Intelligent material declaration early warning system and method based on big data
The big data-based intelligent material declaration and early warning system solves the problems of scattered data storage and reliance on manual experience in traditional material declaration systems. It achieves real-time unified integration and high-precision modeling of data, provides interactive early warning support, and improves the accuracy and efficiency of declarations.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TIANJIN JIANGTIAN ZHIYUN TECHNOLOGY CO LTD
- Filing Date
- 2026-01-07
- Publication Date
- 2026-05-08
AI Technical Summary
Traditional material reporting systems in steel companies suffer from fragmented data storage and a lack of unified data collection and synchronization mechanisms. This results in data fragmentation, and the reliance on manual experience in reporting methods fails to effectively capture the complex relationship between material demand and supply chain fluctuations. Existing early warning mechanisms cannot proactively identify and provide interpretable decision-making basis, leading to delayed detection of reporting anomalies and high correction costs.
The system employs a big data-based intelligent material declaration and early warning system, which includes a data acquisition and synchronization module, a data cleaning and preprocessing module, an algorithm calculation module, and an early warning generation module. Through a hybrid integrated learning model and a knowledge graph dynamic early warning inference engine, it achieves real-time data acquisition, cleaning, analysis, and early warning generation, providing interactive decision support.
It achieves real-time unified integration of cross-system data, improving data reliability and availability. Through multi-algorithm collaborative modeling, it generates quantifiable early warning information, reduces subjective decision-making bias, identifies application anomalies in real time and provides risk alerts, thereby improving the application success rate.
Smart Images
Figure CN121998582A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of enterprise material declaration management applications, and in particular to a material intelligent declaration early warning system and method based on big data. Background Technology
[0002] Material declaration and early warning systems are part of enterprise intelligent manufacturing and supply chain management processes. Due to the complexity of steel enterprise production processes, involving multiple stages such as raw material procurement, production scheduling, and inventory management, traditional declaration methods rely on manual experience and judgment, making them difficult to adapt to dynamic market changes and the demands of refined management. How to efficiently integrate scattered data and achieve intelligent early warning has become a key issue in improving the accuracy of enterprise declarations and decision-making efficiency.
[0003] Common material declaration systems in steel enterprises generally suffer from structural technical bottlenecks: data from various business processes is scattered across production execution systems, warehouse management systems, and supply chain platforms, lacking a unified data collection and synchronization mechanism, resulting in fragmented information required for declaration decisions. The absence or insufficient processing capacity of data cleaning and preprocessing stages leads to persistent problems such as duplicate data, outliers, and format discrepancies, severely undermining the reliability of basic data. At the analytical modeling level, declaration threshold setting methods relying on manual experience or single algorithms cannot effectively capture the complex correlation between material demand and supply chain fluctuations, and the lack of quantitative assessment methods makes it difficult to verify the rationality of declarations. Existing early warning mechanisms mostly adopt post-event statistical methods, which cannot achieve proactive risk identification or provide interpretable decision-making basis, resulting in delayed detection of declaration anomalies and high correction costs, failing to meet the operational requirements of enterprise material declaration management applications. Therefore, a big data-based intelligent material declaration early warning system and method are proposed. Summary of the Invention
[0004] This invention provides the following technical solution: a big data-based intelligent material declaration and early warning system, comprising:
[0005] The system includes a data acquisition and synchronization module, a data cleaning and preprocessing module, an algorithm calculation module, an early warning generation module, and a user interaction module. The data acquisition and synchronization module is connected to the data cleaning and preprocessing module via a data cable. The data acquisition and synchronization module is used to collect and synchronize structured data and ERP data scattered in different systems to the big data platform in real time.
[0006] The data cleaning and preprocessing module is used to clean and preprocess the data during the synchronization process, including removing duplicate records, correcting erroneous data, filling in missing values, and performing standardization.
[0007] The algorithm calculation module is used to analyze and model the cleaned data using machine learning algorithms and generate parameter factors. The algorithm calculation module is connected to the preprocessing module via a data cable and data cleaning. The algorithm calculation module integrates a hybrid ensemble learning model and is equipped with random forest, XGBoost and deep neural networks.
[0008] The early warning generation module is used to generate early warning information based on parameter factors and convert the early warning information into text format using natural language processing technology. The early warning generation module has a big data model application submodule, which takes the constraints in the generated text format as input and inputs them into the big data model for calculation and analysis to generate prediction results.
[0009] The user interaction module is used to feed back warning information to users and provide an interactive interface for users to refer to in decision-making. The warning generation module integrates a dynamic warning inference engine based on a knowledge graph. The dynamic warning inference engine based on the knowledge graph is used to construct an industry knowledge graph and associate it with data on materials, supply chains and market trends.
[0010] This invention provides a material intelligent declaration and early warning method based on big data. Based on the aforementioned big data-based material intelligent declaration and early warning system, the method includes the following steps:
[0011] S1 Data Acquisition and Synchronization:
[0012] The distributed architecture collects structured data and ERP data scattered across different systems in real time and synchronizes them to the big data platform.
[0013] S2 Intelligent Data Cleaning and Preprocessing:
[0014] The synchronous data is cleaned through a multi-stage pipeline. First, a rule engine and pattern recognition technology are used to detect null values, outliers and format errors in the data. Then, an intelligent repair strategy library is called to automatically repair the abnormal data. Finally, the data standardization transformation is completed through the built-in industry standard data dictionary.
[0015] S3 Hybrid Integration Model Analysis and Calculation:
[0016] A hybrid ensemble learning model, including random forest, XGBoost and deep neural network, is used to perform multi-dimensional analysis on the data after preprocessing in step S2. Random forest is responsible for the initial feature screening, XGBoost performs feature optimization and weighting, and deep neural network processes unstructured data features. Finally, a weighted voting mechanism is used to generate a comprehensive parameter factor that includes key indicators such as inventory status and consumption trend.
[0017] S4 Knowledge-Enhanced Intelligent Early Warning Generation:
[0018] The parameter factors generated in step S3 are input into the big data model application submodule based on microservice architecture, and the prediction results are converted into early warning information in text format through natural language processing technology. At the same time, based on the knowledge graph dynamic reasoning engine, material, supply chain and market trend data are associated to generate multi-dimensional early warning rules.
[0019] S5 Interactive Early Warning Decision Support:
[0020] The system uses a responsive interactive interface to display early warning information dashboards, decision support tools, and feedback collection components. It dynamically visualizes the distribution of early warning levels, provides historical case queries and handling suggestions, and records user feedback to optimize the model.
[0021] Preferably, the data acquisition synchronization module adopts a distributed architecture design, which includes data acquisition nodes and a central scheduling server. The acquisition nodes are deployed in the network area where the data source system is located and maintain a connection with the central scheduling server through an asynchronous communication mechanism. The central scheduling server uses a load balancing strategy to dynamically allocate data acquisition tasks. The data acquisition synchronization module also has a built-in breakpoint resume mechanism.
[0022] Preferably, the data cleaning and preprocessing module adopts a multi-level pipeline processing architecture, which includes a data quality detection layer, an anomaly data processing layer, and a standardization transformation layer in sequence. The data quality detection layer automatically identifies null values, outliers, and format errors in the data through a rule engine and pattern recognition technology. The anomaly data processing layer is equipped with an intelligent repair strategy library, and the standardization transformation layer has a built-in industry standard data dictionary.
[0023] Preferably, the big data model application submodule in the early warning generation module adopts a microservice architecture design. The microservice architecture design includes a model loading service, a data preprocessing service, and a prediction execution service. The model loading service is used to manage the deployment and hot switching of different versions of the early warning model. The data preprocessing service is used to convert the input parameter factors into the tensor format required by the model. The prediction execution service is used to realize high-concurrency prediction request processing through a distributed computing framework.
[0024] Preferably, the dynamic early warning reasoning engine of the knowledge graph uses a graph database to store industry knowledge data and constructs a knowledge network containing multi-dimensional relationships such as material entities, supplier nodes, and market indicators. The dynamic early warning reasoning engine of the knowledge graph has built-in rule reasoning and semantic reasoning dual engines. The rule reasoning engine is used to execute logical judgments based on business rules, and the semantic reasoning engine is used to mine potential correlation patterns through graph neural networks.
[0025] Preferably, the user interaction module adopts a responsive design. The user interface of the user interaction module includes a warning information dashboard, a decision support tool, and a feedback collection component. The warning information dashboard dynamically displays the distribution and trend changes of warning levels through visual charts. The decision support tool is used to provide historical case query and disposal suggestion generation functions. The feedback collection component is used to record users' opinions on the processing of warning information.
[0026] Preferably, in the data acquisition and synchronization process of step S1, a distributed acquisition network is constructed using edge computing nodes, and non-intrusive data capture is achieved by deploying a lightweight data proxy module at the business system exit. Asynchronous transmission buffering is achieved by combining message queues, and a two-dimensional data synchronization verification mechanism is constructed based on timestamps and business serial numbers.
[0027] Preferably, the multi-stage pipeline in step S2 adopts a dynamically scalable architecture design, including three parallel processing channels. The first parallel processing channel is configured with a rule engine based on reinforcement learning, which automatically identifies null value types and generates filling strategies by constructing a decision tree containing 3,600 industry data patterns. The second parallel processing channel integrates a convolutional neural network and uses an attention mechanism to locate abnormal expression patterns in text fields. The third parallel processing channel deploys an adversarial generative network and generates synthetic data samples that conform to business rules through adversarial training.
[0028] Preferably, the random forest model in step S3 adopts a dynamic feature importance evaluation mechanism, which updates the feature weights in each iteration cycle through the permutation importance algorithm, and focuses on screening twelve core business indicators such as inventory turnover rate and supplier delivery timeliness rate. When constructing XGBoost, SHAP value interpretability constraints are introduced, and a dedicated time-series processing branch is set up inside the deep neural network. The multi-head attention mechanism of the Transformer architecture is used to parse the unstructured data of procurement contract text and logistics trajectory.
[0029] In summary, compared with the prior art, the present invention provides a material intelligent declaration and early warning system and method based on big data, which has the following beneficial effects:
[0030] 1. This invention achieves real-time collection and synchronization of structured data and ERP data across systems and departments through a data acquisition and synchronization module, effectively breaking down the data silos problem in traditional steel enterprises. Simultaneously, it integrates data scattered across different stages such as production lines, warehousing systems, and supply chain systems into a unified big data platform, providing a complete and consistent data foundation for enterprise reporting and decision-making. This significantly improves decision-making blind spots caused by data scarcity. Furthermore, through a data cleaning and preprocessing module, it automatically removes duplicate records, corrects erroneous data, fills in missing values, and standardizes the data, solving the data disorder problem caused by weak information infrastructure in steel enterprises. This improves data reliability and usability, ensures the accuracy of subsequent analysis, and avoids the risk of discrepancies between reported data and actual conditions due to data distortion.
[0031] 2. This invention integrates a hybrid ensemble learning model of random forest, XGBoost, and deep neural networks through an algorithm calculation module. This enables high-precision modeling of material demand and supply chain fluctuations through multi-algorithm collaboration, and the generated parameter factors can quantitatively reflect the rationality of the declaration data. Combined with the predictive analysis of the big data model application submodule, it provides enterprises with data-driven declaration threshold suggestions, reducing subjective decision-making bias. At the same time, the early warning generation module uses natural language processing technology to transform the analysis results into understandable text warning information, and combines a knowledge graph dynamic reasoning engine to associate multi-dimensional data of materials and market trends. This allows for real-time identification of declaration anomalies and proactive risk warnings, helping enterprises correct data deviations before declaration and improving the success rate of declarations. Attached Figure Description
[0032] Figure 1 This is a schematic diagram of the system structure of the present invention.
[0033] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation
[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0035] Please see Figure 1 This invention provides a technical solution: a big data-based intelligent material reporting and early warning system, comprising:
[0036] The system includes a data acquisition and synchronization module, a data cleaning and preprocessing module, an algorithm calculation module, an early warning generation module, and a user interaction module. The data acquisition and synchronization module is connected to the data cleaning and preprocessing module via a data cable. The data acquisition and synchronization module is used to collect and synchronize structured data and ERP data scattered in different systems to the big data platform in real time.
[0037] The data cleaning and preprocessing module is used to clean and preprocess data during the synchronization process, including removing duplicate records, correcting erroneous data, filling in missing values, and performing standardization. The data acquisition and synchronization module adopts a distributed architecture design, which includes data acquisition nodes and a central scheduling server. The acquisition nodes are deployed in the network area where the data source system is located and maintain a connection with the central scheduling server through an asynchronous communication mechanism. The central scheduling server uses a load balancing strategy to dynamically allocate data acquisition tasks. The data acquisition and synchronization module also has a built-in breakpoint resume mechanism. The data cleaning and preprocessing module adopts a multi-level pipeline processing architecture, which includes a data quality detection layer, an anomaly data processing layer, and a standardization transformation layer. The data quality detection layer automatically identifies null values, outliers, and format errors in the data through a rule engine and pattern recognition technology. The anomaly data processing layer is configured with an intelligent repair strategy library, and the standardization transformation layer has a built-in industry standard data dictionary.
[0038] The algorithm calculation module is used to analyze and model the cleaned data using machine learning algorithms and generate parameter factors. The algorithm calculation module is connected to the preprocessing module via data cable and data cleaning. The algorithm calculation module integrates a hybrid ensemble learning model and is equipped with random forest, XGBoost and deep neural network.
[0039] The early warning generation module generates early warning information based on parameter factors and converts the information into text format using natural language processing technology. Internally, it includes a big data model application submodule. This submodule takes the generated text constraints as input and feeds them into the big data model for calculation and analysis to generate prediction results. The big data model application submodule within the early warning generation module adopts a microservice architecture design, which includes a model loading service, a data preprocessing service, and a prediction execution service. The model loading service manages the deployment and hot-switching of different versions of the early warning model, the data preprocessing service converts the input parameter factors into the tensor format required by the model, and the prediction execution service handles high-concurrency prediction requests through a distributed computing framework.
[0040] The user interaction module is used to provide early warning information to users and provide an interactive interface for users to refer to in decision-making. The user interaction module adopts a responsive design. The interactive interface of the user interaction module includes an early warning information dashboard, decision support tools, and a feedback collection component. The early warning information dashboard dynamically displays the distribution and trend changes of early warning levels through visual charts. The decision support tools are used to provide historical case query and disposal suggestion generation functions. The feedback collection component is used to record users' opinions on the handling of early warning information.
[0041] The early warning generation module integrates a dynamic early warning reasoning engine based on a knowledge graph. This engine constructs an industry knowledge graph and associates it with data on materials, supply chains, and market trends. It uses a graph database to store industry knowledge data and builds a knowledge network that includes multi-dimensional relationships such as material entities, supplier nodes, and market indicators. The engine incorporates both rule-based reasoning and semantic reasoning engines. The rule-based reasoning engine executes logical judgments based on business rules, while the semantic reasoning engine mines potential correlation patterns through graph neural networks.
[0042] Please see Figure 2 This invention provides a material intelligent declaration and early warning method based on big data. Based on the aforementioned big data-based material intelligent declaration and early warning system, the method includes the following steps:
[0043] S1 Data Acquisition and Synchronization:
[0044] The system uses a distributed architecture to collect structured data and ERP data scattered across different systems in real time and synchronizes them to a big data platform. It also uses edge computing nodes to build a distributed collection network and deploys a lightweight data proxy module at the business system exit to achieve non-intrusive data capture. It combines message queues to achieve asynchronous transmission buffering and builds a two-dimensional data synchronization verification mechanism based on timestamps and business serial numbers. The detailed process of the above steps is as follows:
[0045] Construction of Distributed Data Acquisition Network:
[0046] Lightweight data acquisition nodes are deployed at the network edge, close to data sources such as ERP and production management systems. Each node operates independently to avoid single points of failure. The nodes utilize containerization technology, supporting rapid expansion and dynamic resource scheduling. A lightweight proxy module is deployed at the business system egress point to capture data via bypass monitoring or database log parsing, without affecting the normal operation of the original business system. The proxy module only extracts key business fields (such as order number, material code, quantity, etc.) and performs preliminary data format standardization.
[0047] Asynchronous transfer and message queue buffering:
[0048] The collected data is transmitted asynchronously via message queues such as Kafka or RabbitMQ to ensure data stability in high-concurrency scenarios. Messages are stored in partitions according to business type to avoid data backlog. Each edge node records the data transmission progress, and if the network is interrupted, transmission will automatically resume from the breakpoint after recovery to avoid data loss or duplicate collection.
[0049] Two-dimensional data synchronization and verification mechanism:
[0050] Data collection is time-stamped to milliseconds. The central server verifies the temporal continuity of data across different nodes to ensure data is entered into the database in the actual business sequence. A globally unique serial number (e.g., "order number + system identifier") is generated for each transaction. The central server uses idempotency verification of the serial number to prevent duplicate data from being entered into the database. A reconciliation task is performed daily to verify the data consistency between edge nodes and the central database. An alarm is triggered when the difference rate exceeds a threshold.
[0051] Load balancing and dynamic scheduling:
[0052] A dynamic load balancing strategy is adopted to allocate data collection tasks based on the computing power and network conditions of each edge node. Idle nodes receive high-priority data collection requests first, and the CPU, memory, and network load of nodes are monitored in real time. When system resources are scarce, the system automatically switches to sampling collection mode and records a degradation flag for subsequent data repair.
[0053] Data integrity guarantee:
[0054] Before data is written to the big data platform, a final format and logic check is performed to remove abnormal data such as illegal characters and out-of-bounds values. For missing data that fails the check, log backtracking or supplementary data collection through business system interfaces is automatically triggered.
[0055] S2 Intelligent Data Cleaning and Preprocessing:
[0056] The synchronous data is cleaned through a multi-stage pipeline. First, a rule engine and pattern recognition technology are used to detect null values, outliers, and format errors in the data. Then, an intelligent repair strategy library is called to automatically repair the abnormal data. Finally, the data is standardized and transformed through a built-in industry standard data dictionary. The multi-stage pipeline adopts a dynamic and scalable architecture design, including three parallel processing channels. The first parallel processing channel is configured with a rule engine based on reinforcement learning. By building a decision tree containing 3,600 industry data patterns, it automatically identifies null value types and generates filling strategies. The second parallel processing channel integrates a convolutional neural network and uses an attention mechanism to locate abnormal expression patterns in text fields. The third parallel processing channel deploys a generative adversarial network and generates synthetic data samples that conform to business rules through generative adversarial training.
[0057] S3 Hybrid Integration Model Analysis and Calculation:
[0058] A hybrid ensemble learning model incorporating random forest, XGBoost, and deep neural networks is used to perform multi-dimensional analysis on the data preprocessed in step S2. Random forest is responsible for initial feature screening, XGBoost performs feature optimization and weighting, and deep neural networks process unstructured data features. Finally, a weighted voting mechanism is used to generate comprehensive parameter factors that include key indicators such as inventory status and consumption trends. The random forest model adopts a dynamic feature importance evaluation mechanism, updating feature weights in each iteration cycle through a permutation importance algorithm, and focusing on screening twelve core business indicators such as inventory turnover rate and supplier delivery timeliness rate. XGBoost incorporates SHAP value interpretability constraints during construction and sets up a dedicated time-series processing branch within the deep neural network. A multi-head attention mechanism based on the Transformer architecture is used to parse unstructured data such as procurement contract text and logistics trajectory.
[0059] S4 Knowledge-Enhanced Intelligent Early Warning Generation:
[0060] The parameter factors generated in step S3 are input into the big data model application submodule based on microservice architecture, and the prediction results are converted into early warning information in text format through natural language processing technology. At the same time, based on the knowledge graph dynamic reasoning engine, material, supply chain and market trend data are associated to generate multi-dimensional early warning rules. The specific implementation process of the above method is as follows:
[0061] Early warning model calculation under microservice architecture:
[0062] The comprehensive parameter factors (such as inventory status and consumption trends) generated in the S3 stage are transmitted to the big data model application submodule through a standardized interface. The data preprocessing service automatically verifies data integrity, removes outliers, and converts the data into a tensor format that the model can process. The model loading service dynamically calls pre-trained random forest, XGBoost, or deep neural network models based on the business type to perform high-concurrency predictions. The prediction execution service adopts a distributed computing framework, supporting the processing of thousands of requests per second to ensure real-time performance. The outputs of different models generate the final prediction result through a weighted voting mechanism, with weights dynamically adjusted based on historical accuracy (e.g., deep neural networks assign higher weights to unstructured data).
[0063] Natural language warning information generation:
[0064] It uses natural language processing technology to convert numerical prediction results (such as "inventory turnover rate is below the threshold of 15%) into readable text (such as "Warning: There is a high risk of inventory backlog for material A, it is recommended to reduce purchases"). It has multiple built-in warning templates and automatically adapts the expression style according to the user role (such as purchaser, warehouse manager), and supports switching between Chinese and English.
[0065] Knowledge graph dynamic association analysis:
[0066] The knowledge graph engine extracts relevant data such as supplier delivery records and market supply and demand trends for current materials from the graph database and cross-validates them with the prediction results. For example, when a shortage of a certain material is predicted, it automatically associates the inventory status of its substitute materials, executes hard business rules (such as "if the supplier's historical on-time delivery rate is <90% and the market price fluctuation is >5%, trigger a high-risk warning"), and mines potential patterns through graph neural networks (such as identifying the implicit correlation between logistics delays in a certain region and specific weather conditions).
[0067] Multi-dimensional early warning rule generation:
[0068] Combining predicted values with knowledge graph reasoning results, early warnings are categorized into three levels: "high," "medium," and "low." For example, a red warning is issued when both "inventory is below the safety line" and "supplier capacity is insufficient," and countermeasures are recommended based on a historical case database (e.g., "In a similar case in 2023, advance procurement of material B could reduce the risk of production stoppage").
[0069] S5 Interactive Early Warning Decision Support:
[0070] The system features a responsive interactive interface that displays an early warning dashboard, decision support tools, and feedback collection components. It dynamically visualizes the distribution of early warning levels, provides historical case queries and handling suggestions, and records user feedback to optimize the model.
[0071] Feedback loop and model optimization:
[0072] Users can label warning results as "valid" or "invalid". Feedback data is used for incremental model training. The latest market data and supply chain changes are synchronized daily to ensure the timeliness of the reasoning.
[0073] This solution achieves real-time collection and synchronization of structured and ERP data across systems and departments through a data acquisition and synchronization module. This effectively breaks down the data silos present in traditional steel enterprises. Simultaneously, it integrates data scattered across different stages such as production lines, warehousing systems, and supply chain systems into a unified big data platform, providing a complete and consistent data foundation for enterprise reporting and decision-making. This significantly improves decision-making blind spots caused by data scarcity. Furthermore, the data cleaning and preprocessing module automatically removes duplicate records, corrects erroneous data, fills in missing values, and standardizes the data, solving the data disorder problem caused by weak IT infrastructure in steel enterprises. This improves data reliability and usability, ensures the accuracy of subsequent analysis, and avoids the risk of discrepancies between reported data and actual conditions due to data distortion.
[0074] This solution integrates a hybrid ensemble learning model combining random forest, XGBoost, and deep neural networks through an algorithm computation module. This enables high-precision modeling of material demand and supply chain fluctuations through multi-algorithm collaboration, and the generated parameter factors can quantify the rationality of the declared data. Combined with the predictive analysis of the big data model application submodule, it provides enterprises with data-driven declaration threshold suggestions, reducing subjective decision-making bias. Meanwhile, the early warning generation module uses natural language processing technology to transform the analysis results into understandable text warning information, and combines a knowledge graph dynamic inference engine to link multi-dimensional data on materials and market trends. This allows for real-time identification of declaration anomalies and proactive risk alerts, helping enterprises correct data deviations before declaration and improving the success rate of declarations.
[0075] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0076] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A material intelligent declaration and early warning system based on big data, characterized in that, include: The system includes a data acquisition and synchronization module, a data cleaning and preprocessing module, an algorithm calculation module, an early warning generation module, and a user interaction module. The data acquisition and synchronization module is connected to the data cleaning and preprocessing module via a data cable. The data acquisition and synchronization module is used to collect and synchronize structured data and ERP data scattered in different systems to the big data platform in real time. The data cleaning and preprocessing module is used to clean and preprocess the data during the synchronization process, including removing duplicate records, correcting erroneous data, filling in missing values, and performing standardization. The algorithm calculation module is used to analyze and model the cleaned data using machine learning algorithms and generate parameter factors. The algorithm calculation module is connected to the preprocessing module via a data cable and data cleaning. The algorithm calculation module integrates a hybrid ensemble learning model and is equipped with random forest, XGBoost and deep neural networks. The early warning generation module is used to generate early warning information based on parameter factors and convert the early warning information into text format using natural language processing technology. The early warning generation module has a big data model application submodule, which takes the constraints in the generated text format as input and inputs them into the big data model for calculation and analysis to generate prediction results. The user interaction module is used to feed back warning information to users and provide an interactive interface for users to refer to in decision-making. The warning generation module integrates a dynamic warning inference engine based on a knowledge graph. The dynamic warning inference engine based on the knowledge graph is used to construct an industry knowledge graph and associate it with data on materials, supply chains and market trends.
2. The intelligent material reporting and early warning system based on big data according to claim 1, characterized in that: The data acquisition synchronization module adopts a distributed architecture design, which includes data acquisition nodes and a central scheduling server. The acquisition nodes are deployed in the network area where the data source system is located and maintain a connection with the central scheduling server through an asynchronous communication mechanism. The central scheduling server uses a load balancing strategy to dynamically allocate data acquisition tasks. The data acquisition synchronization module also has a built-in breakpoint resume mechanism.
3. The intelligent material reporting and early warning system based on big data according to claim 1, characterized in that: The data cleaning and preprocessing module adopts a multi-level pipeline processing architecture, which includes a data quality detection layer, an anomaly data processing layer, and a standardization transformation layer. The data quality detection layer automatically identifies null values, outliers, and format errors in the data through a rule engine and pattern recognition technology. The anomaly data processing layer is equipped with an intelligent repair strategy library, and the standardization transformation layer has a built-in industry standard data dictionary.
4. The intelligent material reporting and early warning system based on big data according to claim 1, characterized in that: The big data model application submodule in the early warning generation module adopts a microservice architecture design. The microservice architecture design includes a model loading service, a data preprocessing service, and a prediction execution service. The model loading service is used to manage the deployment and hot switching of different versions of the early warning model. The data preprocessing service is used to convert the input parameter factors into tensor formats required by the model. The prediction execution service is used to realize high-concurrency prediction request processing through a distributed computing framework.
5. The intelligent material reporting and early warning system based on big data according to claim 1, characterized in that: The dynamic early warning reasoning engine of the knowledge graph uses a graph database to store industry knowledge data and constructs a knowledge network containing multi-dimensional relationships of material entities, supplier nodes and market indicators. The dynamic early warning reasoning engine of the knowledge graph has built-in rule reasoning and semantic reasoning dual engines. The rule reasoning engine is used to perform logical judgments based on business rules, and the semantic reasoning engine is used to mine potential correlation patterns through graph neural networks.
6. The intelligent material reporting and early warning system based on big data according to claim 1, characterized in that: The user interaction module adopts a responsive design. The user interface of the user interaction module includes an early warning information dashboard, a decision support tool, and a feedback collection component. The early warning information dashboard dynamically displays the distribution and trend changes of early warning levels through visual charts. The decision support tool is used to provide historical case query and disposal suggestion generation functions. The feedback collection component is used to record users' opinions on the processing of early warning information.
7. A material intelligent declaration and early warning method based on big data, based on the material intelligent declaration and early warning system based on big data as described in any one of claims 1-6, characterized in that, Includes the following steps: S1 Data Acquisition and Synchronization: The distributed architecture collects structured data and ERP data scattered across different systems in real time and synchronizes them to the big data platform. S2 Intelligent Data Cleaning and Preprocessing: The synchronous data is cleaned through a multi-stage pipeline. First, a rule engine and pattern recognition technology are used to detect null values, outliers and format errors in the data. Then, an intelligent repair strategy library is called to automatically repair the abnormal data. Finally, the data standardization transformation is completed through the built-in industry standard data dictionary. S3 Hybrid Integration Model Analysis and Calculation: A hybrid ensemble learning model, including random forest, XGBoost and deep neural network, is used to perform multi-dimensional analysis on the data after preprocessing in step S2. Random forest is responsible for the initial feature screening, XGBoost performs feature optimization and weighting, and deep neural network processes unstructured data features. Finally, a weighted voting mechanism is used to generate a comprehensive parameter factor that includes key indicators of inventory status and consumption trend. S4 Knowledge-Enhanced Intelligent Early Warning Generation: The parameter factors generated in step S3 are input into the big data model application submodule based on microservice architecture, and the prediction results are converted into early warning information in text format through natural language processing technology. At the same time, based on the knowledge graph dynamic reasoning engine, material, supply chain and market trend data are associated to generate multi-dimensional early warning rules. S5 Interactive Early Warning Decision Support: The system uses a responsive interactive interface to display early warning information dashboards, decision support tools, and feedback collection components. It dynamically visualizes the distribution of early warning levels, provides historical case queries and handling suggestions, and records user feedback to optimize the model.
8. The intelligent material reporting and early warning method based on big data according to claim 7, characterized in that: In the data acquisition and synchronization process of step S1, a distributed acquisition network is built using edge computing nodes, and non-intrusive data capture is achieved by deploying a lightweight data proxy module at the business system exit. Asynchronous transmission buffering is achieved by combining message queues, and a two-dimensional data synchronization verification mechanism is built based on timestamps and business serial numbers.
9. The intelligent material reporting and early warning method based on big data according to claim 7, characterized in that: The multi-stage pipeline in step S2 adopts a dynamic and scalable architecture design, including three parallel processing channels. The first parallel processing channel is configured with a rule engine based on reinforcement learning, which automatically identifies null value types and generates filling strategies by constructing decision trees. The second parallel processing channel integrates a convolutional neural network and uses an attention mechanism to locate abnormal expression patterns in text fields. The third parallel processing channel deploys an adversarial generative network and generates synthetic data samples that conform to business rules through adversarial training.
10. The intelligent material reporting and early warning method based on big data according to claim 7, characterized in that: The random forest model in step S3 adopts a dynamic feature importance evaluation mechanism. It updates the feature weights in each iteration cycle through the permutation importance algorithm, and focuses on screening the core business indicators such as inventory turnover rate and supplier delivery timeliness rate. When constructing XGBoost, SHAP value interpretation constraints are introduced, and a dedicated time-series processing branch is set up inside the deep neural network. The multi-head attention mechanism of the Transformer architecture is used to parse the unstructured data of procurement contract text and logistics trajectory.