Supply chain collaborative material management system and method

Abstract

The invention provides a supply chain collaborative material management system and method, and relates to the technical field of material management, and the method comprises the steps: receiving a multi-enterprise heterogeneous material data flow, and constructing an industry knowledge graph based on the field-level semantic mapping of a dynamic ontology library; a standardized data table is generated; standardized data is injected into a distributed message queue for parallel processing, dynamic cache partitions are created according to material object labels, timestamp indexes are built in the dynamic cache partitions, a sliding window mechanism is adopted to scan data, and multi-strategy resolution is performed on conflict data in a window; performing conjoint analysis on the cleaned data to generate an enhanced analysis report, and activating a case transfer learning mechanism to update a model when the similarity between new data and historical cases exceeds a threshold value; constructing a material state transfer matrix for the abnormal data, simulating a correction operation in a digital twin environment, and triggering automatic correction; the problems of multi-enterprise heterogeneous data integration, dynamic data processing, intelligent analysis and the like are solved, and the collaboration efficiency and the management level of a supply chain are improved.

Description

Technical Field

[0001] This application relates to the technical field of material management, and particularly to a supply chain collaborative material management system and method. Background Art

[0002] With the in-depth development of global economic integration and the wide application of digital technologies, supply chain management has become a core area for enterprises to enhance their competitiveness. In the modern supply chain system, material management involves collaborative operations among multiple enterprises and multiple links, facing many challenges such as strong data heterogeneity, complex information interaction, and high real-time requirements.

[0003] The traditional supply chain material management mode has obvious limitations. On the one hand, due to differences in business systems, data standards, and coding rules among different enterprises, the material data flow shows heterogeneity, making it difficult to achieve efficient integration and sharing. This data island phenomenon seriously hinders the information flow among various links of the supply chain and reduces the collaborative efficiency. For example, the data formats of suppliers, manufacturers, and logistics enterprises are not unified, making it impossible to accurately match and associate key data such as material categories, supplier information, and logistics paths, increasing the difficulty and cost of material management.

[0004] On the other hand, when dealing with the dynamically changing supply chain environment, the existing material management methods lack effective coping strategies. The entire life cycle of materials involves multiple links such as procurement, production, transportation, and warehousing, and the status and data of each link are constantly changing. Traditional methods are difficult to capture and analyze these dynamic information in real time and cannot detect potential risks and abnormal situations in a timely manner. At the same time, in terms of conflict data processing, anomaly determination, and correction, traditional methods often rely on manual experience and simple rules, lacking intelligent and automated processing capabilities, and are difficult to meet the requirements of efficient and accurate management of modern supply chains.

[0005] In addition, with the continuous expansion of the supply chain scale and the increasing complexity, higher requirements are put forward for the interpretability and decision-making support capabilities of material management. Enterprises need to clearly understand the basis and logic of material management decisions in order to make scientific and reasonable decisions. However, the existing technical means have deficiencies in providing interpretability analysis and intelligent decision-making support, and cannot provide comprehensive and in-depth material management solutions for enterprises.

[0006] Therefore, there is an urgent need for an innovative supply chain collaborative material management method that can effectively solve problems such as multi-enterprise heterogeneous data integration, dynamic data processing, intelligent analysis, and decision-making, and improve the collaborative efficiency and management level of the supply chain. Summary of the Invention

[0007] In view of the above problems, the present application proposes a supply chain collaborative material management system and method, which realizes the efficiency, accuracy, scientificity and intelligence of supply chain collaborative material management through technical means such as multi-enterprise heterogeneous data integration, dynamic data intelligent processing, precise risk control and decision-making, continuous optimization and automatic correction, and effectively improves the collaborative efficiency and management level of the supply chain.

[0008] The purpose of the present application is achieved by the following technical solutions:

[0009] In the first aspect, an embodiment of the present application provides a supply chain collaborative material management method, and the method includes:

[0010] Receiving multi-enterprise heterogeneous material data streams, constructing an industry knowledge graph including material categories, supplier relationships, and logistics paths based on field-level semantic mapping of a dynamic ontology library; and generating a standardized data table with a spatio-temporal joint index;

[0011] Injecting the standardized data into a distributed message queue for parallel processing, creating dynamic cache partitions according to material object tags and building a timestamp index, scanning the data using a sliding window mechanism, and performing multi-strategy resolution on the conflicting data within the window;

[0012] Performing joint analysis on the cleaned data, vertically constructing an LSTM-GNN hybrid model to capture the full life cycle characteristics of materials, horizontally embedding an interpretable business rule engine to define constraint conditions, and quantifying the contribution of each rule to anomaly determination;

[0013] Generating an enhanced analysis report, and activating a case transfer learning mechanism to update the model when the similarity between new data and historical cases exceeds a threshold;

[0014] Constructing a material state transition matrix for abnormal data, simulating correction operations in a digital twin environment and triggering automatic correction.

[0015] In the second aspect, the present application provides a supply chain collaborative material management system for the supply chain collaborative material management method described in the embodiments of the present application, and the system includes:

[0016] A data standardization module for receiving multi-enterprise heterogeneous material data streams, constructing an industry knowledge graph including material categories, supplier relationships, and logistics paths based on field-level semantic mapping of a dynamic ontology library; and generating a standardized data table with a spatio-temporal joint index;

[0017] A data storage resolution module for injecting the standardized data into a distributed message queue for parallel processing, creating dynamic cache partitions according to material object tags and building a timestamp index, scanning the data using a sliding window mechanism, and performing multi-strategy resolution on the conflicting data within the window;

[0018] A joint analysis module is used to perform joint analysis on the cleaned data, vertically construct an LSTM-GNN hybrid model to capture the characteristics of the entire life cycle of materials, horizontally embed an interpretable business rule engine to define constraints, and quantify the contribution of each rule to anomaly determination;

[0019] A report generation module is used to generate an enhanced analysis report, and activate the case transfer learning mechanism to update the model when the similarity between new data and historical cases exceeds the threshold;

[0020] An anomaly handling module is used to construct a material state transition matrix for abnormal data, simulate correction operations in a digital twin environment, and trigger automatic correction.

[0021] The beneficial effects of the present invention include: through the field-level semantic mapping of the dynamic ontology library, it can receive heterogeneous material data streams from multiple enterprises, establish mapping relationships for data with the same semantics but different field names, and use the ontology reasoner for consistency checking and conflict resolution, thereby constructing an industry knowledge graph and a standardized data table, breaking the data barriers between enterprises, realizing the efficient integration and sharing of multi-source heterogeneous data, providing a unified and accurate data basis for subsequent material management, and improving the efficiency of information interaction in each link of the supply chain; injecting standardized data into a distributed message queue for parallel processing, creating a dynamic cache partition according to the material object label and building a timestamp index, using a sliding window mechanism to scan the data, and performing multi-strategy resolution on the conflicting data within the window, which can quickly and accurately process the dynamically changing data in the supply chain, effectively solve the data conflict problem, and ensure the accuracy and consistency of the data; constructing a multi-dimensional evaluation system for material characteristics, determining the weights of each index through principal component analysis and calculating the comprehensive material characteristic value, based on which entering the three-level partition decision logic, designing a differentiated index enhancement strategy for partitions with different risk levels, and realizing the precise control of material risks; at the same time, vertically constructing an LSTM-GNN hybrid model to capture the characteristics of the entire life cycle of materials, horizontally embedding an interpretable business rule engine to define constraints and quantify the rule contribution degree, tracing and verifying when the determination results of the deep learning model and the rule engine conflict, and generating an accurate anomaly determination conclusion, providing scientific and precise decision support for material management. When the similarity between new data and historical cases exceeds the threshold, activate the case transfer learning mechanism to update the model, so that the system can continuously adapt to the changes in the supply chain environment and continuously optimize the management strategy; construct a material state transition matrix for abnormal data, simulate correction operations in a digital twin environment, and automatically execute correction operations and synchronously update the status of all network nodes when the preset conditions are met, realizing the automation and intelligence of material management, and improving the emergency response ability and management efficiency of the supply chain. Description of the Drawings

[0022] Figure 1 It is a schematic diagram of a supply chain collaborative material management method provided by an embodiment of the present application. Specific implementation manners

[0023] Next, in combination with the accompanying drawings and specific implementation manners, the present application will be further described. It should be noted that, on the premise of no conflict, the following-described embodiments or technical features can be arbitrarily combined to form new embodiments with each other.

[0024] Refer to Figure 1 , an embodiment of the present application provides a supply chain collaborative material management method, and the method includes:

[0025] Receiving heterogeneous material data streams of multiple enterprises, and based on field-level semantic mapping of a dynamic ontology library, constructing an industry knowledge graph including material categories, supplier relationships, and logistics paths; and generating a standardized data table with a spatio-temporal joint index;

[0026] Injecting the standardized data into a distributed message queue for parallel processing, creating a dynamic cache partition according to the material object label and building a timestamp index therein, scanning the data by adopting a sliding window mechanism, and performing multi-strategy resolution on the conflicting data within the window;

[0027] Performing joint analysis on the cleaned data, vertically constructing an LSTM-GNN hybrid model to capture the full life cycle characteristics of materials, horizontally embedding an interpretable business rule engine to define constraint conditions, and quantifying the contribution degree of each rule to anomaly determination;

[0028] Generating an enhanced analysis report, and activating a case transfer learning mechanism to update the model when the similarity between the new data and the historical cases exceeds the threshold;

[0029] Constructing a material state transition matrix for the abnormal data, simulating a correction operation in a digital twin environment and triggering an automatic correction.

[0030] The working principle and beneficial effects of the above technical solutions are as follows:

[0031] In a supply chain scenario, the data formats and semantics of different enterprises are different, forming heterogeneous material data streams; first, these data are received, and field-level semantic mapping is performed by using a dynamic ontology library. The dynamic ontology library contains rich semantic information and mapping rules, which can identify fields with the same or similar meanings in the data of different enterprises and uniformly convert them into standard semantic expressions. Based on this, an industry knowledge graph is constructed, which integrates key information such as material categories, supplier relationships, and logistics paths, clearly presenting the associations between various elements of the supply chain in a graphical manner, providing a structured knowledge basis for subsequent analysis; at the same time, a standardized data table with a spatio-temporal joint index is generated, and the data is organized through the spatio-temporal index, facilitating quick positioning and retrieval of material data within a specific time and space range, and realizing the orderly storage and efficient management of the data.

[0032] Inject the standardized data into the distributed message queue, and utilize the parallel computing power of the distributed system to achieve fast data processing. Create dynamic cache partitions according to the material object tags, and build a timestamp index. This storage structure can classify and cache data according to the material type, and at the same time, the timestamp index facilitates processing data in chronological order. Adopt a sliding window mechanism to scan the data. The sliding window takes a certain time interval or data volume as the unit to monitor and analyze the data within the window in real time. When conflict data is found, perform multi-strategy resolution. By reasonably selecting and combining strategies, eliminate data conflicts to ensure the accuracy and consistency of the data.

[0033] Perform joint analysis on the cleaned data, and vertically construct an LSTM-GNN hybrid model. The long short-term memory network (LSTM) is good at processing sequential data and can learn the dynamic change characteristics of materials in the time dimension; the graph neural network (GNN) can mine the relationship network between materials. The combination of the two can comprehensively capture the full life cycle characteristics of materials and conduct in-depth analysis of each link from production, transportation to storage. Horizontally embed an interpretable business rule engine to convert the enterprise's business rules, industry standards, etc. into executable constraint conditions. Through this engine, not only can rule verification be performed on material data, but also the contribution degree of each rule to anomaly determination can be quantified, making the anomaly determination process transparent and interpretable, which is convenient for management personnel to understand and make decisions.

[0034] Generate an enhanced analysis report, which contains key information such as the comprehensive analysis results of material data and early warnings of abnormal situations. When the similarity between new data and historical cases exceeds the set threshold, activate the case transfer learning mechanism. This mechanism migrates the successful processing experience and model parameters in historical cases to the processing of new data, quickly updates the model, enables the model to adapt to newly emerging data patterns and business requirements, and improves the generalization ability and prediction accuracy of the model.

[0035] For abnormal data, construct a material state transition matrix, which describes the conversion probabilities and conditions between different states of materials. In the digital twin environment, by simulating the change process of material states, pre-run various correction operations, and evaluate the effects of different correction strategies. When the optimal correction plan is determined, trigger the automatic correction mechanism to adjust the material states, business processes, etc. in the actual supply chain, achieve rapid response and processing of abnormal situations, and ensure the stable operation of the supply chain.

[0036] In a possible implementation manner, the field-level semantic mapping method of the dynamic ontology library includes:

[0037] Match the predefined semantic model in the dynamic ontology library with the fields of the multi-enterprise heterogeneous material data streams. For data with the same semantics but different field names, establish a semantic mapping relationship;

[0038] Semantically analyze the text fields in the data stream through natural language processing technology, extract key semantic information and map it to the corresponding concepts in the dynamic ontology library;

[0039] Use an ontology reasoner to perform consistency checking and conflict resolution on the semantic mapping relationships to ensure the accuracy of semantic mapping.

[0040] In a possible implementation manner, constructing an industry knowledge graph including material categories, supplier relationships, and logistics paths includes:

[0041] Extract the attribute information of materials from the heterogeneous material data streams of multiple enterprises, classify the materials according to the attribute information, and construct material category nodes;

[0042] Extract data such as the basic information and transaction records of suppliers, construct supplier nodes, and connect the supplier nodes and material category nodes through edges to represent the supply relationship between suppliers and materials;

[0043] Obtain information such as the routes, times, and transportation tools of logistics transportation, construct logistics path nodes, and connect the material category nodes and logistics path nodes through edges to represent the transportation paths of materials;

[0044] Use a graph database to store and manage the industry knowledge graph, and establish indexes for nodes and edges to improve the query and retrieval efficiency of the knowledge graph.

[0045] In a possible implementation manner, the adoption of a transfer learning model to generate a standardized data table with a spatio-temporal joint index includes:

[0046] Select a pre-trained transfer learning model in a related field as the base model;

[0047] Divide the heterogeneous material data streams of multiple enterprises according to the time and space dimensions, and extract time features and space features;

[0048] Input the extracted time features, space features, and material data into the base model for fine-tuning training to make the model adapt to the field differences of different enterprises;

[0049] In the generated standardized data table, establish a spatio-temporal joint index based on timestamps and geographical locations to facilitate the rapid query and retrieval of material data within a specific time and space range.

[0050] The working principle and effects of the above technical solution are as follows: In the scenario of supply chain collaboration, there are situations where the material data streams of different enterprises have different field names and formats but the same semantics. First, match the predefined semantic models in the dynamic ontology library with the fields of the multi-enterprise heterogeneous material data streams. The semantic model is like a "translation dictionary". When encountering data with the same semantics but different field names in different enterprises, it can establish a semantic mapping relationship between the two to achieve unified conversion of data semantics. Second, for the text fields in the data stream, use natural language processing technology for semantic analysis. Through operations such as word segmentation, part-of-speech tagging, and semantic role tagging, extract key semantic information and map it to the corresponding concepts in the dynamic ontology library, so that unstructured text data can be integrated into the semantic system. Finally, with the help of the ontology reasoner, perform consistency checking and conflict resolution on the established semantic mapping relationships. The ontology reasoner, based on the logical rules of the ontology, determines whether there are contradictions in the mapping relationships. If there are, adjust them to ensure the accuracy of the semantic mapping, laying a solid foundation for subsequent data processing.

[0051] From the multi-enterprise heterogeneous material data streams, first extract the attribute information of the materials, such as specifications, models, uses, etc. Classify the materials based on these attributes, construct material category nodes for each category of materials, and form the basic elements of the knowledge graph. Then, extract data such as the basic information of the suppliers (such as name, address, contact information), transaction records, etc., construct supplier nodes, and connect the supplier nodes and the material category nodes through edges. The attributes of the edges can include information such as transaction quantity and transaction time, so as to clearly represent the supply relationship between the suppliers and the materials. Then, obtain information such as the route, time, and transportation means of the logistics transportation, construct logistics path nodes, and also connect the material category nodes and the logistics path nodes through edges to visually present the transportation path of the materials. Finally, use a graph database to store and manage the entire industry knowledge graph. The graph database is good at processing relationship data of nodes and edges, and at the same time establishes indexes for nodes and edges. When querying and retrieving the knowledge graph, it can quickly locate the target nodes and edges according to the indexes, improving the data access efficiency.

[0052] First, select a pre-trained transfer learning model in the relevant field as the base model, which already has certain data processing capabilities and feature extraction capabilities. Divide the heterogeneous material data streams of multiple enterprises according to time dimensions (such as order time, transportation time) and space dimensions (such as production location, transportation route, storage location), and extract time features (such as timestamps, time periods) and space features (such as longitude and latitude, regional codes). Input the extracted time features, space features, and material data into the base model for fine-tuning training. By adjusting the model parameters, make the model adapt to the field differences of different enterprises, and learn the commonalities and characteristics of the data of multiple enterprises, so as to generate a standardized data table. Finally, establish a spatio-temporal joint index based on timestamps and geographical locations in the standardized data table, combine the time and space dimensions, so that when querying material data within a specific time and space range, it can be quickly located and retrieved to meet the needs of real-time monitoring and analysis of the supply chain.

[0053] In a possible implementation manner, the S2 includes:

[0054] Construct a multi-dimensional evaluation system for material features, determine the weights of each first index through principal component analysis, and calculate the comprehensive material feature value;

[0055] Based on the comparison between the comprehensive feature value and the preset threshold, enter the three-level partition decision logic;

[0056] Based on the timestamp index, for partitions with different risk levels, design a differential index enhancement strategy; the differential index enhancement strategy includes: adding a risk event association index and a supplier change record index to the high-risk partition; only retaining the basic timestamp index in the medium and low-risk partitions, but regularly performing index fragmentation sorting and performance optimization.

[0057] In a possible implementation manner, the first index includes supplier credit score, historical delivery delay rate, material transaction frequency, real-time inventory turnover rate, material importance degree, and inventory volatility; the determining the weights of each index through principal component analysis and calculating the comprehensive material feature value includes:

[0058] Calculate the covariance matrix of the indexes, and obtain the eigenvalues and eigenvectors through eigenvalue decomposition;

[0059] Select the principal components according to the cumulative contribution rate of the eigenvalues, and use the eigenvectors corresponding to the principal components as the weights of each index.

[0060] The working principle and effect of the above technical solution are:

[0061] When constructing a multi-dimensional evaluation system for material characteristics, various attributes and influencing factors involved in the supply chain of materials are considered, and multiple dimensions are selected as the first-level indicators. Through linear transformation of multiple indicators, principal component analysis transforms the original multiple related indicators into a few uncorrelated comprehensive indicators (principal components). These principal components can retain as much information of the original indicators as possible. In this process, by calculating the variance contribution rate of each principal component, the weight of each first-level indicator in the comprehensive evaluation is determined. The higher the variance contribution rate, the stronger the importance of the indicator in describing the material characteristics. Finally, based on the weights and actual values of each indicator, the comprehensive characteristic value of the material is calculated. This value serves as a quantitative indicator to measure the overall characteristics of the material and provides data support for subsequent decision-making.

[0062] Based on the calculated comprehensive characteristic value of the material, it is compared with a preset threshold. The preset threshold is a reference value preset according to factors such as industry experience, enterprise historical data, and supply chain management goals. When the comprehensive characteristic value is higher than a certain higher threshold, the material is determined to be of a high-risk level; when it is in the middle range, it is of a medium-risk level; when it is lower than a lower threshold, it is of a low-risk level. Different risk levels correspond to different management strategies and resource allocation methods. After entering the three-level partition decision logic, the system will automatically match the corresponding processing procedures and measures according to the risk level of the material to achieve differential management of the material.

[0063] Based on the timestamp index, a differential index enhancement strategy is implemented for partitions with different risk levels. For materials in the high-risk partition, since they may face various risks such as supply interruption, price fluctuation, and quality problems, a risk event association index and a supplier change record index are added. The risk event association index can record various risk events that occur during the supply chain process of materials, such as transportation delays caused by natural disasters and production accidents of suppliers. Through this index, risk events related to specific materials can be quickly queried, facilitating managers to understand the risk situation in a timely manner and take corresponding measures; the supplier change record index records information such as supplier replacement and contract changes, helping enterprises grasp the dynamic changes in the upstream of the supply chain, evaluate potential risks. For materials in the medium- and low-risk partitions, their risks are relatively low and stable, and only the basic timestamp index is retained, while index fragmentation is regularly sorted out and performance is optimized. During the process of data insertion, deletion, and update, the index will generate fragments, affecting the query efficiency. By regularly sorting out the fragments, the storage structure of the index can be restored, improving the data query performance, ensuring that computing resources are reasonably allocated under the premise of meeting management requirements, and reducing management costs.

[0064] In a possible implementation manner, the three-level partition decision logic includes:

[0065] If the comprehensive eigenvalue ≥ the first threshold (high threshold), and the supplier credit score < the preset credit score, and at the same time the delivery delay rate within the preset time period > the delay rate threshold (e.g., 20%), then partition the material data independently and initiate the supplier risk warning process, increasing the monitoring frequency of the supplier's data.

[0066] If the second threshold < the comprehensive eigenvalue < the first threshold, and the material transaction frequency > x * historical average (x > 1, e.g., x = 1.5), and at the same time the inventory turnover rate < the safety threshold, then preferentially merge the partition with similar high-frequency materials, but implement an isolated storage mark for the material data to avoid read-write conflicts with other data;

[0067] If the comprehensive eigenvalue ≤ the second threshold, use a clustering algorithm (such as K-Means) to cluster the materials, dynamically merge the partitions according to the clustering results, and establish a partition association network to monitor the data interaction of each partition in real time. Once abnormal cross-partition data linkage is found, trigger a re-evaluation of the partition strategy.

[0068] In the above three-level partition decision logic, if only some conditions are met (such as only a single condition is triggered), then perform refined processing through a hierarchical response mechanism and a dynamic weight strategy, including:

[0069] If only the comprehensive eigenvalue ≥ the first threshold is met, then classify the material into the "high-concern partition", but allow it to coexist with similar low-risk materials and trigger the full life cycle tracking of the material (monitoring the entire link from procurement to delivery; and,

[0070] Set a dynamic spot-check frequency, for example, increase the quality inspection ratio per batch to 30% (e.g., the benchmark is 10%);

[0071] If only the supplier credit score < the preset credit score is met; then add a "supplier risk label" to the original partition, but maintain the existing partition structure;

[0072] If only the delivery delay rate > the delay rate threshold is met, then classify the delayed materials into the "logistics emergency partition", independent of the regular logistics data stream.

[0073] The working principle and effect of the above technical solution are:

[0074] For materials that meet the high-risk determination conditions (comprehensive eigenvalue ≥ first threshold, supplier credit score < preset credit score, and delivery delay rate > delay rate threshold within a preset time period), they are independently partitioned and the supplier risk warning process is initiated, increasing the monitoring frequency. This can timely capture potential risks of suppliers, such as capital chain breakage and production capacity decline, and take countermeasures in advance, such as finding alternative suppliers and adjusting the procurement plan, to avoid material supply interruption caused by supplier problems and reduce the operational risk of the supply chain. When only a single high-risk condition is met, such as only the comprehensive eigenvalue ≥ first threshold, the materials are classified into the "high-concern partition" and tracked throughout the life cycle, and the quality inspection ratio is increased, which can also effectively monitor the material quality and supply stability, timely discover and solve potential problems, and comprehensively ensure the safety of the supply chain.

[0075] In the medium-risk determination scenario (second threshold < comprehensive eigenvalue < first threshold, and material transaction frequency > x * historical average, and at the same time inventory turnover rate < safety threshold), it is preferred to merge the partitions with similar high-frequency materials first, reduce the number of data partitions, and lower the data management complexity; implement an isolation storage mark for the material to avoid data read-write conflicts and ensure data processing efficiency. For low-risk materials (comprehensive eigenvalue ≤ second threshold), a clustering algorithm is used to dynamically merge partitions and establish a partition association network to monitor data interaction in real time. This method can flexibly adjust partitions according to material characteristics, avoid resource waste, and achieve efficient utilization of storage resources. At the same time, the dynamically adjusted data management mode enables the system to quickly respond to changes in material data and improve the overall management efficiency.

[0076] Under the hierarchical response mechanism and dynamic weight strategy, the processing methods when only partial conditions are met achieve refined management of data. Add a "supplier risk label" only when the supplier credit score < preset credit score, and classify the delayed materials into the "logistics emergency partition" only when the delivery delay rate > delay rate threshold. This accurate data classification and labeling management enables the enterprise to quickly locate and process specific problem material data. The partition isolation and marking of data prevent interference between material data of different risk levels, ensure the security and stability of data, and also facilitate the enterprise to formulate targeted management strategies according to different data characteristics, improving the accuracy and effectiveness of data processing.

[0077] Through multi-dimensional analysis and partition management of material data, the enterprise can more clearly understand the status and risk situation of materials in the supply chain, providing rich and accurate data support for decisions such as procurement, inventory management, and supplier evaluation.

[0078] In a possible implementation manner, the S2 includes:

[0079] Determine the basic window width according to the material turnover cycle, real-time data flow rate, and the number of supply chain levels;

[0080]

[0081] Among them, is the current base window width (the time span of the sliding window at the current moment); among them, m is the number of material types (that is, the total number of material categories involved in the current window); i is the material type index, is the standard turnover cycle of the i-th material; is the supply chain level number of the i-th material; is the weight of the i-th material in the current window; is the forgetting factor, ∈[0.8, 1.2]; QPS is the data flow rate; t is the current time point (in hours);

[0082] Among them, the weight of the i-th material in the current window is determined by the importance score of the i-th material and the proportion of the total orders of the emergency orders of the i-th material in the current window ;

[0083] In a possible implementation, the weight of the i-th material in the current window is determined by the following formula:

[0084]

[0085] According to the second index of the current data, dynamically adjust the base window width to obtain the adjusted window width; the second index includes the data update frequency, the data fluctuation coefficient, and the proportion of abnormal data;

[0086]

[0087] Among them, is the adjusted window width, is the second index score of the current data, is the first coefficient 0.7 < < 1, is the second coefficient 1 < < 1.3, is the first threshold of the preset second index score of the data, is the second threshold of the preset second index score of the data;

[0088] In a possible implementation, the second index score is determined by the data update frequency, the data fluctuation coefficient, and the proportion of abnormal data.

[0089] In a possible implementation, the second index score of the data is obtained by normalizing the data update frequency, the data fluctuation coefficient, and the proportion of abnormal data respectively and then performing weighted average.

[0090] The working principle and effects of the above technical solution are as follows:

[0091] The determination of the basic window width comprehensively considers key factors such as the material turnover cycle, real-time data flow rate, and the number of supply chain levels. The standard turnover cycle of materials reflects the time required for the materials to complete a cycle from procurement to delivery in the supply chain. Materials with a long turnover cycle mean that they stay in the supply chain for a long time and require a wider window to cover their entire process. The number of supply chain levels reflects the complexity of the material circulation links. The more levels there are, the more complex the involved enterprises and processes are, and a larger window is also needed for data processing.

[0092] The weight of materials in the current window is determined by the importance score of the materials and the proportion of urgent orders. Materials with a high importance score and a large proportion of urgent orders have a greater impact on the stability and efficiency of the supply chain. In the calculation of the window width, a higher weight will be assigned to them, enabling them to obtain more sufficient data processing time and space. The forgetting factor is used to control the influence degree of historical data. The trigonometric function term in the formula introduces a time periodic factor, considering the differences in data processing requirements at different times of the day, such as busy business during the day and relatively flat at night, to achieve the dynamic change of the window width over time.

[0093] Measured from dimensions such as the value, scarcity, and key degree to production of materials, it reflects the strategic position of materials in the supply chain; the proportion of urgent orders reflects the urgency of material demand. Materials with a high proportion of urgent orders need to be processed preferentially. Through this formula, the importance and urgency of materials are quantified and normalized to determine the weight of each material in the calculation of the window width, making the calculation of the window width more in line with the actual business needs and ensuring that key materials and urgent orders can be processed in a timely and sufficient manner.

[0094] The basic window width is dynamically adjusted according to the second indicator of the current data. The second indicator covers the data update frequency, data fluctuation coefficient, and the proportion of abnormal data. By normalizing these three elements separately and then performing a weighted average, the second indicator score of the data is obtained.

[0095] The data update frequency reflects the real-time nature of the data. Frequently updated data requires a narrower window for timely processing to avoid data backlog; the data fluctuation coefficient measures the degree of data change. Data with large fluctuations requires the window to have stronger adaptability; the proportion of abnormal data reflects the quality and potential risks of the data. In the case of a large proportion of abnormal data, a wider window is needed for detailed analysis and processing.

[0096] Based on the comparison result between the second indicator score and the preset threshold, the window width is adjusted to ensure sufficient data processing. Through this dynamic adjustment mechanism, the sliding window width can flexibly change according to the actual state of the data, achieving efficient and accurate processing of material data.

[0097] In a possible implementation, the S3 includes:

[0098] The joint analysis of the cleaned data includes:

[0099] a) Vertical intelligent analysis: Construct a deep learning model that fuses temporal and topological features, where:

[0100] Use the LSTM network to process temporal features such as the inventory level of materials, order frequency, and quality inspection pass rate, and extract the time series patterns of their entire life cycle;

[0101] Model the multi-level association relationship of suppliers - logistics - warehousing through a graph neural network (CNN); the node features include the historical transaction credibility score and real-time logistics delay; the edge weights reflect the tightness of collaboration between nodes;

[0102] Among them, the historical transaction credibility score Is obtained through the following formula:

[0103]

[0104] Design a spatio-temporal cross-attention mechanism to dynamically weight the key time nodes and core topological nodes, and achieve multi-scale fusion of temporal features and network structure.

[0105] b) Horizontal rule analysis: Embed an interpretable business rule engine, where:

[0106] Generate a dynamic business constraint set based on historical data mining and knowledge graph derivation;

[0107] Quantify the contribution of each rule to anomaly determination, and use a dynamic weight strategy to sort the rule priorities;

[0108] c) Joint decision-making mechanism: When there is a conflict in the determination results of the deep learning model and the rule engine, perform the following operations:

[0109] Conduct interpretable traceability of the model output features to locate the key influencing factors;

[0110] Conduct reverse verification on the triggered rules to evaluate the applicability of the rules in the current scenario;

[0111] Generate a final anomaly determination conclusion based on the traceability and verification results.

[0112] In a possible implementation, the implementation method of the spatio-temporal attention mechanism includes:

[0113] Convert the temporal feature vector output by the LSTM into a query vector Q;

[0114] Convert the topological node features output by the GNN into key-value matrices K and V;

[0115] The dynamic weighted fusion of temporal features and topological features is achieved through attention weight calculation.

[0116] In a possible implementation, the joint decision-making mechanism further includes:

[0117] Establish an interactive training mechanism between models and rules, including encoding the rule triggering status as a feature vector and inputting it into the deep learning model; feeding back the feature importance of the model output to the rule engine to optimize the constraints

[0118] Build a conflict case knowledge base for model fine-tuning and rule base version iteration.

[0119] The working principle and effect of the above technical solution are:

[0120] In vertical intelligent analysis, a deep learning model that integrates timing and topological features is constructed, LSTM networks and graph neural networks (GNNs) are used to process different characteristics of materials respectively, and multi-scale fusion is achieved through the spatiotemporal cross-attention mechanism.

[0121] LSTM networks are good at processing data with time series characteristics. For time series characteristics such as inventory, order frequency, and quality inspection pass rate of materials, LSTM networks can effectively capture the long-term dependencies and patterns of time series in the entire life cycle of materials through memory units and gating mechanisms, and extract key time series feature information; GNN is used to model the multi-level relationship between suppliers, logistics, and warehousing, and regards suppliers, logistics nodes, and warehousing nodes as nodes of the graph. Node features include historical transaction credibility scores, which are calculated by the number of successful transactions / total number of transactions × log (1 + years of cooperation), reflecting supplier transaction reliability and cooperation stability) and real-time logistics delays. The edge weight reflects the closeness of collaboration between nodes, thereby constructing a topological structure model of the supply chain and mining the potential relationships between nodes.

[0122] The spatiotemporal cross-attention mechanism converts the time series feature vector output by LSTM into a query vector Q, and the topological node features output by GNN into key-value matrices K and V. By calculating the attention weights, the key time nodes and core topological nodes are dynamically weighted to achieve multi-scale fusion of time series features and network structure, so that the model can simultaneously use the time change law of materials and the topological structure information of the supply chain for analysis.

[0123] The horizontal rule analysis embeds an interpretable business rule engine, which generates a dynamic business constraint set based on historical data mining and knowledge graph derivation. By analyzing historical material data, combining industry knowledge and enterprise business rules, logical relationships and constraint conditions are extracted from the knowledge graph to form a rule set applicable to the current business scenario. At the same time, the contribution degree of each rule to anomaly determination is quantified, and a dynamic weight strategy is adopted to prioritize the rules. According to the importance and effectiveness of the rules in different scenarios, their weights are adjusted to ensure that key rules can play a role first during the anomaly determination process, improving the accuracy and efficiency of rule analysis.

[0124] When there is a conflict between the determination results of the deep learning model and the rule engine, the joint decision-making mechanism is activated. First, perform an interpretable traceability of the model output features, use model interpretation techniques to locate the factors that have a key impact on the determination results, and understand the basis for the model to make decisions; second, perform reverse verification on the triggered rules, combine the actual data and background information of the current business scenario, and evaluate the applicability of the rules in this scenario; finally, generate a final anomaly determination conclusion based on the traceability and verification results, comprehensively consider model analysis and rule judgment, and obtain a more accurate and reliable decision.

[0125] In addition, the joint decision-making mechanism also establishes an interactive training mechanism between the model and the rules. Encoding the rule trigger status as a feature vector and inputting it into the deep learning model enables the model to learn the business logic contained in the rules; feedback the feature importance output by the model to the rule engine to optimize the constraint conditions, realizing the dynamic update and optimization of the rules. At the same time, construct a conflict case knowledge base, record and analyze each conflict case, and use it for model fine-tuning and rule library version iteration to continuously improve the accuracy and adaptability of the model and the rules.

[0126] In a possible implementation, the S4 includes:

[0127] When new material data arrives, extract its spatio-temporal feature vector; the spatio-temporal feature vector includes time phase encoding, geographical grid encoding, and material status encoding;

[0128] When the similarity exceeds the threshold, freeze the underlying feature extraction layer of the model and only fine-tune the top-level decision network.

[0129] In a possible implementation, the similarity calculation adopts a hybrid metric method, including:

[0130] Calculate the improved cosine similarity for structured data :

[0131]

[0132] Among them, 、 The data timestamp for two institutionalized vectors; , are the values on the j-th dimension of two structured vectors;

[0133] Extract the knowledge graph embedding vector for unstructured data and calculate the Euclidean distance;

[0134] Obtain the final similarity by dynamically weighted fusion of structured and unstructured similarities.

[0135] The working principle and effect of the above technical solution are as follows:

[0136] When new material data arrives, the system will extract multi-dimensional features from it to generate spatio-temporal feature vectors. Time-phase encoding digitally converts time information. For example, a day is divided into multiple time periods, and information such as the transaction time and transportation time of materials is encoded into specific numerical vectors to reflect the characteristics of material data in the time series; Geographic grid encoding is to grid the geographical location where the material is located, and each grid corresponds to a code, thus converting the geographical information of the material into a form that can be processed by a computer, facilitating the analysis of the spatial distribution and flow law of materials; Material status encoding is to encode and identify the current status of the material, such as inventory status, transportation status, quality inspection status, etc., so that the status information of the material can be incorporated into the feature vector. Through these three encoding methods, the spatio-temporal information and status information of the new material data are converted into feature vectors that can be calculated and analyzed, providing a basis for subsequent processing.

[0137] The similarity calculation adopts a hybrid metric method, which processes structured data and unstructured data separately and fuses the results. For structured data, an improved cosine similarity formula is used for calculation. On the basis of the traditional cosine similarity, a timestamp factor is introduced to reflect the timeliness of the data. The data closer to the current time has a higher weight in the similarity calculation.

[0138] For unstructured data, first extract the knowledge graph embedding vector, map the unstructured data to a low-dimensional vector space, and then calculate the Euclidean distance. The smaller the Euclidean distance, the higher the similarity of the unstructured data in the feature space. Finally, through dynamic weighting, according to the importance of structured data and unstructured data in different scenarios, the two similarity calculation results are fused to obtain the final similarity value, so as to judge the similarity between the new material data and the historical data. Accurate similarity calculation and targeted model adjustment enable the system to better identify the relationship and differences between the new material data and the historical data.

[0139] When the similarity between the new material data and the historical data exceeds the threshold, in order to ensure that the model can utilize the existing knowledge while quickly adapting to the characteristics of the new data, the system adopts the strategy of freezing the underlying feature extraction layer of the model and only fine-tuning the top-level decision network. The underlying feature extraction layer of the model usually learns the general basic features of the data, and these features have a certain universality in different data scenarios. Freezing the underlying layer can retain the general knowledge learned by the model; while the top-level decision network is responsible for making the final decision based on the features extracted from the underlying layer. Fine-tuning the top-level network for the new data can enable the model to quickly adjust the decision logic on the existing basis, adapt to the characteristics of the new material data, and achieve efficient processing and accurate analysis of the new data.

[0140] By extracting spatio-temporal feature vectors, the material data is transformed into a unified computable form, which is convenient for quickly calculating similarity and model processing. The hybrid metric method adopts the optimal calculation method for different types of data and reduces unnecessary computational complexity by dynamically weighting the fusion results.

[0141] Freezing the underlying feature extraction layer of the model avoids retraining a large number of underlying parameters, saving computational resources and time costs. In practical applications, the training of deep learning models often requires consuming a large amount of computational resources and time, especially for complex underlying network structures. In this way, on the premise of ensuring the model performance, the model training cost is greatly reduced, the resource utilization efficiency is improved, and enterprises can effectively manage and analyze the new material data at a lower cost. Only fine-tuning the top-level decision network enables the model to quickly adapt to the characteristics of the new material data while retaining the existing knowledge, enhances the adaptability of the model to different scenarios and data changes, improves the generalization ability of the model, and enables the model to maintain high accuracy and reliability when processing various new material data.

[0142] In one possible implementation, the S5 includes:

[0143] Generate a material state transition probability model based on the real-time supply chain network topology, where the node state includes three-dimensional features of inventory level, quality inspection status, and in-transit logistics volume; the weight of the transition edge is obtained through the following formula:

[0144]

[0145] In one possible implementation, perform the following simulations in parallel in a virtual environment:

[0146] Monte Carlo simulation: Randomly perturb the key parameters to generate, for example, 500+ risk scenarios;

[0147] Discrete event simulation: Inject typical abnormal events such as equipment failures and sudden surges in demand;

[0148] Reinforcement learning exploration: iteratively optimize and correct the strategy through Q-learning;

[0149] Intelligent correction trigger: when all of the following conditions are met simultaneously: simulated revenue gain ≥ k3 × operating cost, where k3 is a constant greater than 1,

[0150] critical path redundancy ≥ safety threshold and a preset proportion (such as 90%) of the above collaborative nodes reach a consensus, automatically execute the correction operation and synchronously update the status of all network nodes;

[0151] The process of automatic correction execution includes:

[0152] Blockchain evidence storage mechanism, write the key parameters of the correction decision into the consortium chain, including: correction trigger timestamp, hash value of the simulation deduction result, and set of collaborative node signatures.

[0153] The working principle and effect of the above technical solution are:

[0154] The above steps construct a material state transition probability model based on the real-time supply chain network topology. The node state covers the characteristics of three dimensions: inventory level, quality inspection status, and in-transit logistics volume, comprehensively reflecting the actual situation of materials in the supply chain. The weight of the transfer edge is determined by the formula "transfer edge weight = historical successful transfer times / total transfer attempt times × 1 / (1 + current path load rate)". The ratio of historical successful transfer times to total transfer attempt times reflects the historical reliability of this transfer path, while the current path load rate reflects the busyness of the current path. The higher the load rate, the lower the edge weight, so as to dynamically adjust the possibility of transfer.

[0155] Carry out three simulations in parallel in the virtual environment:

[0156] Monte Carlo simulation of various possible supply chain risk situations helps to discover potential problems and challenges.

[0157] Discrete event simulation of the response and changes of the supply chain in case of emergencies to evaluate the robustness of the system.

[0158] Reinforcement learning exploration: use the Q-learning algorithm to iteratively optimize and correct the strategy.

[0159] By continuously trying different correction strategies and adjusting the strategy according to the reward value feedback from the environment, gradually find the optimal correction plan.

[0160] When the following three conditions are met simultaneously, automatically execute the correction operation, including:

[0161] simulated revenue gain ≥ k3 × operating cost (k3 is a constant greater than 1), ensuring that the correction operation is economically feasible, that is, the revenue is greater than the cost;

[0162] The redundancy of the critical path ≥ the safety threshold to ensure that the critical path of the supply chain has sufficient backup capacity to cope with the uncertainties during the correction process;

[0163] More than a preset proportion (e.g., 90%) of the collaborative nodes reach a consensus, which reflects the collaboration and consistency among the nodes in the supply chain. The correction is only carried out when most nodes approve.

[0164] Once the automatic correction is triggered, the blockchain evidence storage mechanism will be adopted; the key parameters of the correction decision, such as the correction trigger timestamp, the hash value of the simulation result, and the set of signatures of the collaborative nodes, will be written into the consortium chain; the comparison between the simulated revenue gain and the operation cost in the intelligent correction trigger condition avoids unnecessary resource waste, improves the operation efficiency and economic benefits of the supply chain, and the correction operation can only be triggered when a preset proportion of the collaborative nodes reach a consensus, which promotes the communication and collaboration among the nodes in the supply chain.

[0165] The embodiment of the present application also provides a supply chain collaborative material management system for the supply chain collaborative material management method described in the embodiment of the present application. The system includes:

[0166] A data standardization module, which is used to receive multi-enterprise heterogeneous material data streams, construct an industry knowledge graph including material categories, supplier relationships, and logistics paths based on the field-level semantic mapping of the dynamic ontology library; and generate a standardized data table with a spatio-temporal joint index;

[0167] A data storage and resolution module, which is used to inject the standardized data into a distributed message queue for parallel processing, create a dynamic cache partition according to the material object label and build a timestamp index inside, scan the data using a sliding window mechanism, and perform multi-strategy resolution on the conflicting data within the window;

[0168] A joint analysis module, which is used to perform joint analysis on the cleaned data, vertically construct an LSTM-GNN hybrid model to capture the full life cycle characteristics of materials, horizontally embed an interpretable business rule engine to define constraint conditions, and quantify the contribution of each rule to the anomaly determination;

[0169] A report generation module, which is used to generate an enhanced analysis report and activate the case transfer learning mechanism to update the model when the similarity between the new data and the historical cases exceeds the threshold;

[0170] An anomaly handling module, which is used to construct a material state transition matrix for the abnormal data, simulate the correction operation in the digital twin environment and trigger automatic correction.

[0171] The working principles and effects of the above technical solutions are the same as those in the method embodiment of the present application and will not be elaborated here.

[0172] This application is described from the perspectives of purpose of use, efficacy, progress, and novelty, and has met the functional improvement and usage requirements emphasized by the patent law. The above description and accompanying drawings of this application are only preferred embodiments of this application and do not limit this application thereby. Therefore, all those that are similar or identical to the structure, device, features, etc. of this application, that is, all equivalent substitutions or modifications made according to the scope of the patent application of this application, shall fall within the scope of patent application protection of this application.

Claims

1. A supply chain collaborative material management method, characterized in that: The method comprises: Receive heterogeneous material data streams from multiple enterprises, build an industry knowledge graph including material categories, supplier relationships, and logistics paths based on field-level semantic mapping of the dynamic ontology library, and generate standardized data tables with spatiotemporal joint indexes; Inject standardized data into distributed message queues for parallel processing, create dynamic cache partitions based on material object tags and build in timestamp indexes, use a sliding window mechanism to scan data, and perform multi-strategy resolution on conflicting data within the window; Conduct joint analysis on the cleaned data, build a LSTM-GNN hybrid model vertically to capture the characteristics of the entire life cycle of materials, embed an interpretable business rule engine horizontally to define constraints, and quantify the contribution of each rule to anomaly determination; Generate enhanced analysis reports, and activate the case transfer learning mechanism to update the model when the similarity between new data and historical cases exceeds the threshold; Construct a material state transfer matrix for abnormal data, simulate correction operations in the digital twin environment and trigger automatic correction.

2. The supply chain collaborative material management method according to claim 1, characterized in that: The construction of the industry knowledge graph including material categories, supplier relationships and logistics paths includes: Extract the attribute information of materials from the heterogeneous material data streams of multiple enterprises, classify the materials according to the attribute information, and construct material category nodes; Build supplier nodes through supplier information, and connect supplier nodes with material category nodes through edges; Through logistics transportation information, the logistics path nodes are constructed, and the material category nodes and logistics path nodes are connected through edges; A graph database is used to store and manage the industry knowledge graph, and to establish an index of nodes and edges.

3. The supply chain collaborative material management method according to claim 1, characterized in that: The S2 includes: Construct a multi-dimensional evaluation system for material characteristics, determine the weight of each first indicator through principal component analysis, and calculate the comprehensive characteristic value of materials; the first indicators include supplier credit score, historical delivery delay rate, material transaction frequency, real-time inventory turnover rate, material importance and inventory volatility; Based on the comparison between the comprehensive feature value and the preset threshold, the three-level partition decision logic is entered; Based on the timestamp index, differentiated index enhancement strategies are designed for partitions of different risk levels.

4. The supply chain collaborative material management method according to claim 3 is characterized in that: The three-level partition decision logic includes: If the comprehensive characteristic value is ≥ the first threshold, and the supplier's credit score is less than the preset credit score, and the delivery delay rate within the preset time period is greater than the delay rate threshold, the material data will be partitioned independently, and the supplier risk warning process will be initiated to increase the monitoring frequency of the supplier's data; If the second threshold is less than the comprehensive characteristic value and less than the first threshold, and the material transaction frequency is greater than x*historical mean, and the inventory turnover rate is less than the safety threshold, the material will be merged with the high-frequency materials of the same type and partitioned, but the material data will be marked for isolated storage; If the comprehensive characteristic value is ≤ the second threshold, a clustering algorithm is used to cluster the materials. The partitions are dynamically merged according to the clustering results, and a partition association network is established to monitor the data interaction of each partition in real time. If the cross-partition data is abnormally linked, the partition strategy is triggered to be re-evaluated.

5. The supply chain collaborative material management method according to claim 1, characterized in that: The S2 includes: Determine the basic window width based on the material turnover cycle, real-time data flow rate, and the number of supply chain levels; ； in, is the current basic window width; m is the number of material types; i is the material type index, is the standard turnover cycle of the ith material; is the number of supply chain levels of the i-th material; is the weight of the i-th material in the current window; Forgetting Factor ∈[0.8,1.2]; QPS is the real-time data flow rate; t is the current time point; According to a second indicator of the current data, the basic window width is dynamically adjusted to obtain the adjusted window width; the second indicator includes the data update frequency, the data fluctuation coefficient and the proportion of abnormal data.

6. The supply chain collaborative material management method according to claim 1, characterized in that: The S2 includes: Perform confidence attenuation fusion processing on the overlapping data at the window edge; Identify numerical, temporal, and logical conflicts through attention mechanism models; Layered execution of basic and enhanced strategies. The basic strategy includes timestamp priority coverage and weighted average of data sources. The enhanced strategy includes path backtracking verification and multi-version data fusion. Dynamically select strategy combinations based on the Q-learning model to generate conflict resolution traceability maps.

7. The supply chain collaborative material management method according to claim 1, characterized in that: The S3 includes: Construct a deep learning model that integrates time series and topological features, including: using LSTM network to extract the time series pattern of the entire life cycle of materials; modeling the multi-level relationship between suppliers, logistics and warehousing through graph neural network; designing a spatiotemporal attention mechanism to dynamically integrate time series features and network topological structure; Embedded explainable business rule engine, including: generating dynamic business constraint sets based on historical data mining and knowledge graph derivation to quantify the contribution of each rule to anomaly determination, and using dynamic weight strategy to prioritize rules; When the deep learning model conflicts with the rule engine's decision results, the following actions are performed: Conduct interpretable traceability of model output features and locate key influencing factors; Reversely verify the triggering rules and evaluate their applicability in the current scenario; Generate the final abnormality judgment conclusion based on the traceability and verification results.

8. The supply chain collaborative material management method according to claim 1, characterized in that: The S4 includes: When new material data arrives, extract its spatiotemporal feature vector; the spatiotemporal feature vector includes time phase code, geographic grid code, and material status code; When the similarity exceeds the threshold, the underlying feature extraction layer of the model is frozen and only the top-level decision network is fine-tuned.

9. The supply chain collaborative material management method according to claim 1, characterized in that: The S5 includes: The material status transfer probability model is generated based on the real-time supply chain network topology, where the node status includes three-dimensional characteristics of inventory level, quality inspection status, and logistics in-transit quantity; the transfer edge weight is obtained by the following formula: ； When the simulated revenue gain ≥ k3 × operation cost, critical path redundancy ≥ safety threshold, and the preset proportion of the above collaborative nodes reach a consensus, the correction operation is automatically performed and the node status of the entire network is updated synchronously; k3 is a constant greater than 1.

10. A supply chain collaborative material management system, used to implement the supply chain collaborative material management method according to claim 1, characterized in that: The system comprises: The data standardization module is used to receive heterogeneous material data streams from multiple enterprises, build an industry knowledge graph containing material categories, supplier relationships, and logistics paths based on field-level semantic mapping of the dynamic ontology library, and generate a standardized data table with spatiotemporal joint indexes; The data storage and resolution module is used to inject standardized data into the distributed message queue for parallel processing, create dynamic cache partitions according to material object tags and build in timestamp indexes, use a sliding window mechanism to scan data, and perform multi-strategy resolution on conflicting data within the window; The joint analysis module is used to conduct joint analysis on the cleaned data, vertically construct an LSTM-GNN hybrid model to capture the characteristics of the entire life cycle of materials, horizontally embed an interpretable business rule engine to define constraints, and quantify the contribution of each rule to anomaly determination; Report generation module, used to generate enhanced analysis reports. When the similarity between new data and historical cases exceeds the threshold, the case transfer learning mechanism is activated to update the model. The exception handling module is used to construct a material state transfer matrix for abnormal data, simulate correction operations in the digital twin environment, and trigger automatic corrections.

Images