the immediate impact of emergencies on the supply chain; The second stage integrates the second time data, fine-tunes the model parameters through incremental learning, and outputs the calibrated risk assessment results.

7. The supply chain health assessment and risk warning method based on graph database according to claim 1 is characterized in that: In step S4, the node scores of the enterprise financial relationship sub-network are calculated as follows: Corporate financial health score = ω1×stock price volatility + ω2×market capitalization rate + ω3×cash flow stability + ω4×credit rating change; Financial entity health score = ω5 × change in capital adequacy ratio + ω6 × change in non-performing loan ratio + ω7 × change in provision coverage ratio; The edge score calculation method of the enterprise financial relationship sub-network is: Financial relationship strength score = ω8×(financial relationship amount / enterprise total assets×100)+ω9×(financial relationship amount / financial entity total assets×100)+ω10×recent activity of financial relationship The node score calculation method of the enterprise supply relationship sub-network is: Enterprise supply health score = ω11 × delivery on-time rate + ω12 × inventory level change + ω13 × transportation time change; The edge score calculation method of the enterprise supply relationship subnetwork is: Supply relationship strength score = ω14 × (recent supply / total enterprise demand × 100) + ω15 × (recent supply / total supplier supply × 100) + ω16 × recent stability of supply relationship The node score calculation method of the national trade relationship sub-network is: National trade health score = ω17 × exchange rate volatility + ω18 × frequency of trade policy changes + ω19 × changes in logistics time + ω20 × changes in regulatory compliance. The edge score calculation method includes: Financial relationship strength score = ω8×(financial amount / total assets of the enterprise)+ω9×(financial amount / total assets of the financial entity)+ω10×trading activity; Supply relationship strength score = ω14×(supply volume / total enterprise demand) + ω15×(supply volume / total number of suppliers) + ω16×supply stability. The edge score calculation method of the national trade relationship subnetwork is: Trade relationship strength score = ω21 × (recent trade volume / country A’s total trade volume × 100) + ω22 × (recent trade volume / country B’s total trade volume × 100) + ω23 × recent stability of trade relationship Among them, ω1 to ω23 are the weights of each dimension respectively.

8. The supply chain health assessment and risk warning method based on graph database according to claim 1 is characterized in that: The enterprise supply chain health index includes the supply relationship health score, the supplier financial health score and the supplier's country health score; The supply relationship health score uses the average score of the edges in the enterprise supply relationship subnetwork: The supplier financial health score uses the average of the scores of the nodes in the enterprise financial relationship sub-network. The health score of the supplier's country is the average of the scores of the nodes in the country's trade relationship sub-network.

9. The supply chain health assessment and risk warning method based on graph database according to claim 8 is characterized in that: The enterprise comprehensive health score is calculated as follows: The enterprise comprehensive health score = ω24 × supply relationship health score + ω25 × supplier financial health score + ω26 × supplier country health score; Among them, ω24 to ω26 are the weights of each dimension respectively.

10. The supply chain health assessment and risk warning method based on graph database according to claim 9 is characterized in that: The supply chain network score is obtained by averaging the comprehensive health scores of all enterprises.

Images