Automobile part enterprise supply chain risk early warning method based on artificial intelligence

Abstract

The invention belongs to the technical field of automobile parts, and discloses an automobile part enterprise supply chain risk early warning method based on artificial intelligence. Comprising the steps that supply chain data are collected and processed, and a graph is constructed; evaluating the suppliers based on the atlas to generate a portrait matrix; on the basis of the portrait matrix and in combination with the production parameters, model training is performed, a prediction engine is constructed, and a part quality risk prediction result is generated; performing anomaly detection on the nodes to form a monitoring network, and generating a risk assessment result; constructing a supply chain network topology model based on the map, and performing risk propagation path analysis to generate a risk conduction map; establishing a risk assessment model, integrating the risk prediction result, the risk assessment result and the risk conduction diagram, and performing integrated assessment on the risk of each link of the supply chain to form a scoring system; based on a scoring system, a dynamic risk early warning threshold is generated, a risk response decision tree is constructed, intelligent risk response suggestions are provided, and the enterprise risk disposal efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of automotive parts, and more specifically, to an artificial intelligence-based early warning method for supply chain risks of automotive parts enterprises. Background Art

[0002] With the rapid development of the automotive industry and the acceleration of globalization, the automotive parts supply chain has become massive, complex, and cross-sector. According to statistics, a typical passenger car contains approximately 30,000 parts, involving the coordinated efforts of hundreds of suppliers. Against this backdrop, supply chain risk management has become a core operational challenge for automotive parts companies.

[0003] The automotive parts supply chain is vast and complex, involving numerous suppliers, logistics nodes, and production processes. This results in fragmented data sources and non-standard formats. Each supplier generates a massive amount of heterogeneous data daily, originating from ERP systems, quality inspection systems, logistics tracking systems, and other sources. This data is stored in diverse systems with varying formats and standards, hindering unified analysis and risk monitoring, and hindering effective integration and utilization. Traditional risk assessment methods rely heavily on manual judgment and experience, failing to scientifically quantify and dynamically assess multi-dimensional risks such as supplier qualifications, production capacity, and financial status. Furthermore, assessment results often lag behind actual risk occurrences. When a supplier experiences a risk event, such as a sudden financial crisis or a sudden drop in production capacity, traditional methods are unable to quickly assess the scope of the risk, leading to collateral impacts on multiple downstream supporting companies. In actual production operations, the lack of predictive capabilities for component quality fluctuations often leads to quality issues not being detected and addressed in a timely manner. Risk monitoring in the logistics process is also relatively weak, failing to track and issue warnings of abnormal conditions in real time. At the same time, supply chain networks are complex, with intricate relationships between nodes. Traditional methods struggle to effectively analyze the transmission paths and diffusion effects of risks within supply chain networks, often leading to chain reactions by the time risks are discovered. Furthermore, existing early warning mechanisms often suffer from single indicators, a lack of scientific basis for threshold setting, and inaccurate early warning results. Furthermore, after risks occur, they lack systematic response strategies, making it impossible to quickly develop effective risk response plans. This forces companies to passively respond to risk events, impacting the stability of production and operations. In actual production, these issues can lead to a range of serious consequences, including supply disruptions, quality incidents, and delivery delays, resulting in significant economic losses for companies.

[0004] In view of this, the present invention proposes an artificial intelligence-based supply chain risk early warning method for automotive parts enterprises to solve the above problems. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned objectives, the present invention provides the following technical solution: an artificial intelligence-based supply chain risk early warning method for automotive parts enterprises, comprising: Step 1: Collect and integrate multi-source heterogeneous data from the supply chain to build a supply chain knowledge graph; Step 2: Based on the supply chain knowledge graph, perform multi-dimensional risk feature extraction and quantitative assessment on suppliers to generate a supplier risk profile matrix; Step 3: Based on the supplier risk profile matrix and in combination with the parts production process parameters, a quality fluctuation prediction model is trained to build a parts quality risk prediction engine and generate parts quality risk prediction results; Step 4: Perform spatiotemporal feature analysis and anomaly detection on supply chain logistics nodes to form a logistics risk monitoring network and generate logistics node risk assessment results. Step 5: Based on the supply chain knowledge graph, a supply chain network topology model is constructed, and risk propagation path analysis is performed to generate a risk transmission map; Step 6: Establish a multi-agent collaborative risk assessment model, integrate the component quality risk prediction results, logistics node risk assessment results and risk transmission map, conduct an integrated assessment of the risks in each link of the supply chain, and form a comprehensive risk scoring system; Step 7: Based on the comprehensive risk scoring system and combined with multivariate time series analysis, a dynamic risk warning threshold is generated, and then a warning result is obtained; based on the warning result and the preset historical response strategy library, a risk response decision tree is constructed to provide intelligent risk response suggestions.

[0006] Furthermore, the collection and fusion processing of multi-source heterogeneous data of the supply chain to construct a supply chain knowledge graph includes: Set up a multi-source data acquisition interface to obtain raw data from the enterprise resource planning system, supplier management system, quality management system and external database to form an initial data set; Performing data cleaning and standardization on the initial data set to obtain a normalized data set; Performing entity recognition and relationship extraction on the normalized data set to construct an initial semantic network; Based on the initial semantic network, an ontology model is constructed to determine entity categories, attributes and relationship types to form a supply chain ontology framework; Mapping the normalized data set to the supply chain ontology framework to generate an instantiated knowledge base; Knowledge reasoning and consistency verification are performed on the instantiated knowledge base to obtain an optimized supply chain knowledge graph.

[0007] Furthermore, based on the supply chain knowledge graph, multi-dimensional risk feature extraction and quantitative assessment of suppliers are performed to generate a supplier risk profile matrix, including: Extracting supplier basic information, historical delivery records, quality performance, financial status, and geopolitical factors from the supply chain knowledge graph to form a supplier feature set; Performing principal component analysis on the supplier feature set to determine key risk dimensions and obtain a dimensionality-reduced feature space; Based on the dimensionality reduction feature space, a supplier risk assessment index system is constructed, and the weight of each index is determined by using the hierarchical analysis method, thereby forming a weighted index system; Quantitatively score suppliers on each risk dimension to obtain a multi-dimensional risk score vector; Cluster analysis is performed on the multidimensional risk score vector to identify supplier risk patterns, form risk category divisions, and obtain risk category information; the supplier's multidimensional risk score vector is combined with the risk category information to generate a supplier risk profile matrix.

[0008] Furthermore, based on the supplier risk profile matrix and in combination with the parts production process parameters, a quality fluctuation prediction model is trained to build a parts quality risk prediction engine and generate parts quality risk prediction results, including: Collect process parameters, equipment status, environmental conditions and test data during the parts production process to form a production process feature set; Fusing the production process feature set with the supplier risk profile matrix to construct a quality prediction training dataset; Based on the quality prediction training data set, a deep learning method is used to train a quality fluctuation prediction basic model; transfer learning is performed on the quality fluctuation prediction basic model to obtain a multimodal prediction model group; Based on the multimodal prediction model group, an integrated learning framework is constructed, and then a sustainable and self-optimizing component quality risk prediction engine is constructed. The component quality risk prediction engine is used to evaluate the current and future component quality status and generate component quality risk prediction results, including quality fluctuation probability, potential defect type, quality risk level and risk occurrence time prediction.

[0009] Furthermore, the spatiotemporal feature analysis and anomaly detection of supply chain logistics nodes are performed to form a logistics risk monitoring network and output logistics node risk assessment results, including: Collecting the location information, transportation time, transit status, and environmental condition data of logistics nodes to form a logistics spatiotemporal data set; extracting time series features from the logistics spatiotemporal data set to identify seasonal patterns and periodic changes in logistics operations and obtain a time series feature vector; Based on the time series feature vector, a logistics node time series prediction model is constructed to generate a normal logistics behavior baseline; a multivariate anomaly detection algorithm is defined to compare the deviation between the current logistics status and the normal logistics behavior baseline in real time to identify potential anomalies; the identified potential anomalies are classified and risk level assessed to form a logistics anomaly risk library; based on the logistics anomaly risk library, risk network modeling is performed on the logistics nodes to construct a logistics risk monitoring network; based on the logistics risk monitoring network, the node comprehensive risk index is calculated, and the exponential smoothing method is used to perform time series prediction on the comprehensive risk index of each node to obtain a predicted risk index; based on the predicted risk index, a set of key risk nodes is identified, and the set of key risk nodes and the corresponding predicted risk index are the logistics node risk assessment results.

[0010] Furthermore, based on the supply chain knowledge graph, a supply chain network topology model is constructed, and risk propagation path analysis is performed to generate a risk transmission map, including: Extract the association relationships among suppliers, manufacturers, and logistics service providers from the supply chain knowledge graph to construct an initial network graph; Calculate the centrality index and critical paths of network nodes to identify key nodes in the supply chain; based on the key nodes, construct a network vulnerability assessment model to quantitatively analyze the stability of the network structure; then obtain the network vulnerability assessment results; define a risk propagation dynamics model to simulate the diffusion process of risk events in the supply chain network and identify the risk cascade effect; perform a probability analysis on the propagation paths of different risk events to obtain a risk propagation probability matrix; based on the risk propagation probability matrix and the network vulnerability assessment results, construct a dynamic risk transmission map.

[0011] Furthermore, the multi-agent collaborative risk assessment model is established to integrate the component quality risk prediction results, logistics node risk assessment results and risk transmission map, and conduct an integrated assessment of the risks in each link of the supply chain to form a comprehensive risk scoring system, including: Define supplier risk assessment agents, quality risk assessment agents, logistics risk assessment agents, and market risk assessment agents, each responsible for risk assessment in different dimensions; Configure an expert knowledge base and a machine learning model for each agent; input the component quality risk prediction results into the quality risk assessment agent, input the logistics node risk assessment results into the logistics risk assessment agent, and input the risk transmission map into each agent; Build inter-agent communication protocols and collaborative decision-making mechanisms to enable different agents to exchange risk information and coordinate assessment results, forming a self-evolving comprehensive risk scoring system.

[0012] Furthermore, based on the comprehensive risk scoring system, combined with multivariate time series analysis, a dynamic risk warning threshold is generated, and then a warning result is obtained; based on the warning result and the preset historical response strategy library, a risk response decision tree is constructed to provide intelligent risk response suggestions, including: Collect historical risk event data and corresponding risk indicator time series to construct a risk-indicator correlation data set; perform time series pattern mining on the risk-indicator correlation data set to identify risk precursor characteristics and form a warning leading indicator library; Based on the early warning leading indicator library, a multi-level early warning threshold calculation model is defined to generate differentiated thresholds for different types of parts and suppliers; Constructing a Bayesian network early warning model, combined with the comprehensive risk scoring system, to obtain the risk probability and early warning level, i.e., the early warning result; Establish a supply chain risk classification system, perform structured storage of historical risk events and their response measures, and form a historical response strategy library; based on the historical response strategy library, use case-based reasoning methods to extract the association rules between risk scenarios and response measures; use the warning results, combined with the critical path analysis of the risk transmission map, to identify the current risk situation feature vector; based on the risk situation feature vector, use the C4.5 algorithm to construct a risk response decision tree, and map risk attributes to a set of response strategies; analyze the risk propagation path and impact nodes according to the risk transmission map, and prioritize the response strategy set of the risk response decision tree; form a final risk response plan; convert the risk response plan into a specific execution task, and generate intelligent risk response suggestions including action steps, responsible persons and time nodes.

[0013] An artificial intelligence-based early warning device for supply chain risks of automobile parts enterprises, the artificial intelligence-based early warning device for supply chain risks of automobile parts enterprises comprising: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory to enable the artificial intelligence-based automobile parts enterprise supply chain risk early warning device to execute the artificial intelligence-based automobile parts enterprise supply chain risk early warning method as described.

[0014] A computer-readable storage medium stores instructions, which, when executed by a processor, implement the aforementioned artificial intelligence-based supply chain risk early warning method for automotive parts enterprises.

[0015] The technical effects and advantages of the artificial intelligence-based early warning method for automobile parts enterprise supply chain risks of the present invention are as follows: This application comprehensively enhances the accuracy, foresight, and decision-making efficiency of supply chain risk management for automotive parts companies. By deeply integrating multi-dimensional data sources with a dynamic knowledge graph, it breaks through traditional information silos and achieves comprehensive insight into risks throughout the entire supply chain, from supplier qualifications to production quality fluctuations and logistics anomalies. This allows for precise identification of potential risk transmission paths and cascading effects, shortening risk warning response times. The multimodal risk assessment system, which adaptively adjusts warning thresholds based on real-time context, effectively addresses the industry's long-standing challenges of false positives and missed alerts, improves the accuracy of risk identification at key nodes, and significantly reduces production losses caused by supply chain disruptions. An intelligent decision-making support mechanism, deeply integrating historical experience with real-time trends, automatically generates a combination of response strategies tailored to the company's resource endowment, shifting risk management from reactive response to proactive prevention and control, and assisting companies in rapidly building resilient supply chain networks in complex market environments. Furthermore, through risk transmission visualization and traceability and attribution analysis, management is empowered to precisely identify the root causes of risk, optimize supplier strategic layouts and inventory strategies, and systematically reduce supply chain vulnerabilities. This not only improves the efficiency of risk management but also reduces the hidden costs associated with quality defects and logistics delays through preventive measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of the artificial intelligence-based supply chain risk early warning method for automotive parts companies of the present invention. DETAILED DESCRIPTION

[0017] The embodiments of the present application provide an artificial intelligence-based supply chain risk early warning method for automotive parts companies. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, the artificial intelligence-based supply chain risk early warning method for automobile parts enterprises, in a specific embodiment, the process of executing step 1 "collecting and integrating multi-source heterogeneous data of the supply chain to construct a supply chain knowledge graph" specifically includes the following steps: Set up a multi-source data acquisition interface to obtain raw data from the enterprise resource planning system, supplier management system, quality management system and external database to form an initial data set; Perform data cleaning and standardization on the initial data set to eliminate redundancy and outliers and obtain a normalized data set; Perform entity recognition and relationship extraction on the normalized dataset to build an initial semantic network; Based on the initial semantic network, the ontology model is constructed to determine the entity categories, attributes and relationship types to form the supply chain ontology framework; Map the normalized dataset to the supply chain ontology framework to generate an instantiated knowledge base; Knowledge reasoning and consistency verification are performed on the instantiated knowledge base to obtain the optimized supply chain knowledge graph.

[0019] Specifically, the multi-source data acquisition interface design utilizes a distributed acquisition architecture, acquiring data from various systems through REST APIs, ODBC connections, and ETL tools. Purchase orders, inventory levels, and production plan data are extracted from the enterprise resource planning system; basic supplier information, qualification certificates, and evaluation records are obtained from the supplier management system; quality inspection reports, records of non-conforming products, and quality improvement measures are extracted from the quality management system; and information such as market trends, macroeconomic indicators, and geopolitical risks is obtained from external databases. All of this data constitutes the initial dataset, providing raw material for the subsequent knowledge graph construction.

[0020] The data cleaning process uses a three-stage approach. First, missing values are handled. For data with a missing rate below 5%, mean / mode imputation is used. For data with a missing rate between 5% and 20%, KNN interpolation based on similar records is used. Features with a missing rate exceeding 20% are evaluated for importance and then retained. Next, outlier detection is performed, using the Z-score method to identify numerical anomalies and frequency analysis to identify categorical anomalies. Finally, data normalization is performed to obtain a normalized dataset; the Min-Max normalization method is used for numerical data:

[0021] ;in, represents the normalized value, Represents the original value, and represent the minimum and maximum values of the feature respectively.

[0022] A supply chain entity dictionary was constructed, containing specialized terms for categories such as suppliers, parts, production equipment, and quality indicators. A BiLSTM-CRF model was then used for text entity recognition. The model input was a text sequence T = {t1, t2, ..., tn}, and the output was a sequence of entity labels Y = {y1, y2, ..., yn}.

[0023] The BiLSTM-CRF model consists of a BiLSTM layer and a CRF layer; the BiLSTM layer encodes context information to obtain a feature vector sequence H = {h1, h2, ..., hn}, and the CRF layer calculates the probability of the optimal label sequence: ;in, represents the label transfer matrix (from Transfer to probability), represents the emission probability, is the normalization factor, is the text sequence, the entity label sequence corresponding to the text sequence, and the index in the feature vector sequence; and Belong to the entity label sequence and feature vector sequence respectively; is the corresponding sequence length; Define the relationship type set R in the supply chain domain = {supply relationship, cooperation relationship, competition relationship, geographical location relationship, ...}. Then, build a remote supervision training set based on the seed relationship instance and use the graph convolutional network model to learn the relationship between entities. For the entity pair (e1, e2), construct the feature vector by its context representation c and entity representation e1, e2. , calculate the relationship type probability: ;in, is the weight matrix, is the bias vector. By setting the relationship threshold Determine the relationship between entities: ; like There is no relationship between the corresponding entities; then the initial semantic network is constructed.

[0024] The ontology model is constructed using a combination of bottom-up and top-down approaches. Based on the identified entities and relationships (i.e., the initial semantic network), a preliminary class hierarchy is constructed. Domain expert knowledge is then incorporated to define core concepts, their attributes, and relationships. Finally, the ontology model is described using the formal language OWL (Web Ontology Language). The ontology framework defines four basic components: classes, object attributes, data attributes, and individuals, forming a conceptual framework for the supply chain domain, known as the supply chain ontology framework.

[0025] Define a mapping rule set R = {r1, r2, ..., rn}, where each rule defines the conversion relationship from the source data model to the target ontology. Then use the entity linking technology to map the identified entities to the target ontology. and the classes in the ontology Perform matching and calculate the matching degree: ;in, 、 、 is the weight coefficient and , is the string similarity, is the context similarity, is the attribute similarity, and Entity With class String, and For Entity With class The context representation of and For Entity With class Attributes.

[0026] Preset similarity threshold ,when When the entity Mapping to Class , and then obtain the instantiated knowledge base.

[0027] Knowledge reasoning uses a description logic-based reasoning mechanism and a rule-based reasoning method. Subclass reasoning, attribute inheritance, and constraint verification are performed through the ontology reasoning rule set, and specific domain knowledge is derived through the business rule set.

[0028] It should be explained that the ontology reasoning rule set is a logical reasoning rule based on the knowledge graph, which is used to verify the consistency of knowledge and deduce implicit knowledge. For example: Subclass reasoning: If A is a subclass of B, and B has attribute P, then A inherits P.

[0029] (For example, if "Brake Supplier" is a subcategory of "First-Tier Supplier" and "First-Tier Supplier" is required to be ISO certified, then "Brake Supplier" automatically inherits this requirement).

[0030] Business rule sets are industry-specific empirical rules that supplement logical reasoning. For example, "If a supplier's delivery delay rate is greater than 15% over the past three months, mark it as high risk," or "Second-tier suppliers of critical components must pass on-site audits."

[0031] Consistency testing mainly includes two parts: logical consistency testing and business rule consistency testing. Logical consistency testing uses a descriptive logic inference engine to determine whether there are any contradictions in the ontology; business rule consistency testing verifies whether the knowledge conforms to the business logic through domain constraints, thereby obtaining an optimized supply chain knowledge graph.

[0032] In a specific embodiment, the process of executing step 2, "extracting and quantitatively evaluating multi-dimensional risk characteristics of suppliers based on the supply chain knowledge graph to generate a supplier risk profile matrix," specifically includes the following steps: Extract supplier basic information, historical delivery records, quality performance, financial status, and geopolitical factors from the supply chain knowledge graph to form a supplier feature set; Conduct principal component analysis on the supplier feature set to identify key risk dimensions and obtain a reduced-dimensional feature space. Based on the reduced-dimensional feature space, construct a supplier risk assessment indicator system and use the analytic hierarchy process to determine the weight of each indicator to form a weighted indicator system. Using a fuzzy comprehensive evaluation method, we quantitatively score suppliers on each risk dimension to obtain a multidimensional risk score vector. We then perform cluster analysis on the multidimensional risk score vector to identify supplier risk patterns and form risk classifications. The supplier's multi-dimensional risk score vector is combined with the risk category information to generate a supplier risk profile matrix.

[0033] Specifically, when extracting supplier characteristics from the supply chain knowledge graph, a multi-path query and attribute aggregation method is used. For the supplier entity S, a multi-level query is constructed through the graph query language SPARQL to obtain direct attributes and related entity information. Basic information characteristics include supplier size, establishment time, main business, technical capability rating, etc.; historical delivery record characteristics include historical delivery timeliness, average delay time, order response speed, etc.; quality performance characteristics include product qualification rate, quality problem solving cycle, quality system certification level, etc.; financial status characteristics include asset-liability ratio, current ratio, accounts receivable turnover rate, etc.; geopolitical factor characteristics include political stability of the supplier's region, natural disaster risk, labor policy impact, etc. Through graph query, a supplier feature set is constructed. ,in to Indicates the value of the 1st feature to the value of the Nth feature.

[0034] The principal component analysis is performed on the supplier feature set to reduce the feature dimension and identify the key risk dimensions. The original feature space of the supplier feature set is N-dimensional. First, the feature covariance matrix C is calculated; ;in, is the number of suppliers, is the feature mean vector, The first The value of the feature, To transpose the brackets.

[0035] Then, the eigenvalues and corresponding eigenvectors of the covariance matrix C are calculated. The first k eigenvectors are selected to form the projection matrix, where k retains at least 85% of the information. The first k eigenvectors correspond to the key risk dimensions. The supplier feature set is projected into a low-dimensional space to obtain the reduced-dimensional feature vector F'; that is, in the reduced-dimensional feature space.

[0036] Based on the reduced-dimensional feature space, a hierarchical supplier risk assessment indicator system is constructed. The supplier risk assessment indicator system is divided into two levels: primary and secondary indicators. The primary indicators include five dimensions: operational risk, financial risk, quality risk, supply risk, and geopolitical risk. Each primary indicator is divided into 3-5 secondary indicators. The analytic hierarchy process (AHP) is used to determine the indicator weights for the primary and secondary indicators. First, a judgment matrix A is constructed. The element a_ii' in the judgment matrix represents the importance of indicator i relative to indicator i'. The maximum eigenvalue λmax of A and the corresponding eigenvector w are then calculated. After normalization, the indicator weights are obtained. The corresponding sum indicators are multiplied together to form a weighted indicator system.

[0037] Construct an evaluation factor set U and an evaluation level set V. Usually, the evaluation levels are {very low, low, medium, high, very high}. The evaluation factor set is a set of risk assessment dimensions, for example: U = {quality qualification rate (Q1), delivery timeliness rate (Q2), financial liquidity ratio (Q3), geopolitical stability (Q4)}. Then construct a fuzzy relationship matrix R. The element r_ll' in the fuzzy relationship matrix R represents the membership of the lth factor to the l'th evaluation level. For quantitative indicators (numerical quantifiable indicators), the fuzzy membership function is used to calculate the membership; for qualitative indicators (textual or graded indicators), the membership is determined through expert evaluation (for example, the membership is determined through the Delphi method). Then the comprehensive evaluation results are obtained. ;in," " indicates fuzzy synthesis operation, is the weight vector of the evaluation factors, indicating the relative importance of each evaluation factor; a weighted average operation (M(·,⊕)) is used. The final score S is calculated by weighting the membership degree: ;in, For the The quantitative value of the evaluation level, The comprehensive evaluation results indicate The membership of the evaluation level, The above process is repeated for each risk dimension to obtain the supplier's multidimensional risk score vector VS = [s1, s2, ..., sk], where sk represents the supplier's score on the kth risk dimension; and then the multidimensional risk score vector is obtained.

[0038] Cluster analysis is performed on the multidimensional risk score vector to identify supplier risk patterns. Using the K-means++ clustering algorithm, the optimal number of clusters, K, is first determined using the silhouette coefficient method. The K value that maximizes the silhouette coefficient is selected as the optimal number of clusters. The K-means++ algorithm is then run to cluster the results into K risk categories {C1, C2, ..., CK}, and each supplier is assigned to a risk category.

[0039] Finally, the supplier's multidimensional risk score vector is combined with the risk category information (generated through cluster analysis (K-means++)) to generate a supplier risk profile matrix. The rows of the supplier risk profile matrix represent different suppliers, while the columns include the multidimensional risk score and risk category identifier. The corresponding matrix element is M[f]=[VS, f, Cf], where VS, f is the risk score vector of the f-th supplier, and Cf is its risk category. The supplier risk profile matrix provides a comprehensive view of the supplier's risk profile, supporting subsequent risk management and decision-making.

[0040] In a specific embodiment, the process of executing step 3, "training a quality fluctuation prediction model based on the supplier risk profile matrix and in combination with component production process parameters, building a component quality risk prediction engine, and generating component quality risk prediction results," specifically includes the following steps: Collect process parameters, equipment status, environmental conditions and test data during the parts production process to form a production process feature set; The production process feature set and the supplier risk profile matrix are integrated to construct a quality prediction training dataset. Based on the quality prediction training dataset, a deep learning method is used to train a basic model for quality fluctuation prediction. Transfer learning is performed on the basic model to adapt to the quality prediction requirements of different component types, resulting in a multimodal prediction model group. Based on the multimodal prediction model group, an integrated learning framework is constructed to improve prediction accuracy through a model-weighted voting mechanism. The integrated learning framework is integrated with the real-time data processing module to build a sustainable and self-optimizing component quality risk prediction engine. The component quality risk prediction engine is used to evaluate the current and future component quality status and generate component quality risk prediction results, including quality fluctuation probability, potential defect type, quality risk level and risk occurrence time prediction.

[0041] Specifically, multi-level sensor networks and industrial Internet of Things technologies are used to collect parameters during the parts production process. Process parameters include processing temperature, pressure, rotation speed, feed rate, etc.; equipment status includes equipment operating time, vibration frequency, temperature rise value, energy consumption data, etc.; environmental conditions include workshop temperature and humidity, air quality, cleanliness, etc.; test data include dimensional deviation, surface roughness, material hardness, fatigue strength, etc. For continuously produced parts, the sliding window method is used to extract time series features. The window size w is determined according to the production cycle. For each time point u, the feature vector is F(u)=[p(u-w+1), p(u-w+2), ..., p(u)]; where p(u) represents the parameter value vector at time point u. For discretely produced parts, the batch aggregation method is used to extract features. For each batch s, the feature vector is: F(s)=[mean(p(s)),std(p(s)),min(p(s)),max(p(s)),skew(p(s))]; where p(b) represents the parameter value sequence of batch b, mean, std, min, max, and skew represent the mean, standard deviation, minimum, maximum, and skew, respectively.

[0042] Feature fusion adopts a multi-level joint embedding method. First, the production process feature F_prod and the supplier risk feature F_supp are normalized, and then the two types of features are projected into a common feature space using nonlinear mapping to obtain the common feature ; To enhance the feature expression capability while retaining the original feature information, residual connections are added to obtain enhanced features: ;in, is the weight matrix of the residual connection, Denotes vector concatenation. Construct a training dataset D = {(F_final(h), y(h))}, where F_final(h) is the h-th training sample (enhanced feature) and y(h) is the corresponding quality fluctuation label, which can be a continuous value (such as defect rate) or a discrete value (such as quality grade). This yields the quality prediction training dataset.

[0043] The basic model for quality fluctuation prediction employs a deep neural network architecture. For time series data, a Long-Term Memory (LSTM) network is used to capture temporal dependencies. For non-time series data, a Multi-Layer Perceptron (MLP) is used. The model is trained using the Adam optimizer, using the mean squared error loss function for regression tasks and the cross-entropy loss function for classification tasks. This is then integrated into the basic model for quality fluctuation prediction.

[0044] Transfer learning uses domain adaptation techniques to address distribution differences between different component types. The source domain data distribution and the target domain data distribution are defined. First, a basic quality fluctuation prediction model is trained on the source domain data, and then adapted to the target domain through the following steps: 1. Freeze the parameters of the feature extraction layer and only fine-tune the task-related layers; 2. Add a domain discriminator d to make the features generated by the feature extractor distributed consistently in the source and target domains; 3. Optimize the joint loss function: ;in, is the task-related loss (MSE or CE), is the balance parameter, is the domain discrimination loss: ;in, is the number of source domain samples, which is used to normalize the source domain loss to ensure that the contribution of the source domain is proportional to its sample number; is the number of samples in the target domain, Indicates the first samples. The first samples, f_feat is the feature extractor; The information in the brackets is processed by the domain discriminator d. By applying the above transfer learning process to different component types, a multimodal prediction model group {M_1, M_2, ..., M_G} is obtained, where M_G is the prediction model for the G-th type of component.

[0045] Build an integrated learning framework. Specifically, define the weight of each prediction model and calculate the formula: ;in, It is The error of the prediction model of the same type of parts on the validation set, is a parameter that controls the weight distribution. Based on this weight, a weighted prediction is performed, multiplying the prediction result by the weight. To further improve performance, a meta-learner is used to combine the predictions of the prediction model, which is an ensemble learning framework. The meta-learner can be a simple model such as logistic regression or random forest.

[0046] The component quality risk prediction engine architecture utilizes a microservice design, comprising five core components: data access service, feature processing service, model inference service, result feedback service, and self-optimization service. The data access service is responsible for collecting production data and supplier risk data in real time; the feature processing service performs data cleaning, feature extraction, and feature fusion; the model inference service invokes appropriate prediction models to generate quality risk prediction results; the result feedback service pushes prediction results to relevant systems and collects feedback; and the self-optimization service regularly retrains the model and updates parameters based on the discrepancy between the predicted results and actual quality performance. Self-optimization utilizes an online learning approach. For each batch of new data (X_new, Y_new), X_new represents the new input data, and Y_new represents the corresponding true label or target value.

[0047] To perform an incremental update: ;in, Indicates that at time step The model parameters of , which is a vector containing all the learnable parameters of the model; Indicates that at time step The model parameters, is the learning rate, is the gradient of the loss function; to prevent catastrophic forgetting, elastic weight integration is introduced: ;in, is a parameter The Fisher information matrix of are the original parameters, is the regularization strength, is the integrated parameter; Real-time collection of component process parameters and the latest supplier risk status in current production; data preprocessing and feature fusion through feature processing services; selection of appropriate prediction models or model combinations based on component type; and generation of multi-dimensional quality risk assessment results through model inference services: Quality fluctuation probability: indicates the probability of a component experiencing quality abnormalities within a future time window; Potential defect types: predict the specific defect types that may occur and their probability distribution; Quality risk level: Quantify the risk into three levels: high, medium, and low for decision-making reference; Risk occurrence time prediction: Estimate the time point or time range when quality problems may occur.

[0048] Pattern matching is performed between the prediction results and historical quality problem data to enhance the interpretability of the prediction; formatted component quality risk prediction results are output as key input for subsequent supply chain risk integration assessment.

[0049] In a specific embodiment, the process of executing step 4 of "performing spatiotemporal feature analysis and anomaly detection on supply chain logistics nodes, forming a logistics risk monitoring network, and outputting logistics node risk assessment results" specifically includes the following steps: Collect the location information, transportation time, transit status and environmental condition data of logistics nodes to form a logistics spatiotemporal data set; extract the time series features of the logistics spatiotemporal data set to identify the seasonal patterns and periodic changes of logistics operations and obtain the time series feature vector; Based on time series feature vectors, a logistics node time series prediction model is constructed to generate a baseline for normal logistics behavior. A multivariate anomaly detection algorithm is defined to compare the deviation between the current logistics status and the baseline for normal logistics behavior in real time to identify potential anomalies. Identified potential anomalies are classified and risk-level assessed to form a logistics anomaly risk library. Based on the logistics abnormality risk database, risk network modeling is carried out for logistics nodes, and a logistics risk monitoring network is constructed. According to the logistics risk monitoring network, the comprehensive risk index of the node is calculated. The exponential smoothing method is used to perform time series prediction on the comprehensive risk index of each node to obtain the predicted risk index. Based on the predicted risk index, the key risk node set is identified. The key risk node set and the corresponding predicted risk index are the logistics node risk assessment results.

[0050] Specifically, logistics spatiotemporal data collection utilizes multi-source sensing technology and IoT devices. Location information is obtained in real time using GPS / Beidou positioning systems to obtain the spatial coordinates of logistics vehicles and cargo. Transportation time includes departure time, arrival time, dwell time, and total transportation time. Transit status includes loading and unloading, storage, and in-transit status. Environmental conditions include temperature, humidity, vibration intensity, and air pressure. For critical components, special environmental parameters such as electromagnetic radiation and chemical exposure are also collected. The data collection frequency is dynamically adjusted based on the importance of the logistics link, with high-frequency sampling (1 minute / time) used for key nodes and low-frequency sampling (30 minutes / time) for common nodes. All data is transmitted to the cloud platform in real time via an IoT gateway, forming a logistics spatiotemporal dataset D1 = {(p_i, t_i, s_i, e_i)}, where p_i represents location information, t_i represents time information, s_i represents status information, and e_i represents environmental information.

[0051] First, data is segmented. Based on the logistics node type, time series data is divided into loading, transportation, transit, and unloading segments. Statistical features (such as mean, variance, and quantiles), frequency domain features (using fast Fourier transform to obtain spectral features), and morphological features (such as trend, seasonality, and periodicity) are then extracted for each data segment. For location data, motion features such as speed, acceleration, and dwell point distribution are calculated. For time data, delay features such as average delay time and delay standard deviation are calculated. For state data, the state transition frequency and duration distribution are extracted. For environmental data, the frequency of exceeding standards and fluctuation characteristics are calculated.

[0052] Identify seasonal patterns and cyclical changes in logistics operations: ;in, is a trend item, is the seasonal term, is the residual term, is the time index, The raw time series data of logistics operations (logistics operation indicators that vary over time, such as transportation time, arrival rate, delay frequency, and other logistics key performance indicators) are decomposed using the STL method. Spectral analysis of seasonal terms is then performed to identify the main cycles (seasonal patterns and cyclical changes). These main cycles are combined with statistical features, frequency domain features, and morphological features (vector-level concatenation) to form the time series feature vector F_time. The logistics node time series forecasting model uses an integrated approach of the Prophet algorithm and the ARIMA model. The Prophet algorithm is based on the additive model framework: ;in, is the trend function, is a seasonal function, is the holiday effect function, Is the error term. The ARIMA model is based on the integration of difference equations and determines the model parameters by minimizing the information criterion (such as AIC or BIC). The prediction results of the two models are fused by weighted averaging, and the output is then used to form the baseline of normal logistics behavior. ,in is the predicted value, is the forecast standard deviation.

[0053] Define point anomaly detection rules based on the 3σ principle; then define contextual anomaly detection, calculated using the local outlier factor (LOF) algorithm. Finally, define aggregate anomaly detection, implemented using the subsequence density distribution change detection algorithm (DPARSAD). Combined with point anomaly detection rules, contextual anomaly detection, and aggregate anomaly detection, three types of anomalies are detected. An anomaly score (the weighted sum of the three anomaly types) is calculated. When the anomaly score is greater than the preset anomaly threshold, it is marked as an anomaly point, i.e., a potential anomaly.

[0054] Define the anomaly category set C1 = {delay anomaly, path deviation, environment violation, state anomaly, ...}. Use the decision tree model to classify the detected potential anomalies: ;in, is the abnormal feature vector, f_DT is the decision tree model, Is the exception belonging to the category Then, the risk level of each anomaly is evaluated, taking into account three factors: anomaly severity S1, impact range R1, and duration D2. The risk score Ris is calculated as follows: ;in, 、 、 All anomalies are normalized to the interval [0, 1]. Based on the risk score, anomalies are divided into five risk levels: very low (0-0.2), low (0.2-0.4), medium (0.4-0.6), high (0.6-0.8), and very high (0.8-1.0). All detected anomalies, their classification, and risk level information are stored in the logistics anomaly risk database L1 = {(ai, ci, ri)}, where ai is the anomaly feature, ci is the anomaly category, and ri is the risk level.

[0055] Based on the logistics anomaly risk database, a logistics node network G1 = (V1, E1) is constructed, where V1 is the node set and E1 is the edge set. Nodes represent entities in the logistics system (such as suppliers, warehouses, distribution centers, customers, etc.), and edges represent logistics paths. Then, based on the logistics anomaly risk database, a risk index is calculated for each node v1: ;in, is the set of anomalies associated with node v1, t_a1 is the time of occurrence of anomaly a1, t_now is the current time, and λ1 is the time decay factor. Calculate the risk propagation probability P_risk(e) for each edge e=(u1, v1), where u1 and v1 are the nodes connected by edge e; ;in, 、 and is the weight coefficient and , is the historical risk level of edge e (obtained based on machine learning); is the risk index of u1.

[0056] Construct a logistics risk monitoring network G_risk=(V1, E1, R2, P2), where R2 is the node risk index mapping and P2 is the edge risk propagation probability mapping. Using visualization technology, the logistics risk monitoring network is updated and displayed in real time. Different risk levels are marked with different colors, and the risk propagation probability is represented by the thickness of the edge.

[0057] Based on the constructed logistics risk monitoring network, the node comprehensive risk index is calculated, which is the weighted sum of the node centrality index, node vulnerability index and risk index; the key risk node set is identified by the threshold method; that is, a high risk threshold is set, which is usually set to the upper quartile of the risk distribution. The node whose comprehensive risk index is greater than the high risk threshold is the key risk node.

[0058] In a specific embodiment, the process of executing step 5 of "building a supply chain network topology model based on the supply chain knowledge graph, performing risk propagation path analysis, and generating a risk transmission map" specifically includes the following steps: Extract the relationships between suppliers, manufacturers, and logistics service providers from the supply chain knowledge graph and construct an initial network graph; Using complex network analysis methods to calculate the centrality index and critical paths of network nodes and identify key nodes in the supply chain; Based on key nodes, a network vulnerability assessment model is constructed to quantitatively analyze the stability of the network structure and obtain the network vulnerability assessment results. Define a risk propagation dynamics model to simulate the diffusion process of risk events in the supply chain network and identify the risk cascade effect; Through Monte Carlo simulation, the probability analysis of the propagation paths of different risk events is carried out to obtain the risk propagation probability matrix; Based on the risk propagation probability matrix and network vulnerability assessment results, a dynamic risk transmission map is constructed to realize the visualization of risk propagation.

[0059] Specifically, supply chain network relationship extraction utilizes knowledge graph query and relationship mining techniques. Direct relationships between entities, including supply, contract, and ownership relationships, are extracted from the supply chain knowledge graph. Implicit relationships, such as geographic location, shared customers, and technological dependencies, are also mined. First-order relationships are directly retrieved through SPARQL queries.

[0060] For higher-order relationships, path analysis algorithms such as Floyd-Warshall are used to identify multi-hop paths connecting two entities. All extracted relationships form an edge set E = {(u2, v2, r2, w2)}, where u2 and v2 are entities, r2 is the relationship type, and w2 is the relationship strength. Relationship strength is calculated (weighted sum) based on relationship type, historical transaction frequency, and importance score. The weighted sum is based on the relationship type (e.g., supplier relationship = 1.0, partner relationship = 0.8), historical transaction frequency (normalized to [0, 1], e.g., 5 transactions per month → 0.7), and importance score (derived from a ranking, e.g., key supplier = 1.0).

[0061] In the initial network graph G2=(V3, E3), V3 is the entity set, including suppliers, manufacturers, logistics service providers, etc., and E3 is the extracted relationship set.

[0062] Four types of centrality metrics are calculated, including degree centrality, closeness centrality, betweenness centrality, and eigenvector centrality; degree centrality is the number of connections of a node, closeness centrality is the inverse of the average shortest path length from a node to other nodes, betweenness centrality is the frequency of a node being on the shortest path, and eigenvector centrality is a centrality measure that takes into account the importance of neighbors; Identify the critical path in a network diagram: Where P3 is the set of all paths, and Duration(p2) is the total duration of path p2. By combining centrality and critical paths, we identify a set of supply chain key nodes, K3. Specifically, we filter by defining thresholds. For example, if the centrality index is > 0.7 and the critical path is on ≥ 3 critical paths, we mark it as a critical node.

[0063] Define node deletion strategies, including random deletion, degree-first deletion, and centrality-first deletion. For each node deletion strategy, gradually remove nodes from G2 and calculate the size of the large connected component of G2, G2(p4), where p4 is the proportion of nodes retained. The percolation threshold p_c is defined as the critical point at which G2(p4) suddenly changes from 0 to a positive value. The larger the p_c, the more fragile the network. Define the network robustness index: ;in, is the number of sampling points, The number of nodes to be retained; , The larger the value, the more robust the network. For a set of key nodes in the supply chain, the vulnerability coefficient is defined as: ;in, The size of the largest connected component when all nodes are retained, The value represents the relative decrease in network connectivity after removing all key nodes. The aforementioned network robustness indicators and vulnerability coefficients are used to quantitatively analyze the stability of the network structure. Specifically, thresholds are also defined for screening. For example, if the network robustness indicator is <0.5 and the vulnerability coefficient is >0.6, the network structure is fragile. A heat map is generated to display the connectivity changes after node removal, forming the network vulnerability assessment results.

[0064] The risk propagation dynamics model uses a variant of the SIR model. Define the node state set S3 = {Healthy (H), Exposed (E), Impacted (I), Recovered (R)}. For any time t, the state of node v1 is S3_v1(t)∈S, and the state transition probability is determined by the node's own attributes and the states of its neighbors: ;in, For the moment , the state transition probability of node v1, Is the risk from the node Spread to The probability of is the set of neighbor nodes of node v1; is the indicator function, when the node When the state is affected (I) at time t, the function value is 1; otherwise, it is 0; At time t, node Status; is the product of the corresponding edge weight and the vulnerability coefficient of node v1.

[0065] By repeatedly iterating the above state transfer process, the propagation of risk in the network is simulated, and the risk cascade path and the set of affected nodes, namely the risk cascade effect, are identified.

[0066] To perform a probabilistic analysis of the propagation paths of different risk events, we first define a set of risk events. Each event in the set has an initial outbreak node set Oi and a risk characteristic parameter θi. For each risk event ei, we perform N3 independent simulations (e.g., N = 1000), recording the risk propagation path and affected nodes for each simulation. For any two nodes u1 and v1, we define the risk propagation probability as the number of times the risk propagates from u1 to v1 in N3 simulations divided by N3. We then construct a preliminary risk propagation probability matrix Ti, where the element Ti[u1, v1] represents the risk propagation probability between nodes u1 and v1. For all risk events, we calculate the risk propagation probability matrix by multiplying each risk propagation probability by its corresponding weight.

[0067] The risk transmission diagram utilizes multi-layer network visualization technology. The base network layer, G_base, consists of supply chain entities and relationships; the risk assessment layer, G_risk, includes node risk scores and edge propagation probabilities; and the temporal dynamic layer, G_time, displays the evolution of risk over time. A dynamic risk transmission diagram, G_cascade, is constructed, comprising a node set, an edge set, a node risk mapping, an edge propagation probability mapping (a risk propagation probability matrix), and a time evolution function (e.g., a Gaussian function). Through interactive visualization technology, a dynamic display of risk propagation is achieved, including: 1. Node color coding: gradient from green (low) to red (high) according to risk level; 2. Edge width encoding: adjust the thickness of the edge according to the propagation probability; 3. Node size encoding: adjust the size of the node according to its importance; 4. Timeline control: Use the sliding timeline to view the spread of risks at different time points; 5. Risk profile: shows the risk transmission path and probability distribution between any two points; Through the risk transmission map, decision makers can intuitively understand the possible paths, key nodes and transmission speeds of risk transmission, providing decision support for risk prevention and control.

[0068] In one specific embodiment, the process of executing step 6, "establishing a multi-agent collaborative risk assessment model, integrating component quality risk prediction results, logistics node risk assessment results, and risk transmission maps, performing an integrated assessment of risks in each link of the supply chain, and forming a comprehensive risk scoring system," specifically includes the following steps: Define supplier risk assessment agents, quality risk assessment agents, logistics risk assessment agents, and market risk assessment agents, each responsible for risk assessment in different dimensions; Configure expert knowledge base and machine learning model for each agent to achieve semi-supervised risk learning capabilities; Input the parts quality risk prediction results into the quality risk assessment agent, and input the logistics node risk assessment results into the logistics risk assessment agent; Input the risk transmission map into each agent to evaluate the potential impact of risk transmission on each link; Build inter-agent communication protocols and collaborative decision-making mechanisms to enable different agents to exchange risk information and coordinate assessment results, forming a self-evolving comprehensive risk scoring system.

[0069] Specifically, the multi-agent system architecture adopts a layered distributed design. The bottom layer is the data acquisition layer, which includes various data interfaces and sensor networks; the middle layer is the agent layer, which includes multiple professional agents; and the top layer is the collaborative decision-making layer, which is responsible for communication between agents and result fusion. Each agent is defined as a triple A3=(P4, K4, M4), where P4 is the perception module, K4 is the knowledge base, and M4 is the evaluation model. The supplier risk assessment agent A_sup is responsible for assessing risks such as the supplier's financial status, production capacity, and management level; the quality risk assessment agent A_qua is responsible for assessing risks such as product quality fluctuations, defect rates, and reliability; the logistics risk assessment agent A_log is responsible for assessing risks such as transportation delays, inventory shortages, and distribution anomalies; and the market risk assessment agent A_mkt is responsible for assessing risks such as demand fluctuations, price changes, and competitive situations.

[0070] Each intelligent agent has a specific mechanism for processing input data: the quality risk assessment intelligent agent A_qua receives the component quality risk prediction results generated in step 3, uses them as the main input data for assessing quality risks, and combines them with historical quality problem data to perform risk enhancement analysis; the logistics risk assessment intelligent agent A_log receives the logistics node risk assessment results produced in step 4, integrates them with historical logistics performance data, and forms a complete logistics risk view; all intelligent agents receive the risk transmission map generated in step 5, and analyze the risk propagation path in the network and the potential impact of the cascade effect on their respective areas of responsibility.

[0071] The expert knowledge base utilizes a combination of ontological reasoning and case-based reasoning. The ontology knowledge base, K_ont, uses OWL-DL to describe domain concepts and rules; the case base, K_case, stores historical risk events and response measures; and the rule base, K_rule, contains a set of if-then rules for risk assessment. The machine learning model employs semi-supervised learning to address the scarcity of labeled data. Model training utilizes federated learning technology, enabling agents to collaboratively train the model without sharing the original data. Each agent, A_i, trains a model, M_i, on local data, D_i, and then uploads the model parameters, θ_i, to the coordinator for aggregation (weighted sum). The aggregated global model parameters are distributed to each agent for the next round of training.

[0072] The application mechanism of the risk transmission map includes: Node impact analysis: Each agent analyzes the vulnerability indicators of nodes related to its responsibilities in the risk transmission diagram; Path dependency modeling: Calculate the diffusion probability and time delay of risk events based on the topological structure of the risk transmission graph; Cross-impact assessment: Analyze the mutual impact between different risk areas, such as the transmission effect of supplier risks on logistics and quality; Critical link monitoring: Identify the critical links in the risk transmission diagram and prioritize the risk status of these links.

[0073] The inter-agent communication protocol utilizes an asynchronous messaging system based on a publish / subscribe model. A message type set M5 is defined, consisting of {risk warning, status update, model parameter, collaboration request, ...}. Each message contains a sender ID, a receiver ID, a message type, content, and a timestamp. A message bus is established to implement many-to-many message routing and subscription mechanisms. To address communication delays and loss, a message confirmation and retransmission mechanism is introduced. A message history queue is maintained to support message backtracking and analysis.

[0074] The collaborative decision-making mechanism adopts a two-layer game model. The bottom layer is the internal decision-making of the intelligent agent, which is modeled using the Markov decision process (MDP): MDP=(S5, A5, P5, R5, γ5) Where S5 is the state space, A5 is the action space, P5 is the state transition probability, R5 is the reward function, and γ5 is the discount factor. The agent solves the optimal policy π* through dynamic programming.

[0075] The top layer is the collaboration between agents. An integrated voting mechanism is used to integrate the evaluation results of each agent. When disagreements arise, consensus is reached through a negotiation mechanism. 1. Each agent o submits the evaluation result r1_o and confidence c1_o; 2. Preliminary Integration Results ; 3. Calculate the divergence ; 4. If the degree of disagreement When the risk is greater than the preset threshold, a negotiation mechanism is triggered: each agent shares the evaluation basis, re-evaluates and submits the updated results, and then forms a comprehensive risk scoring system.

[0076] In a specific embodiment, the process of executing step 7, "generating dynamic risk warning thresholds based on a comprehensive risk scoring system in combination with multivariate time series analysis, and then obtaining warning results to achieve differentiated risk warnings; constructing a risk response decision tree based on the warning results and a historical response strategy library to provide intelligent risk response recommendations," specifically includes the following steps: Collect historical risk event data and corresponding risk indicator time series to construct a risk-indicator correlation dataset; conduct time series pattern mining on the risk-indicator correlation dataset to identify risk precursor characteristics and form a warning leading indicator library; Based on the early warning leading indicator library, a multi-level early warning threshold calculation model is defined to generate differentiated thresholds for different types of parts and suppliers; Construct a Bayesian network early warning model and combine it with a comprehensive risk scoring system to obtain the risk probability and warning level, i.e. the warning result; Establish a supply chain risk classification system, carry out structured storage of historical risk events and their response measures, and form a historical response strategy library; based on the historical response strategy library, use case reasoning methods to extract the association rules between risk scenarios and response measures; use early warning results, combined with the critical path analysis of the risk transmission map, to identify the current risk situation feature vector; based on the risk situation feature vector, use the C4.5 algorithm to construct a risk response decision tree, and map risk attributes to a set of response strategies; analyze the risk propagation path and impact nodes according to the risk transmission map, and prioritize the response strategy set of the risk response decision tree; form a final risk response plan; convert the risk response plan into a specific execution task, and generate intelligent risk response suggestions including action steps, responsible persons and time nodes.

[0077] Historical risk event data includes detailed information on specific risk events, such as supply chain disruptions, supplier defaults, quality issues, and logistics delays, including the time of occurrence, scope of impact, duration, and extent of loss. Risk indicator time series data includes various indicator data corresponding to these events, such as supplier financial ratios, quality compliance rates, logistics timeliness metrics, and market volatility indices. By correlating risk events with indicator data from corresponding time periods, we construct a risk-indicator correlation dataset, providing foundational data for subsequent time series pattern mining.

[0078] Time series pattern mining is performed on risk-indicator correlation datasets to identify risk precursors and form a library of early warning leading indicators. This method utilizes a long short-term memory (LSTM) network architecture, effectively capturing long-term dependencies and temporal patterns within indicator sequences—those precursors to risk. It can identify typical indicator change patterns prior to risk events, thereby identifying early warning leading indicators. These leading indicators are ranked based on their predictive power for different risk types to form a library of early warning leading indicators.

[0079] Based on the early warning leading indicator library, a multi-level early warning threshold calculation model is defined to generate differentiated thresholds for different types of parts and suppliers. The multi-level early warning threshold model uses a combination of decision trees and fuzzy logic to generate customized early warning thresholds for different combinations based on factors such as the importance level of the parts (critical parts, important parts, general parts), the risk category of the supplier (high risk, medium risk, low risk), and the criticality of the production line (main line, auxiliary line). The decision tree structure is used to divide the different hierarchical categories, and each leaf node corresponds to a fuzzy logic system, which is used to generate specific threshold parameters. The membership function of the fuzzy logic system is: ;in, is the risk indicator value, As the center point, and To control the parameters, the shape of the membership function is determined. Through the multi-level warning threshold calculation model, the system can generate differentiated warning thresholds for different types of parts and suppliers, achieving refined risk management.

[0080] A Bayesian network early warning model, combined with a comprehensive risk scoring system, enables risk probability reasoning and early warning level determination. The Bayesian network early warning model consists of a graph structure consisting of nodes (representing risk factors) and directed edges (representing causal relationships). Each node has a conditional probability table, representing the probability distribution of that node under different states of its parent node. The posterior probability of a risk event is calculated using Bayes' theorem. Through inference and calculation, the system can determine the risk probability of different risk events and determine the early warning level (normal, concern, warning, emergency) based on the risk probability value. The early warning level classification criteria are based on a comprehensive consideration of risk probability and potential impact, ensuring that the system can issue appropriate early warning levels for different risk situations.

[0081] The supply chain risk classification system utilizes a multi-tiered structure. The first-level classification includes supplier risk, quality risk, logistics risk, and market risk. The second-level classification is further refined, such as supplier risk being divided into financial risk, production capacity risk, and management risk. The third-level classification is categorized into specific risk event types, such as financial risk being divided into cash flow constraints and credit rating declines. For each risk event type, a standardized description template is established, including attribute fields such as risk code, risk description, trigger conditions, scope of impact, severity, duration, and frequency.

[0082] The historical response strategy database utilizes a storage method that combines a relational database and a knowledge graph. The relational database stores structured information, including a table of basic risk information, a table of response measures, a table of execution records, and a table of effectiveness evaluations. The knowledge graph stores complex relationships, including causal relationships between risks, matching relationships between risks and response measures, and synergistic relationships between response measures. Each response strategy record includes fields such as the strategy ID, applicable risk type, preconditions, execution steps, resource requirements, expected results, actual results, and lessons learned. Historical response strategies are sourced from internal enterprise risk response records, industry best practices, expert experience summaries, and academic research findings.

[0083] The case-based reasoning method is based on similarity measurement and knowledge transfer principles. First, a risk scenario similarity calculation framework is defined, comprehensively considering the similarity of risk type, impact scope, severity, and environmental conditions. Different similarity calculation methods are used for different attribute types: tree-based semantic distance is used for categorical attributes; normalized Euclidean distance is used for numerical attributes; and semantic vector cosine similarity is used for textual attributes. Case database retrieval uses the KNN algorithm, searching the historical response strategy library for several historical cases that are most similar to the current risk scenario.

[0084] Association rule extraction utilizes a modified version of the Apriori algorithm. The risk scenario attribute set serves as the antecedent item set, and the response measure attribute set serves as the consequent item set. Association rules such as "Under X risk scenario, Y response measure should be taken" are mined. To improve rule quality, three evaluation metrics—support, confidence, and lift—are introduced. Only rules that simultaneously meet minimum support, confidence, and lift thresholds are retained. Rule conflicts are resolved using a rule ranking mechanism based on confidence, historical success rate, and expert confirmation. This ultimately creates an association rule base, providing the rule foundation for decision tree construction.

[0085] The process of constructing a risk scenario feature vector consists of two stages: feature selection and vector generation. Feature selection uses the information gain method to extract the most discriminative attributes from the early warning results and risk transmission map. Common features include risk type, risk level, scope of impact, urgency, speed of transmission, degree of damage to key nodes, and feasibility of alternative solutions. The critical path analysis provided by the risk transmission map includes risk source node identification, risk diffusion path prediction, affected node assessment, and network vulnerability analysis. Feature vector generation uses a hybrid encoding method, using one-hot encoding for categorical features, normalization for numerical features, and ordinal encoding for sequential features. This ultimately forms a fixed-dimensional feature vector that fully represents the current risk scenario.

[0086] The C4.5 decision tree is constructed based on the principles of information entropy and information gain ratio. First, a training dataset is constructed, with samples in the form of <feature vector, response strategy label>, extracted from a historical response strategy library. The information gain ratio of each feature is then calculated, and the feature with the largest gain ratio is selected as the basis for splitting the root node. Subtrees are then recursively constructed until the termination condition is met. To handle the problem of continuous and missing values of features, the C4.5 algorithm uses the principle of minimizing information entropy to determine the optimal split point and introduces a weight distribution mechanism to handle missing values. The constructed decision tree is pruned and optimized, including pre-pruning and post-pruning. Pre-pruning limits the growth of the tree by early stopping conditions, while post-pruning removes unreliable branches through error estimation, balancing the complexity and accuracy of the model.

[0087] The decision tree mapping process propagates the input risk scenario feature vector from the root node along the decision path to the leaf nodes, obtaining the corresponding set of response strategies. For situations with incomplete feature values, a multi-path exploration method is used to make weighted decisions based on the probability distribution of different feature values. For new scenarios not covered by the decision tree, a nearest neighbor interpolation method is used to find the most similar known scenario in the feature space and draw on its response strategy.

[0088] The prioritization of response strategies takes into account multiple factors. First, the risk transmission path is analyzed using a risk transmission map, calculating the risk exposure and time urgency of each node. Second, the effectiveness of each response strategy is evaluated, including implementation cost, implementation time, expected results, and probability of success. Third, the balance between short-term response and long-term prevention is considered, taking into account the company's resource constraints and business objectives. Finally, a comprehensive scoring model is established using the Analytic Hierarchy Process (AHP) to rank the alternative response strategies and form a priority queue.

[0089] The process of developing risk response plans adheres to the principles of systematicity and synergy. First, response strategies are categorized into three categories based on the systemic impact of the risk: direct intervention, mitigation, and backup. Direct intervention targets the source of the risk, such as replacing problematic suppliers; mitigation targets the impact of the risk, such as adjusting production plans; and backup provides safeguards, such as activating alternative suppliers. Next, the different strategies are combined and optimized to resolve dependencies and conflicts between them, resulting in a coordinated response plan. Finally, the plan undergoes feasibility verification, including resource adequacy testing, time constraint verification, and risk reassessment, to ensure its feasibility and effectiveness.

[0090] Intelligent risk response recommendation generation utilizes a method that combines structured templates with natural language generation technology. First, the risk response plan is broken down into specific execution tasks. Each task includes attributes such as task description, execution steps, resource requirements, responsible and collaborating departments, start and completion time, priority, and verification criteria. Then, task leaders are automatically assigned based on the company's organizational structure and personnel responsibilities, and reasonable timelines are set. Finally, natural language generation technology is used to convert the structured task information into clear and concise text descriptions, generating a standardized risk response recommendation report. The report includes sections such as risk overview, impact assessment, response strategy, task breakdown, resource allocation, timeline, and effectiveness monitoring.

[0091] Intelligent push notifications for risk response recommendations utilize a multi-channel, tiered push strategy. Different notification methods are selected based on the urgency and importance of the risk. High-urgency risks receive instant messaging and phone notifications, while general risks receive email and system message notifications. Information is pushed at different levels based on user roles and permissions, with management-level users receiving decision-making-level information and executive-level users receiving operational-level guidance.

[0092] The closed-loop risk response management mechanism consists of three components: execution monitoring, effectiveness evaluation, and knowledge updating. Execution monitoring tracks the execution status of each response task through the task management system, promptly identifying and resolving implementation obstacles. Effectiveness evaluation comprehensively assesses the effectiveness of response measures using qualitative and quantitative indicators, including the degree of risk mitigation, resource consumption, and business impact. Knowledge updating feeds new response experience back into the historical response strategy library, enriching the case library, optimizing association rules, and improving the accuracy of the decision tree, ultimately forming a continuously improving intelligent risk management system.

[0093] The present application also provides an artificial intelligence-based automobile parts enterprise supply chain risk warning device, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the artificial intelligence-based automobile parts enterprise supply chain risk warning method in the above-mentioned embodiments.

[0094] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the computer executes the steps of the artificial intelligence-based automotive parts enterprise supply chain risk warning method.

[0095] This application comprehensively enhances the accuracy, foresight, and decision-making efficiency of supply chain risk management for automotive parts companies. By deeply integrating multi-dimensional data sources with a dynamic knowledge graph, it breaks through traditional information silos and achieves comprehensive insight into risks throughout the entire supply chain, from supplier qualifications to production quality fluctuations and logistics anomalies. This allows for precise identification of potential risk transmission paths and cascading effects, shortening risk warning response times. The multimodal risk assessment system, which adaptively adjusts warning thresholds based on real-time context, effectively addresses the industry's long-standing challenges of false positives and missed alerts, improves the accuracy of risk identification at key nodes, and significantly reduces production losses caused by supply chain disruptions. An intelligent decision-making support mechanism, deeply integrating historical experience with real-time trends, automatically generates a combination of response strategies tailored to the company's resource endowment, shifting risk management from reactive response to proactive prevention and control, and assisting companies in rapidly building resilient supply chain networks in complex market environments. Furthermore, through risk transmission visualization and traceability and attribution analysis, management is empowered to precisely identify the root causes of risk, optimize supplier strategic layouts and inventory strategies, and systematically reduce supply chain vulnerabilities. This not only improves the efficiency of risk management but also reduces the hidden costs associated with quality defects and logistics delays through preventive measures.

[0096] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0097] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program code.

[0098] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0099] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An artificial intelligence-based early warning method for supply chain risks in automotive parts companies, characterized by: include: Step 1: Collect and integrate multi-source heterogeneous data from the supply chain to build a supply chain knowledge graph; Step 2: Based on the supply chain knowledge graph, perform multi-dimensional risk feature extraction and quantitative assessment on suppliers to generate a supplier risk profile matrix; Step 3: Based on the supplier risk profile matrix and in combination with the parts production process parameters, a quality fluctuation prediction model is trained to build a parts quality risk prediction engine and generate parts quality risk prediction results; Step 4: Perform spatiotemporal feature analysis and anomaly detection on supply chain logistics nodes to form a logistics risk monitoring network and generate logistics node risk assessment results. Step 5: Based on the supply chain knowledge graph, a supply chain network topology model is constructed, and risk propagation path analysis is performed to generate a risk transmission map; Step 6: Establish a multi-agent collaborative risk assessment model, integrate the component quality risk prediction results, logistics node risk assessment results and risk transmission map, conduct an integrated assessment of the risks in each link of the supply chain, and form a comprehensive risk scoring system; Step 7: Based on the comprehensive risk scoring system and combined with multivariate time series analysis, a dynamic risk warning threshold is generated, and then a warning result is obtained; Based on the early warning results and the preset historical response strategy library, a risk response decision tree is constructed to provide intelligent risk response recommendations.

2. The artificial intelligence-based early warning method for automobile parts enterprise supply chain risks according to claim 1 is characterized in that: The collection and fusion processing of multi-source heterogeneous data of the supply chain to construct a supply chain knowledge graph includes: Set up a multi-source data acquisition interface to obtain raw data from the enterprise resource planning system, supplier management system, quality management system and external database to form an initial data set; Performing data cleaning and standardization on the initial data set to obtain a normalized data set; Performing entity recognition and relationship extraction on the normalized data set to construct an initial semantic network; Based on the initial semantic network, an ontology model is constructed to determine entity categories, attributes and relationship types to form a supply chain ontology framework; Mapping the normalized data set to the supply chain ontology framework to generate an instantiated knowledge base; Knowledge reasoning and consistency verification are performed on the instantiated knowledge base to obtain an optimized supply chain knowledge graph.

3. The artificial intelligence-based early warning method for automobile parts enterprise supply chain risks according to claim 2 is characterized in that: Based on the supply chain knowledge graph, multi-dimensional risk feature extraction and quantitative assessment of suppliers are performed to generate a supplier risk profile matrix, including: Extracting supplier basic information, historical delivery records, quality performance, financial status, and geopolitical factors from the supply chain knowledge graph to form a supplier feature set; Performing principal component analysis on the supplier feature set to determine key risk dimensions and obtain a dimensionality-reduced feature space; Based on the dimensionality reduction feature space, a supplier risk assessment index system is constructed, and the weight of each index is determined by using the hierarchical analysis method, thereby forming a weighted index system; Quantitatively score suppliers on each risk dimension to obtain a multi-dimensional risk score vector; Cluster analysis is performed on the multidimensional risk score vector to identify supplier risk patterns, form risk category divisions, and obtain risk category information; the supplier's multidimensional risk score vector is combined with the risk category information to generate a supplier risk profile matrix.

4. The artificial intelligence-based early warning method for automobile parts enterprise supply chain risks according to claim 3 is characterized in that: The quality fluctuation prediction model training is performed based on the supplier risk profile matrix and combined with the parts production process parameters, and a parts quality risk prediction engine is constructed to generate parts quality risk prediction results, including: Collect process parameters, equipment status, environmental conditions and test data during the parts production process to form a production process feature set; Fusing the production process feature set with the supplier risk profile matrix to construct a quality prediction training dataset; Based on the quality prediction training data set, a deep learning method is used to train a quality fluctuation prediction basic model; transfer learning is performed on the quality fluctuation prediction basic model to obtain a multimodal prediction model group; Based on the multimodal prediction model group, an integrated learning framework is constructed, and then a sustainable and self-optimizing component quality risk prediction engine is constructed. The component quality risk prediction engine is used to evaluate the current and future component quality status and generate component quality risk prediction results, including quality fluctuation probability, potential defect type, quality risk level and risk occurrence time prediction.

5. The artificial intelligence-based early warning method for automobile parts enterprise supply chain risks according to claim 4 is characterized in that: The spatiotemporal feature analysis and anomaly detection of supply chain logistics nodes are performed to form a logistics risk monitoring network and output logistics node risk assessment results, including: Collecting the location information, transportation time, transit status, and environmental condition data of logistics nodes to form a logistics spatiotemporal data set; extracting time series features from the logistics spatiotemporal data set to identify seasonal patterns and periodic changes in logistics operations and obtain a time series feature vector; Based on the time series feature vector, a logistics node time series prediction model is constructed to generate a normal logistics behavior baseline; a multivariate anomaly detection algorithm is defined to compare the deviation between the current logistics status and the normal logistics behavior baseline in real time to identify potential anomalies; the identified potential anomalies are classified and risk level assessed to form a logistics anomaly risk library; based on the logistics anomaly risk library, risk network modeling is performed on the logistics nodes to construct a logistics risk monitoring network; based on the logistics risk monitoring network, the node comprehensive risk index is calculated, and the exponential smoothing method is used to perform time series prediction on the comprehensive risk index of each node to obtain a predicted risk index; based on the predicted risk index, a set of key risk nodes is identified, and the set of key risk nodes and the corresponding predicted risk index are the logistics node risk assessment results.

6. The artificial intelligence-based early warning method for automobile parts enterprise supply chain risks according to claim 5 is characterized in that: The supply chain network topology model is constructed based on the supply chain knowledge graph, and risk propagation path analysis is performed to generate a risk transmission map, including: Extract the association relationships among suppliers, manufacturers, and logistics service providers from the supply chain knowledge graph to construct an initial network graph; Calculate the centrality index and critical paths of network nodes to identify key nodes in the supply chain; based on the key nodes, construct a network vulnerability assessment model to quantitatively analyze the stability of the network structure; then obtain the network vulnerability assessment results; define a risk propagation dynamics model to simulate the diffusion process of risk events in the supply chain network and identify the risk cascade effect; perform a probability analysis on the propagation paths of different risk events to obtain a risk propagation probability matrix; based on the risk propagation probability matrix and the network vulnerability assessment results, construct a dynamic risk transmission map.

7. The artificial intelligence-based early warning method for automobile parts enterprise supply chain risks according to claim 6 is characterized in that: The multi-agent collaborative risk assessment model is established to integrate the component quality risk prediction results, logistics node risk assessment results and risk transmission map, and conduct an integrated assessment of the risks in each link of the supply chain to form a comprehensive risk scoring system, including: Define supplier risk assessment agents, quality risk assessment agents, logistics risk assessment agents, and market risk assessment agents, each responsible for risk assessment in different dimensions; Configure an expert knowledge base and a machine learning model for each agent; input the component quality risk prediction results into the quality risk assessment agent, input the logistics node risk assessment results into the logistics risk assessment agent, and input the risk transmission map into each agent; Build inter-agent communication protocols and collaborative decision-making mechanisms to enable different agents to exchange risk information and coordinate assessment results, forming a self-evolving comprehensive risk scoring system.

8. The artificial intelligence-based early warning method for automobile parts enterprise supply chain risks according to claim 7 is characterized in that: Based on the comprehensive risk scoring system, combined with multivariate time series analysis, a dynamic risk warning threshold is generated, and then a warning result is obtained; Based on the early warning results and the preset historical response strategy library, a risk response decision tree is constructed to provide intelligent risk response suggestions, including: Collect historical risk event data and corresponding risk indicator time series to construct a risk-indicator correlation data set; perform time series pattern mining on the risk-indicator correlation data set to identify risk precursor characteristics and form a warning leading indicator library; Based on the early warning leading indicator library, a multi-level early warning threshold calculation model is defined to generate differentiated thresholds for different types of parts and suppliers; Constructing a Bayesian network early warning model, combined with the comprehensive risk scoring system, to obtain the risk probability and early warning level, i.e., the early warning result; Establish a supply chain risk classification system, perform structured storage of historical risk events and their response measures, and form a historical response strategy library; based on the historical response strategy library, use case-based reasoning methods to extract the association rules between risk scenarios and response measures; use the warning results, combined with the critical path analysis of the risk transmission map, to identify the current risk situation feature vector; based on the risk situation feature vector, use the C4.5 algorithm to construct a risk response decision tree, and map risk attributes to a set of response strategies; analyze the risk propagation path and impact nodes according to the risk transmission map, and prioritize the response strategy set of the risk response decision tree; form a final risk response plan; convert the risk response plan into a specific execution task, and generate intelligent risk response suggestions including action steps, responsible persons and time nodes.

9. Artificial intelligence-based early warning equipment for automotive parts supply chain risks, characterized by: The artificial intelligence-based automotive parts enterprise supply chain risk early warning device includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory so that the artificial intelligence-based automobile parts enterprise supply chain risk early warning device executes the artificial intelligence-based automobile parts enterprise supply chain risk early warning method as described in any one of claims 1-8.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the artificial intelligence-based automobile parts enterprise supply chain risk early warning method according to any one of claims 1 to 8 is implemented.

Images