A method and system for identifying supply chain disruption risks

Abstract

The application discloses a supply chain interruption risk identification method and system, belongs to the technical field of supply chain risk intelligent control, constructs an ontology knowledge base and calculates the risk correlation score of articles in the ontology knowledge base, constructs a Markov decision process, and models a supply chain interruption risk environment into a quadruple based on reinforcement learning recommendation update, wherein a state space contains an article feature vector, a candidate article set in a current article database and scores of the candidate article set; an action space contains an action of recommending or not recommending an article to a risk manager; a reward function rewards or negatively rewards according to user feedback; a state transition probability updates the ontology knowledge base based on a recommendation result to generate a new state; a Q value is calculated in combination with a current article score and a maximum expected reward of a future state; an action is selected based on the Q value; a risk manager labels recommended articles as relevant and irrelevant, and then updates the ontology knowledge base according to a feedback result.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for supply chain risks, and in particular to a method and system for identifying supply chain disruption risks. Background Technology

[0002] Current terminology databases for supply chain risk identification mainly rely on static databases. Methods based on static databases for risk identification have obvious limitations, lacking the ability to dynamically expand semantically related risk events. Traditional methods match article content with a predefined set of fixed terms, which can lead to the failure to identify emerging or indirectly related risks.

[0003] The lack of timeliness is a significant flaw in current supply chain risk identification models. This is mainly manifested in the absence of a time decay mechanism, which leads to an excessive weighting of historical news and thus interferes with real-time risk assessment. This flaw is particularly prominent in rapidly changing supply chain environments, such as policy changes under special circumstances or time-sensitive risk events like extreme weather. In these situations, the news value will decline exponentially over time. Summary of the Invention

[0004] To address the shortcomings of existing technologies and improve the breadth and timeliness of supply chain risk identification, this invention adopts the following technical solution:

[0005] A method for identifying supply chain disruption risks includes the following steps:

[0006] Step 1: Construct an ontology knowledge base;

[0007] Step 2: Calculate the risk relevance score for each article in the ontology knowledge base to obtain the relevance score;

[0008] Step 3: Recommendation update based on reinforcement learning, which includes the following steps:

[0009] Step 3.1: Construct a Markov decision process M, modeling the supply chain disruption risk environment as a quadruple M = (S, A, R, P), where S represents the state space, containing article feature vectors, the set of candidate articles in the current article database, and their scores (ns values); A represents the action space, recommending or not recommending articles to the risk manager; R represents the reward function, rewarding or disrewarding based on user feedback (√ is +1, × is -1); P represents the state transition probability, updating the ontology knowledge base based on the recommendation results to generate new states.

[0010] Step 3.2: Calculate the Q-value by combining the current article rating with the maximum expected reward for the future state;

[0011] Step 3.3: Select action based on Q value;

[0012] Step 3.4: Risk managers mark the recommended articles as relevant or irrelevant, and then update the ontology knowledge base based on the feedback results.

[0013] Furthermore, the ontology knowledge base construction in step 1 specifically includes the following steps:

[0014] Step 1.1: Knowledge Acquisition; Integrate diverse data sources, including structured data, unstructured text, and domain knowledge bases, where semantic features of unstructured text are extracted using natural language processing techniques;

[0015] The types of data required for supply chain risk identification include structured data, unstructured data, and domain knowledge bases. Unstructured data is parsed and semantically extracted using natural language processing (NLP) technology. Domain knowledge bases integrate authoritative knowledge such as industry risk standards and historical risk case databases to provide a standardized reference framework for risk classification and reasoning.

[0016] Step 1.2: Construct a heterogeneous data interface for knowledge input; structured data is accessed through API or ETL tools, and unstructured text is input through web crawlers or file upload channels. Then, data cleaning and standardization are performed to remove noisy data and unify data formats.

[0017] Step 1.3: Knowledge Extraction; Using dependency parsing and semantic extraction techniques, extract supply chain entities, risk relationships, and spatiotemporal attributes of risk events to generate triples in the form of subject, relation, and object;

[0018] Step 1.4: Ontology modeling; Modeling is performed based on the ontology structure O = (C, R, A, T, I). Class C represents the basic concepts or categories of knowledge in the supply chain risk domain, relation R represents the relationship between classes and serves as an object attribute to describe the hierarchical relationship between supply chain nodes, attribute A represents the different characteristics of the category and is a data attribute for quantifying risk characteristic indicators, object T represents the data type value of the attribute, and instance identifier I represents the content and context of each class.

[0019] Step 1.5: Knowledge Storage; Store ontology knowledge in the database as RDF triples and create a timestamp index;

[0020] Step 1.6: Dynamic Updates; Data is updated in real time through automated web crawlers and API interfaces (which enable different software components or applications to communicate and interact with each other), ensuring the dynamic updates of the knowledge base.

[0021] Furthermore, in step 1.4, class C includes the following:

[0022] Participants: refers to legal entities that participate in the supply chain;

[0023] Supplier: refers to an organization that provides products and services;

[0024] Acquirer: refers to an organization or individual that obtains products and services from other organizations; in the supply chain, the acquirer of an upstream organization is also the supplier of a downstream organization;

[0025] Procurement activities refer to the transactions between the purchaser and the supplier, which can be used to conduct business, provide products and services, and realize commercial interests.

[0026] Products: refers to the products or services provided by the supplier in the procurement process;

[0027] Risk: refers to the possibility that supply chain security threats may exploit vulnerabilities in supply chain management to cause supply chain security incidents, and the resulting impact on the organization;

[0028] Risk Levels: Different concepts are defined into different risk levels based on documents or expert standards;

[0029] Risk factors: These refer to the conditions that trigger various supply chain risks; clearly defining the types of risks and the relationships between their corresponding risk factors is crucial for effective risk identification.

[0030] Object properties R help link classes and subclasses of an ontology. Defining object properties is used to describe the association between enterprise nodes and their upstream suppliers, enabling an explicit expression of hierarchical relationships between supply chain nodes.

[0031] Data attribute A is used to capture various indicators in the supply chain, and the quantitative characterization of risk features can provide a data foundation for subsequent risk assessment and decision-making.

[0032] Furthermore, in step 2, an exponential decay function is added to the relevance score to dynamically reduce the weight of older articles, thereby improving timeliness sensitivity. Based on the frequency of risk events, risk level scores, and the time decay function, a risk relevance score for each article is generated.

[0033]

[0034] Where n represents the number of risk events of interest in the article, f i Indicates the frequency of the risk term 'i' in the article, tr i The risk level score for term i is represented by Δt, which represents the number of days between the current time and the article's publication time. Different decay rates λ are set for different types of risks, and e represents the base of the natural exponential function.

[0035] Furthermore, the calculation formula in step 3.2 is as follows:

[0036]

[0037] Among them, s t Let a represent the state at time t. t ns represents the action at time t. t This represents the current article rating. γ is a discount factor, a parameter between 0 and 1, used to control how much importance the agent places on future rewards. When γ is close to 1, the agent tends to consider long-term gains, while when γ is close to 0, the agent focuses more on immediate rewards.

[0038] Furthermore, in step 3.3, an ε-greedy strategy is adopted to select actions. Initially, the correlation between different articles and the needs of risk managers should be explored extensively. An action (exploration) is randomly selected with probability ε, and scoring and feedback data are collected to identify valuable article types.

[0039] Furthermore, in step 3.3, in the later stages, as the knowledge base feedback data accumulates, the focus shifts to using historically highly relevant articles for recommendations:

[0040]

[0041] Among them, Q * Let a represent the value function (optimal state-action value function) that maximizes the cumulative reward among all possible strategies. *t Represents the optimal value function Q * The selected action (optimal action) is indicated by the subscript *.

[0042] Furthermore, in the later stages, periodic exploration is still conducted to select the action with the highest current Q value (utilization) with a probability of 1-ε, to capture new risk information in a timely manner, update the knowledge base, and ensure that the recommendation system can adapt to environmental changes. The initial value of the ε parameter (0<=ε<=1) can be set to 0.1 to focus on exploration, and can gradually decrease to 0.05 as the number of iterations increases to focus on utilization.

[0043] Furthermore, in step 3.4, the risk level score update formula for the ontology knowledge base is as follows:

[0044]

[0045] Among them, tr i * indicates the updated risk level score, f i This indicates the frequency of the term 'i' in the article.

[0046] A supply chain disruption risk identification system includes an ontology knowledge base, a relevance scoring module, and a recommendation and update module. The ontology knowledge base, the relevance scoring module, and the recommendation and update module sequentially perform ontology knowledge base construction, relevance scoring, and reinforcement learning-based recommendation and update according to the aforementioned supply chain disruption risk identification method.

[0047] The advantages and beneficial effects of this invention are as follows:

[0048] This invention utilizes a collaborative mechanism of constructing a dynamically expanding ontology knowledge base and reinforcement learning. By leveraging structured and unstructured data within the ontology base, it defines classes and attributes, rates risk-related words, calculates article relevance scores, considers the impact of time factors on risk events, and then, based on reinforcement learning, explores first and then utilizes, recommending suitable articles and providing timely feedback. This improves the accuracy and timeliness of supply chain risk identification, achieving breakthroughs in semantic depth, timeliness, and practicality. It is applicable to various complex and ever-changing supply chain environments, providing an efficient solution for supply chain risk control. Attached Figure Description

[0049] Figure 1 This is a flowchart of the risk identification method in an embodiment of the present invention.

[0050] Figure 2 This is a diagram illustrating the knowledge base construction process in an embodiment of the present invention.

[0051] Figure 3 This is a flowchart of reinforcement learning recommendation in an embodiment of the present invention.

[0052] Figure 4 This is a schematic diagram of the risk identification system in an embodiment of the present invention. Detailed Implementation

[0053] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0054] like Figure 1 As shown, a supply chain disruption risk identification method, based on ontology knowledge base, time decay function, and reinforcement learning, includes the following steps:

[0055] Step 1: Constructing an ontology knowledge base; the ontology knowledge base is a hierarchical knowledge network dynamically generated through semantic extension technology based on standard terminology databases in the supply chain domain (such as the Cambridge Classification of Business Risks (CTBR)). The construction of the ontology knowledge base involves acquiring data and performing knowledge extraction, ontology modeling, knowledge storage, and dynamic updates, such as... Figure 2 As shown, the specific steps include the following:

[0056] Step 1.1: Knowledge Acquisition; Integrate diverse data sources, including structured data, unstructured text, and domain knowledge bases, where semantic features of unstructured text are extracted using natural language processing techniques;

[0057] The types of data required for supply chain risk identification are diverse, including structured data such as supply chain node information in enterprise ERP systems and transportation timeliness and abnormal events in logistics tracking data; unstructured data, including geopolitical conflicts implied in news texts and market demand fluctuations or brand reputation crises reflected in social media public opinion, which require the use of natural language processing (NLP) technology for text parsing and semantic extraction; and domain knowledge bases that integrate authoritative knowledge such as industry risk standards and historical risk case databases to provide a standardized reference framework for risk classification and reasoning.

[0058] Step 1.2: Knowledge Input; Construct heterogeneous data interfaces, with structured data accessed through API or ETL tools, and unstructured text input through web crawlers or file upload channels. Then, data cleaning and standardization are performed to remove noisy data and unify data formats.

[0059] Step 1.3: Knowledge Extraction; Using dependency parsing and semantic extraction techniques, extract supply chain entities, risk relationships, and spatiotemporal attributes of risk events to generate triples in the form of "<subject, relationship, object>".

[0060] The Protege plugin enables semi-automatic knowledge extraction, extracting supply chain entities such as suppliers, materials, and transportation nodes. Based on dependency parsing, it extracts risk relationships such as "supplier A depends on raw material B" and identifies risk events such as strikes and natural disasters and their spatiotemporal attributes. Finally, it constructs supply chain risk triples such as "<supplier X, limited by geopolitical risk Y>".

[0061] Step 1.4: Ontology Modeling; Modeling is performed based on the ontology structure O = (C, R, A, T, I). Classes (C) represent basic concepts or categories of knowledge in the supply chain risk domain, such as supply chain participants, products, risk levels, and risk factors; Relationships (R) represent the relationships between classes, serving as object attributes to describe the hierarchical relationships between supply chain nodes; Attributes (A) represent different characteristics of categories, considered as data attributes for quantifying risk characteristics; Target (T) represents the data type value of the attribute; Instance identifiers (I) represent the content and context of each class.

[0062] Step 1.4.1: Class (C);

[0063] Participants: refers to legal entities that participate in the supply chain, which can take various forms such as enterprises, institutions, and schools;

[0064] Suppliers: refers to organizations that provide products and services, including product suppliers, service providers, system integrators, manufacturers, distributors, agents, etc.

[0065] Acquirer: refers to an organization or individual that obtains products and services from other organizations. In the supply chain, the acquirer of an upstream organization is also the supplier of a downstream organization;

[0066] Procurement activities refer to the transactions between the purchaser and the supplier, which can be used to conduct business, provide products and services, and realize commercial interests.

[0067] Products: refers to the products or services provided by the supplier in the procurement process;

[0068] Risk: refers to the possibility that supply chain security threats may exploit vulnerabilities in supply chain management to cause supply chain security incidents, and the resulting impact on the organization;

[0069] Risk Level: Based on documents or expert standards, different concepts are defined as different risk levels;

[0070] Risk factors refer to the conditions that trigger various supply chain risks. Clearly defining the types of risks and the relationships between their corresponding risk factors is crucial for effective risk identification.

[0071] Step 1.4.2: Object Properties (R);

[0072] Object properties help link classes and subclasses of an ontology. Defining object properties describes the association between enterprise nodes and their upstream suppliers, enabling an explicit expression of hierarchical relationships between supply chain nodes.

[0073] Step 1.4.3: Data Attributes (A);

[0074] Data attributes are used to capture various indicators in the supply chain, and the quantitative characterization of risk features can provide a data foundation for subsequent risk assessment and decision-making.

[0075] Step 1.5: Knowledge Storage; Store ontology knowledge in the database as RDF triples and create a timestamp index;

[0076] The OWL-encoded ontology is converted into RDF triples and stored in an RDF database. The RDF triples are then mapped to a "node-relationship-node" format to store risk knowledge. A fast retrieval index is built based on the timestamps of risk events.

[0077] Step 1.6: Dynamic Updates; Data is updated in real time through automated web crawlers and API interfaces (which enable different software components or applications to communicate and interact with each other), ensuring the dynamic updates of the knowledge base.

[0078] Step 2: Calculate the relevance score; calculate the risk relevance score for each article in the knowledge base, and incorporate a time decay function into the relevance score. The time decay function is an exponential function used to dynamically reduce the weight of older news. This invention primarily uses the exponential decay function to improve timeliness sensitivity; based on the frequency of risk events, risk level scores, and the time decay function, generate the risk relevance score for each news article:

[0079]

[0080] Where n is the number of risk events of interest in the article, and f i It refers to the frequency of the risk term "i" in news reports, and "tr" i Δt represents the risk level score of term i, Δt represents the number of days between the current time and the time of article publication, and different decay rates λ can be set for different types of risks. For example, λ = 0.1 (rapid decay) for natural disaster news and λ = 0.05 (slow decay) for economic policy news. e is the base of the natural exponential function.

[0081] Step 3: Recommendation updates based on reinforcement learning; such as Figure 3 As shown, the specific steps include the following:

[0082] Step 3.1: Construct the Markov decision process M;

[0083] The supply chain disruption risk environment is modeled as a quadruple M = (S, A, R, P), where:

[0084] State space S: contains news feature vectors, the candidate news set in the current news database and their scores (ns values);

[0085] Action Space A: Recommend or not recommend news to risk managers;

[0086] Reward function R: Rewards or punishes users based on their feedback (√ is +1, × is -1);

[0087] State transition probability P: Update the ontology knowledge base based on the recommendation results to generate a new state;

[0088] Step 3.2: Q-value calculation;

[0089] The formula for calculating the maximum expected reward based on the current news rating and the future state is as follows:

[0090]

[0091] Among them, s t Let a represent the state at time t. t ns represents the action at time t. tThis represents the current news rating. γ is a discount factor, a parameter between 0 and 1, used to control how much importance the agent places on future rewards. When γ is close to 1, the agent tends to consider long-term gains, while when γ is close to 0, the agent focuses more on immediate rewards.

[0092] Step 3.3: Action selection; An ε-greedy strategy is used to select actions. In the early stage, random exploration is carried out with a probability of ε, and in the later stage, actions with high Q values ​​are selected with a probability of (1-ε).

[0093] When recommending risk articles to risk managers, a balance needs to be struck between exploration and utilization. Initially, the relevance of different articles to the needs of risk managers should be explored extensively. An action (exploration) should be randomly selected with probability ε to collect ratings and feedback data, identifying valuable article types. Later, as feedback data accumulates in the knowledge base, recommendations can focus on utilizing historically highly relevant articles.

[0094]

[0095] Among them, Q * Let a represent the value function (optimal state-action value function) that maximizes the cumulative reward among all possible strategies. *t Represents the optimal value function Q * The selected action (optimal action), the subscript * indicates "optimal";

[0096] However, it is still necessary to explore regularly, select the action with the highest current Q value with a probability of 1-∈ (exploitation), capture new risk information in a timely manner, update the knowledge base, and ensure that the recommendation system can adapt to environmental changes. The initial value of the ε parameter (0<=ε<=1) can be set to 0.1 to focus on exploration, and can be gradually reduced to 0.05 as the number of iterations increases to focus on exploitation.

[0097] Step 3.4: Feedback from the risk manager;

[0098] Risk managers mark recommended news items as "relevant" (√) or "irrelevant". Then, the ontology knowledge base is updated based on the feedback results:

[0099]

[0100] Among them, tr i * indicates the updated risk level rating, f i The frequency of occurrence of the term "i" in the news.

[0101] like Figure 4As shown, a supply chain disruption risk identification system includes an ontology knowledge base, a relevance scoring module, and a recommendation and update module. The ontology knowledge base, the relevance scoring module, and the recommendation and update module sequentially perform the construction of the ontology knowledge base, the relevance scoring, and the recommendation and update based on reinforcement learning according to a supply chain disruption risk identification method.

[0102] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying supply chain disruption risks, characterized in that... Includes the following steps: Step 1: Construct the ontology knowledge base; the construction of the ontology knowledge base includes the following steps: Step 1.1: Knowledge Acquisition; Integrate diverse data sources, including structured data, unstructured text, and domain knowledge bases, where semantic features of unstructured text are extracted using natural language processing techniques; The types of data required for supply chain risk identification include structured data, unstructured data, and domain knowledge bases; Step 1.2: Construct heterogeneous data interfaces for knowledge input; Step 1.3: Knowledge Extraction; Extract the spatiotemporal attributes of supply chain entities, risk relationships, and risk events to generate triplets in the form of subject, relationship, and object; Step 1.4: Ontology modeling; Modeling is based on the ontology structure O=(C,R,A,T,I). Class C represents the basic concepts or categories of knowledge in the supply chain risk domain. Relationship R represents the relationship between classes and serves as an object attribute to describe the hierarchical relationship between supply chain nodes. Attribute A represents the different characteristics of the category and is a data attribute for quantifying risk characteristic indicators. Object T represents the data type value of the attribute. Instance identifier I represents the content and context of each class. Step 1.5: Knowledge Storage; Store ontology knowledge in the database in the form of triples and create a timestamp index; Step 1.6: Dynamic update; Step 2: Calculate the risk relevance score of articles in the ontology knowledge base to obtain the relevance score; Step 3: Recommendation update based on reinforcement learning, which includes the following steps: Step 3.1: Construct a Markov decision process, modeling the supply chain disruption risk environment as a quadruple M=(S,A,R,P), where S represents the state space, containing article feature vectors, the set of candidate articles in the current article database, and their scores; A represents the action space, recommending or not recommending articles to the risk manager; R represents the reward function, rewarding or disrewarding based on user feedback; and P represents the state transition probability, updating the ontology knowledge base based on the recommendation results to generate new states. Step 3.2: Calculate the Q-value by combining the current article rating with the maximum expected reward for the future state; Step 3.3: Select action based on Q value; Step 3.4: Risk managers mark the recommended articles as relevant or irrelevant, and then update the ontology knowledge base based on the feedback results.

2. The supply chain disruption risk identification method according to claim 1, characterized in that: In step 1.4, class C includes the following: Participants: refers to legal entities that participate in the supply chain; Supplier: refers to an organization that provides products and services; Acquirer: refers to an organization or individual that obtains products and services from other organizations; in the supply chain, the acquirer of an upstream organization is also the supplier of a downstream organization; Procurement activities: refer to transactions between the purchaser and the supplier; Products: refers to the products or services provided by the supplier in the procurement process; Risk: refers to the possibility that supply chain security threats may exploit vulnerabilities in supply chain management to cause supply chain security incidents, and the resulting impact on the organization; Risk Levels: Different concepts are defined into different risk levels based on documents or expert standards; Risk factors: These refer to the conditions that trigger various supply chain risks; The object attribute R is used to describe the relationship between an enterprise node and its upstream suppliers; Data attribute A is used to capture various metrics in the supply chain.

3. The supply chain disruption risk identification method according to claim 1, characterized in that: In step 2, an exponential decay function is added to the relevance score to dynamically reduce the weight of older articles. Based on the frequency of risk events, risk level scores, and the time decay function, a risk relevance score for each article is generated. (λ>0, >=0, 0 <e -λ∆t <=1) Where, 𝑛 represents the number of risk events of interest in the article, f i Indicates the frequency of the risk term 'i' in the article, tr i The risk level score for term i is represented by Δt, which represents the number of days between the current time and the article's publication time. Different decay rates λ are set for different types of risks, and e represents the base of the natural exponential function.

4. The supply chain disruption risk identification method according to claim 1, characterized in that: In step 3.3, an ε-greedy strategy is used to select actions. Initially, the correlation between different articles and the needs of risk managers should be explored extensively. An action is randomly selected with probability ε, and scoring and feedback data are collected to identify valuable article types.

5. The supply chain disruption risk identification method according to claim 4, characterized in that: In step 3.3, later on, as knowledge base feedback data accumulates, the focus shifts to using historically highly relevant articles for recommendations: in, This represents the value function that maximizes the cumulative reward among all possible strategies. Represents the optimal value function The selected action, indicated by the index *, represents the optimal action. Indicates the state at time t. This represents the action at time t.

6. The supply chain disruption risk identification method according to claim 5, characterized in that: In the later stages, periodic exploration is still conducted, and the action with the highest current Q value is selected with a probability of 1−ε.

7. The supply chain disruption risk identification method according to claim 5, characterized in that: In step 3.4, the risk level score update formula for the ontology knowledge base is as follows: Among them, tr i * indicates the updated risk level score, f i This indicates the frequency of the term 'i' in the article.

8. A supply chain disruption risk identification system, comprising an ontology knowledge base, a relevance scoring module, and a recommendation and update module, characterized in that: The ontology knowledge base, relevance scoring module, and recommendation update module, according to any one of claims 1 to 7, in a supply chain disruption risk identification method, sequentially perform the construction of the ontology knowledge base, relevance scoring, and recommendation update based on reinforcement learning.

Images