Decision recommendation system and method for preferentially processing key data of supply chain of manufacturing enterprise

Abstract

The invention relates to a decision recommendation system and method for preferentially processing key data of a supply chain of a manufacturing enterprise. According to the decision recommendation system and method, through a data fusion module, a priority evaluation module, a decision recommendation module and a self-adaptive feedback module, supply chain key data management realizes collection, analysis and decision; the invention discloses an optimized decision recommendation system and method for an automatic closed-loop process. The system is characterized by comprising a data fusion module, a priority evaluation module, a decision recommendation module and a self-adaptive feedback module, the data fusion module outputs structured data to the priority evaluation module, and the decision recommendation module generates a scheme according to the priority and feeds back an actual effect to the self-adaptive feedback module after execution. The self-adaptive feedback module reversely updates the priority rule; the data fusion module is composed of a data acquisition unit, a data preprocessing unit and a data storage unit, and the priority evaluation module is composed of a rule engine unit, a machine learning model unit and a dynamic weight adjustment unit.

Description

Technical Field

[0001] This invention relates to a decision recommendation system and method for prioritizing key data in the manufacturing enterprise supply chain. It is an automated and intelligent decision recommendation system and method for prioritizing key data in the supply chain, belonging to the field of supply chain data management technology. In particular, it relates to a decision recommendation system and method that enables the automated closed-loop process of collection, analysis, decision-making, and optimization of key data in the supply chain through a data fusion module, a priority evaluation module, a decision recommendation module, and an adaptive feedback module. Background Technology

[0002] In the operation of manufacturing enterprises, supply chain management is crucial, and prioritizing key data and making scientific decision recommendations is key to improving supply chain efficiency and enterprise competitiveness. However, current decision recommendation systems and methods for prioritizing key supply chain data have bottlenecks. First, in terms of priority assessment, most enterprises usually prioritize data based on simple rules or human experience, failing to comprehensively consider the impact of multiple factors on the importance of data. Moreover, differences in experience among different personnel may lead to inconsistent assessment results, lacking objectivity and stability. Second, decision recommendations rely heavily on historical experience and simple data analysis, which cannot cope with complex and ever-changing supply chain environments, especially when faced with sudden disruptions in raw material supply, failing to provide quick and effective response decisions.

[0003] Announcement No. CN119904257A discloses a supply chain data management method and system based on data analysis. First, it filters information at each node in the supply chain, removing duplicate information and shortening subsequent data processing time. Second, it analyzes the non-duplicate node information to identify core nodes in the supply chain, making subsequent data retrieval more convenient. Then, it compares these core nodes to determine the central node of the supply chain. This solution enables unified management of supply chain data, avoiding multiple data retrievals from different companies' databases. However, this solution lacks key data identification and priority assessment, and it cannot adapt to dynamic changes, failing to provide direct support for enterprise decision-making. Summary of the Invention

[0004] To improve the above situation, the present invention provides a decision recommendation system and method for prioritizing the processing of key data in the supply chain of manufacturing enterprises. This system and method enables the automated closed-loop process of collection, analysis, decision-making, and optimization of key data in the supply chain through a data fusion module, a priority evaluation module, a decision recommendation module, and an adaptive feedback module.

[0005] The decision recommendation system and method for prioritizing key data processing in the manufacturing enterprise supply chain of the present invention are implemented as follows: The decision recommendation system for prioritizing key data processing in the manufacturing enterprise supply chain of the present invention includes a data fusion module, a priority evaluation module, a decision recommendation module, and an adaptive feedback module; The feature is that the data fusion module outputs structured data to the priority evaluation module, the decision recommendation module generates a scheme according to the priority, and after execution, it feeds back the actual effect to the adaptive feedback module, and the adaptive feedback module updates the priority rules in reverse. The data fusion module consists of a data acquisition unit, a data preprocessing unit, and a data storage unit. The data acquisition unit collects production data, logistics data, market data, and supplier data in real time. The data acquisition unit outputs the raw data to the data preprocessing unit, which performs data cleaning, data standardization, unstructured data processing, and data association on the raw data. The data preprocessing unit outputs the preprocessed data to the data storage unit, which stores and manages the data through a distributed storage architecture, blockchain notarization, and caching mechanisms. The priority evaluation module consists of a rule engine unit, a machine learning model unit, and a dynamic weight adjustment unit. The rule engine unit establishes a rule base, performs real-time rule matching and dynamic rule loading to classify supply chain events into preliminary priorities, and provides a rapid response mechanism. The real-time rule matching extracts event attributes and parses key information in the event, including event type, timestamp, related resource (order, warehouse, supplier, etc.) identifiers, and event context parameters, forming a standard structure for rule matching. The matching methods support sequential matching, index matching (creating an index based on rule tags or event types), and conditional expression evaluation. After a successful match, the corresponding priority level and processing suggestions are immediately extracted to generate a preliminary response task. The dynamic rule loading means that the rule engine unit supports dynamically adding, modifying, and deleting rules without interrupting system operation; The machine learning model unit performs intelligent priority assessment of supply chain events through model architecture, feature engineering, and online prediction methods, while the dynamic weight adjustment unit dynamically optimizes the priority assessment strategy based on real-time feedback. The decision recommendation module consists of a multi-objective optimization unit, a scenario simulation unit, and a recommendation output unit. The multi-objective optimization unit generates an optimized decision scheme set based on the priority evaluation results through objective system construction, optimization algorithm, and constraint modeling. The scenario simulation unit simulates the candidate decision schemes through digital twin construction, predicts the execution results, and outputs optimized schemes with risk assessment. The recommendation output unit transforms the optimized schemes into executable decision suggestions and outputs them through scheme formatting and visualization. The constraint modeling submodule establishes linear or nonlinear constraints based on the actual supply chain business logic, including but not limited to resource capacity constraints (including manpower, logistics, and capital), time window constraints (such as the requirement to respond within 48 hours), concurrent event priority conflicts, SLA (Service Level Agreement) restrictions, and indivisible task conditions (such as the requirement that a certain type of event must be processed as a whole). The adaptive feedback module consists of an execution feedback unit, a model iteration unit, and a knowledge base construction unit. The execution feedback unit collects decision execution effect data in real time and outputs cleaned feedback data to the model iteration unit. The model iteration unit continuously optimizes each algorithm model based on the feedback data. The knowledge base construction unit accumulates supply chain decision knowledge and provides training samples for model iteration. The optimization results of the adaptive feedback module are used to empower the data fusion module, the priority evaluation module, and the decision recommendation module. This invention also relates to a decision-making recommendation method for prioritizing key data processing in a manufacturing enterprise's supply chain, characterized by utilizing a decision-making recommendation system for prioritizing key supply chain data processing, comprising the following steps: The raw data acquired by the data acquisition unit is transmitted to the data preprocessing unit. The data preprocessing unit preprocesses the raw data and then outputs the preprocessed data to the data storage unit for storage and management. The processed data is output to the priority evaluation module. The rule engine unit and machine learning model realize the priority evaluation of supply chain events. The multi-objective optimization unit generates a set of optimized decision schemes. The scenario simulation unit simulates the candidate decision schemes, predicts the execution results, and outputs the optimized scheme with risk assessment. The recommended output unit formats the optimization scheme and transforms it into actionable decision recommendations in a visual presentation. The execution feedback unit collects decision execution effect data in real time and outputs the feedback data to the model iteration unit. The model iteration unit continuously optimizes each algorithm model based on this feedback data to improve the accuracy and adaptability of the model. The knowledge base construction unit accumulates supply chain decision knowledge and provides training samples for model iteration. Ultimately, a closed-loop process of "collection-analysis-decision-optimization" is achieved. Beneficial effects

[0006] First, by working together with the rule engine unit and the machine learning model unit, and by comprehensively considering multiple factors, the priority evaluation results are ensured to be unaffected by subjective experience, thus achieving objectivity in the evaluation process and stability in the evaluation results.

[0007] Second, when faced with complex emergencies such as raw material supply disruptions, the system can quickly generate optimized decision-making solutions that include risk assessments, providing accurate and effective response strategies, and improving the speed and quality of enterprise decision-making response in complex supply chain environments.

[0008] Third, the dynamic weight adjustment unit dynamically optimizes the priority evaluation strategy in real time based on the decision execution effect data collected by the adaptive feedback module. This enables the priority evaluation rules to be continuously adjusted according to actual business changes, maintaining the timeliness and accuracy of priority evaluation of supply chain events. Detailed Implementation

[0009] The decision recommendation system for prioritizing key data processing in the manufacturing enterprise supply chain of the present invention is implemented as follows: The decision recommendation system for prioritizing key data processing in the manufacturing enterprise supply chain of the present invention includes a data fusion module, a priority evaluation module, a decision recommendation module, and an adaptive feedback module; The feature is that the data fusion module outputs structured data to the priority evaluation module, the decision recommendation module generates a scheme according to the priority, and after execution, it feeds back the actual effect to the adaptive feedback module, and the adaptive feedback module updates the priority rules in reverse. The data fusion module consists of a data acquisition unit, a data preprocessing unit, and a data storage unit. The data acquisition unit collects production data, logistics data, market data, and supplier data in real time. The data acquisition unit outputs the raw data to the data preprocessing unit, which performs data cleaning, data standardization, unstructured data processing, and data association on the raw data. The data preprocessing unit outputs the preprocessed data to the data storage unit, which stores and manages the data through a distributed storage architecture, blockchain notarization, and caching mechanisms. Furthermore, the production data is collected in real time by IoT sensors deployed on the production line to monitor the equipment's operating status. Furthermore, the logistics data is collected by connecting to the transportation management system and the warehouse management system to obtain real-time transportation trajectory, inventory status, and abnormal events. Preferably, the distributed storage architecture includes real-time data storage in a time-series database and supports high-concurrency writing, while historical data is archived to a data lake; Preferably, the caching mechanism is to cache frequently accessed data (such as the daily inventory status) in Redis; The priority evaluation module consists of a rule engine unit, a machine learning model unit, and a dynamic weight adjustment unit. The rule engine unit establishes a rule base, performs real-time rule matching and dynamic rule loading to classify supply chain events into preliminary priorities, and provides a rapid response mechanism. Furthermore, the rule base establishment includes rule data structure design. The rules are defined in a structured way, including: rule ID, triggering condition (a Boolean expression consisting of multiple attribute variables, such as "order delay > 48 hours AND supplier rating = low"), applicable scenario tags (logistics anomaly", "inventory shortage", "quality complaint", etc.), priority level, and processing suggestions / instructions. The real-time rule matching extracts event attributes and parses key information in the event, including event type, timestamp, related resource (order, warehouse, supplier, etc.) identifiers, and event context parameters, forming a standard structure for rule matching. The matching methods support sequential matching, index matching (creating an index based on rule tags or event types), and conditional expression evaluation. After a successful match, the corresponding priority level and processing suggestions are immediately extracted to generate a preliminary response task. The dynamic rule loading means that the rule engine unit supports dynamically adding, modifying, and deleting rules without interrupting system operation; The machine learning model unit performs intelligent priority assessment of supply chain events through model architecture, feature engineering, and online prediction methods, while the dynamic weight adjustment unit dynamically optimizes the priority assessment strategy based on real-time feedback. Preferably, the model architecture includes urgency assessment: LSTM neural network (temporal feature analysis), influence assessment: random forest (multi-feature association analysis), and timeliness assessment: XGBoost (structured feature processing). Furthermore, the urgency assessment sub-model is used to identify event development trends, abnormal fluctuations, and suddenness characteristics. The input features are event evolution data constructed based on time series, including delayed time series, inventory fluctuation records, historical alarm frequencies, etc. A multi-layer LSTM (Long Short-Term Memory) network structure is adopted to extract time-dependent information, and finally outputs the urgency assessment results. Furthermore, the impact assessment sub-model is used to analyze the potential impact of events on key nodes in the supply chain (including order fulfillment, customer satisfaction, inventory costs, etc.). Input features include order amount, key materials involved, supplier level, number of upstream and downstream links, customer level, etc. The model is constructed using a multi-decision tree random ensemble and finally outputs the impact assessment results. Preferably, the timeliness assessment sub-model is used to assess whether an event needs to be responded to or processed in a short period of time; the input features include event duration, historical average processing time, resource availability, number of concurrent events, etc., and the structured features are modeled using a gradient boosting tree (XGBoost) to achieve high fitting accuracy; the final output is the timeliness assessment result. Preferably, the outputs of the urgency assessment sub-model, the influence assessment sub-model, and the timeliness assessment sub-model are weighted and fused to form a comprehensive priority assessment result; Preferably, the feature engineering module includes: a) static feature extraction: including event type, supplier level, customer level, product category, etc.; b) dynamic feature extraction: including real-time event indicators (delay minutes, order backlog), abnormal growth rate, upstream and downstream response time, etc.; c) derived feature construction: including event periodic statistical features, moving average, growth rate, Z-score anomaly score, etc.; d) feature standardization processing: normalizing or standardizing continuous features to adapt to the input requirements of different models. Preferably, the weight management setting of the dynamic weight adjustment unit is an initial weight and a dynamic adjustment weight. The initial weight is set by the experience of professionals, and the dynamic adjustment weight is based on the weight optimization of PPO algorithm reinforcement learning. The decision recommendation module consists of a multi-objective optimization unit, a scenario simulation unit, and a recommendation output unit. The multi-objective optimization unit generates an optimized decision scheme set based on the priority evaluation results through objective system construction, optimization algorithm, and constraint modeling. The scenario simulation unit simulates the candidate decision schemes through digital twin construction, predicts the execution results, and outputs optimized schemes with risk assessment. The recommendation output unit transforms the optimized schemes into executable decision suggestions and outputs them through scheme formatting and visualization. Furthermore, the target system construction submodule constructs an objective function based on enterprise operation indicators, including but not limited to minimizing processing time, minimizing cost, maximizing customer satisfaction, maximizing resource utilization balance, and maximizing event risk mitigation rate, and jointly optimizing them in a Pareto front manner. The constraint modeling submodule establishes linear or nonlinear constraints based on the actual supply chain business logic, including but not limited to resource capacity constraints (including manpower, logistics, and capital), time window constraints (such as the requirement to respond within 48 hours), concurrent event priority conflicts, SLA (Service Level Agreement) restrictions, and indivisible task conditions (such as the requirement that a certain type of event must be processed as a whole). Preferably, the multi-objective optimization unit uses a genetic algorithm to generate a set of optimization decision schemes; Furthermore, the digital twin is constructed through supply chain entity modeling and dynamic parameter configuration; Preferred output channels for the recommended output unit include desktop decision dashboards, mobile alert push notifications, and automatic reports via email or SMS. The adaptive feedback module consists of an execution feedback unit, a model iteration unit, and a knowledge base construction unit. The execution feedback unit collects decision execution effect data in real time and outputs cleaned feedback data to the model iteration unit. The model iteration unit continuously optimizes each algorithm model based on the feedback data. The knowledge base construction unit accumulates supply chain decision knowledge and provides training samples for model iteration. The optimization results of the adaptive feedback module are used to empower the data fusion module, the priority evaluation module, and the decision recommendation module. Furthermore, the reverse empowerment of the adaptive feedback module includes updating the data acquisition strategy and optimizing the data quality rules by the data fusion module; adjusting the feature weights of the priority evaluation module, hot updating the model parameters, reconstructing the objective function of the decision recommendation module, and calibrating the simulation parameters. This invention discloses a decision-making recommendation method for prioritizing key data processing in a manufacturing enterprise's supply chain. The method utilizes a decision-making recommendation system for prioritizing key supply chain data processing to make decisions, and includes the following steps: The raw data acquired by the data acquisition unit is transmitted to the data preprocessing unit. The data preprocessing unit preprocesses the raw data and then outputs the preprocessed data to the data storage unit for storage and management. The processed data is output to the priority evaluation module. The rule engine unit and machine learning model realize the priority evaluation of supply chain events. The multi-objective optimization unit generates a set of optimized decision schemes. The scenario simulation unit simulates the candidate decision schemes, predicts the execution results, and outputs the optimized scheme with risk assessment. The recommended output unit formats the optimization scheme and transforms it into actionable decision recommendations in a visual presentation. The execution feedback unit collects decision execution effect data in real time and outputs the feedback data to the model iteration unit. The model iteration unit continuously optimizes each algorithm model based on this feedback data to improve the accuracy and adaptability of the model. The knowledge base construction unit accumulates supply chain decision knowledge and provides training samples for model iteration. Ultimately, this achieves a closed-loop process of "collection-analysis-decision-optimization"; The distributed storage architecture includes storing real-time data in a time-series database and supporting high-concurrency writing, while archiving historical data to a data lake. It can store real-time data and historical data separately, select appropriate storage methods based on data characteristics, optimize the allocation of storage resources, and improve the overall performance and efficiency of the storage system. The caching mechanism caches frequently accessed data in Redis, which can shorten the data access response time. Relevant staff can quickly obtain real-time inventory information from the cache and respond to inquiries quickly. The model architecture includes urgency assessment: LSTM neural network (temporal feature analysis), impact assessment: random forest (multi-feature association analysis), and timeliness assessment: XGBoost (structured feature processing). The multi-model architecture can comprehensively assess supply chain events from different dimensions. The three complement each other and work together to improve the ability to cope with supply chain risks and changes. The dynamic weight adjustment unit sets initial weights and dynamic adjustment weights. The initial weights are set by professional experience, while the dynamic adjustment weights are optimized based on PPO algorithm reinforcement learning. The initial weights provide an experience-based framework to ensure that the system can provide reasonable evaluation results in the early stages of operation. The dynamic adjustment weights utilize the learning ability of reinforcement learning on actual business data to continuously optimize the evaluation model. The combination of the two can avoid the lag caused by relying solely on experience and prevent the blindness of relying solely on data learning in the early stages of the system due to a lack of guidance. The output channels of the recommended output unit include desktop decision dashboards, mobile alert push notifications, and automatic reports via email or SMS, which can ensure timely information delivery while meeting the needs of diverse office scenarios and improving decision-making efficiency and response speed. The goal is to achieve an automated closed-loop process for the collection, analysis, decision-making, and optimization of key supply chain data management through data fusion, priority evaluation, decision recommendation, and adaptive feedback modules.

[0010] It should be noted that, unless otherwise explicitly specified and limited, the terms "placed," "connected," and "linked" should be interpreted broadly. For example, they can refer to fixed connections such as folded edges, rivets, pins, adhesives, and welds; detachable connections such as threaded connections, snap-fit ​​connections, and hinges; integral connections; electrical connections; direct connections; or indirect connections via an intermediate medium; or internal connections between two components. Those skilled in the art can understand the specific meaning of these terms in this invention based on the specific circumstances.

[0011] It should be further noted that, in order to keep the description simple and clear, the above specific embodiments only describe the differences between them and other embodiments. However, those skilled in the art should know that the above specific embodiments are also independent technical solutions.

Claims

1. A decision-making recommendation system for prioritizing key data in a manufacturing enterprise's supply chain, comprising a data fusion module, a priority evaluation module, a decision recommendation module, and an adaptive feedback module; characterized in that: The data fusion module outputs structured data to the priority evaluation module, the decision recommendation module generates a scheme based on the priority, and after execution, feeds back the actual effect to the adaptive feedback module, which in turn updates the priority rules. The data fusion module consists of a data acquisition unit, a data preprocessing unit, and a data storage unit. The priority evaluation module consists of a rule engine unit, a machine learning model unit, and a dynamic weight adjustment unit. The decision recommendation module consists of a multi-objective optimization unit, a scenario simulation unit, and a recommendation output unit. The adaptive feedback module consists of an execution feedback unit, a model iteration unit, and a knowledge base construction unit.

2. The decision recommendation system for prioritizing key data processing in the supply chain of a manufacturing enterprise according to claim 1, characterized in that... The data acquisition unit collects production data, logistics data, market data, and supplier data in real time. The data acquisition unit outputs the raw data to the data preprocessing unit, which performs data cleaning, data standardization, unstructured data processing, and data association on the raw data. The data preprocessing unit outputs the preprocessed data to the data storage unit, which stores and manages the data through a distributed storage architecture, blockchain notarization, and caching mechanisms.

3. The decision recommendation system for prioritizing key data processing in the supply chain of a manufacturing enterprise according to claim 1, characterized in that... The rule engine unit performs preliminary priority classification of supply chain events through rule base establishment, real-time rule matching, and dynamic rule loading, providing a rapid response mechanism. The machine learning model unit performs intelligent priority evaluation of supply chain events through model architecture, feature engineering, and online prediction. The dynamic weight adjustment unit dynamically optimizes the priority evaluation strategy based on real-time feedback. The dynamic weight adjustment unit sets initial weights and dynamically adjusted weights. The initial weights are set by professional experience, and the dynamic weights are optimized based on PPO algorithm reinforcement learning.

4. A decision recommendation system for prioritizing key data processing in a manufacturing enterprise's supply chain, as described in claim 1, is characterized in that... The multi-objective optimization unit generates an optimization decision scheme set based on the priority evaluation results through objective system construction, optimization algorithm and constraint modeling. The scenario simulation unit simulates the candidate decision schemes through digital twin construction, predicts the execution results, and outputs the optimization scheme with risk assessment. The recommendation output unit transforms the optimization scheme into an executable decision suggestion and outputs it through scheme formatting and visualization. The digital twin is constructed through supply chain entity modeling and dynamic parameter configuration. Recommended output channels for the output unit include desktop decision dashboards, mobile alert push notifications, and automatic reports via email or SMS.

5. A decision recommendation system for prioritizing key data processing in a manufacturing enterprise's supply chain, as described in claim 1, is characterized in that... The execution feedback unit collects decision execution effect data in real time and outputs cleaned feedback data to the model iteration unit. The model iteration unit continuously optimizes each algorithm model based on the feedback data. The knowledge base construction unit accumulates supply chain decision knowledge and provides training samples for model iteration. The optimization results of the adaptive feedback module are used to back-empower the data fusion module, priority evaluation module, and decision recommendation module.

6. A decision recommendation system for prioritizing key data processing in a manufacturing enterprise's supply chain, as described in claim 2, is characterized in that... The production data is collected in real time by IoT sensors deployed on the production line to monitor the equipment's operating status; the logistics data is collected by connecting to the transportation management system and the warehouse management system to obtain real-time transportation trajectory, inventory status, and abnormal events; the distributed storage architecture includes storing real-time data in a time-series database and supporting high-concurrency writing, and archiving historical data to a data lake; the caching mechanism caches frequently accessed data (such as the daily inventory status) in Redis.

7. A decision recommendation system for prioritizing key data processing in a manufacturing enterprise's supply chain, as described in claim 3, is characterized in that... The rule base establishment includes rule data structure design. Rules are defined using a structured expression, including: rule ID, triggering conditions (a Boolean expression consisting of multiple attribute variables, such as "order delay > 48 hours AND supplier rating = low"), applicable scenario tags (logistics anomaly, inventory shortage, quality complaint, etc.), priority level, and processing suggestions / instructions. Real-time rule matching extracts event attributes, parses key information from the event, including event type, timestamp, related resource (order, warehouse, supplier, etc.) identifiers, and event context parameters, forming a standard structure for rule matching. Matching methods support sequential matching, indexed matching (creating an index based on rule tags or event types), and conditional expression evaluation. Upon successful matching, the corresponding priority level and processing suggestions are immediately extracted, generating a preliminary response task. Dynamic rule loading, i.e., the rule engine unit, supports dynamically adding, modifying, and deleting rules without interrupting system operation.

8. A decision recommendation system for prioritizing key data processing in a manufacturing enterprise's supply chain, as described in claim 3, is characterized in that... The model architecture includes urgency assessment: LSTM neural network (time series feature analysis), impact assessment: random forest (multi-feature association analysis), and timeliness assessment: XGBoost (structured feature processing). The urgency assessment sub-model identifies event development trends, abnormal fluctuations, and suddenness characteristics. Input features are event evolution data constructed based on time series, including delayed time series, inventory fluctuation records, historical alarm frequencies, etc. A multi-layer LSTM (Long Short-Term Memory) network structure is used to extract time-dependent information, ultimately outputting the urgency assessment result. The impact assessment sub-model analyzes the potential impact of events on key nodes in the supply chain (including order fulfillment, customer satisfaction, inventory costs, etc.). Input features include order amount, key materials involved, supplier level, number of upstream and downstream links, customer level, etc. A multi-decision tree random ensemble is used to construct the model, ultimately outputting the impact assessment result. The timeliness assessment sub-model assesses whether an event needs to be responded to or handled within a short period. Input features include event duration, historical average processing time, resource availability, and number of concurrent events. Gradient Boosting Tree (XGBoost) is used to model the structured features, achieving high fitting accuracy. The final output is a timeliness assessment result. The outputs of the urgency assessment sub-model, the influence assessment sub-model, and the timeliness assessment sub-model are weighted and fused to form a comprehensive priority assessment result. The feature engineering module includes: a) Static feature extraction: including event type, supplier level, customer classification, product category, etc.; b. Dynamic feature extraction: including real-time event metrics (delay minutes, order backlog), abnormal growth rate, upstream and downstream response time, etc.; c. Derived feature construction: including event periodic statistical features, moving average, growth rate, Z-score anomaly score, etc.; d. Feature standardization: normalizing or standardizing continuous features to adapt to the input requirements of different models.

9. A decision recommendation system for prioritizing key data processing in a manufacturing enterprise's supply chain, as described in claim 5, is characterized in that... The target system construction submodule constructs an objective function based on enterprise operation indicators, including but not limited to minimizing processing time, minimizing cost, maximizing customer satisfaction, maximizing resource utilization balance, and maximizing event risk mitigation rate, all optimized using the Pareto front approach. The constraint modeling submodule establishes linear or nonlinear constraints based on actual supply chain business logic, including but not limited to resource capacity constraints (including manpower, logistics, and capital), time window constraints (e.g., requiring response within 48 hours), concurrent event priority conflicts, SLA (Service Level Agreement) limitations, and task indivisibility conditions (e.g., certain types of events must be processed as a whole). The multi-objective optimization unit uses a genetic algorithm to generate a set of optimized decision schemes. The adaptive feedback module's reverse empowerment includes the data fusion module updating data acquisition strategies and optimizing data quality rules; the priority evaluation module adjusting feature weights, hot updating model parameters, the decision recommendation module reconstructing the objective function, and calibrating simulation parameters.

10. A decision recommendation system for prioritizing key data processing in a manufacturing enterprise's supply chain according to claim 1, characterized in that... The aforementioned decision-making recommendation method for prioritizing key data processing in the manufacturing enterprise supply chain utilizes a decision-making recommendation system for prioritizing key data processing in the supply chain, and includes the following steps: (1) The raw data collected by the data acquisition unit is transmitted to the data preprocessing unit. The data preprocessing unit preprocesses the raw data and outputs the preprocessed data to the data storage unit for storage and management. (2) The processed data is output to the priority evaluation module, the rule engine unit and the machine learning model realize the priority evaluation of supply chain events, the multi-objective optimization unit generates an optimization decision scheme set, the scenario simulation unit simulates the candidate decision schemes, predicts the execution results, and outputs the optimization scheme with risk assessment. (3) The recommended output unit will format the optimization scheme and transform it into an actionable decision suggestion output in a visual presentation; (4) The execution feedback unit collects decision execution effect data in real time and outputs the feedback data to the model iteration unit. The model iteration unit continuously optimizes each algorithm model based on these feedback data to improve the accuracy and adaptability of the model. The knowledge base construction unit accumulates supply chain decision knowledge to provide training samples for model iteration. (5) Finally, a closed-loop process of "collection-analysis-decision-optimization" is achieved.