Group-level component cigarette management method, device, equipment and storage medium

Abstract

The application discloses a group-level piece cigarette management method, device and equipment and a storage medium, relates to the tobacco technical field, and the method generates a standardized data stream through a dynamic ETL technology, breaks data barriers, and lays a foundation for group-level data fusion. Based on this, a multi-objective optimization scheduling model is constructed, equipment utilization, inventory turnover rate and transportation cost are coordinated, scheduling decisions are changed from experience-driven to global optimization. The non-dominated sorting genetic algorithm is used to solve the model, realize intelligent collaboration and dynamic optimal allocation of resources. Through multi-dimensional data dynamic calculation, safety inventory and inventory age warning line are calculated, the accuracy and adaptability of inventory risk warning are improved. The linear regression model is used to monitor and predict the delivery progress in real time, and the process controllability is enhanced. Finally, the decision tree model is used to automatically identify abnormal orders and classify the root causes, realizing the change from post-response to proactive identification.

Description

Technical Field

[0001] This application relates to the field of tobacco technology, and in particular to a method, apparatus, equipment and storage medium for managing group-level cigarette packs. Background Technology

[0002] Currently, large-scale tobacco companies struggle to meet the demands of large-scale, refined management in the management of cigarette cartons. A significant issue is the existence of data silos in carton management, making integration difficult. Many production units and remote warehouses within the group utilize independent management systems, such as different versions of WMS, customized warehouse systems, and third-party logistics cloud platforms. Furthermore, data standards are inconsistent, with variations in inventory field definitions, document formats, and equipment status codes, resulting in fragmented storage of inventory data, inbound / outbound details, and equipment operation data. During manual verification, data is first aggregated and then checked one by one, leading to a general 4-8 hour delay in data timeliness, rendering it unreliable and inconsistent. Combined with subjective errors and mistakes in manual verification, the error rate becomes significant, hindering real-time decision-making at the group level.

[0003] Meanwhile, the allocation of sales contracts and transfer orders relies solely on inventory data from a single production unit, without taking into account factors such as actual equipment capacity, operating status, and transportation costs. This results in some production units having idle equipment and others operating beyond their capacity, leading to crude decision-making in cigarette dispatching and low resource utilization.

[0004] In summary, current technologies mostly focus on the management of single production units or local warehouses, failing to form a unified group-level architecture. This makes it difficult for group-level tobacco companies to meet the needs of large-scale and refined management in the management of cigarette packs. Summary of the Invention

[0005] The main purpose of this application is to provide a method, device, equipment and storage medium for managing cigarette packs at the group level, so as to solve the problem that group-level tobacco enterprises cannot meet the needs of large-scale and refined management in the management of cigarette packs in the prior art.

[0006] To achieve the above objectives, this application provides the following technical solution: A group-level cigarette pack management method, applicable to production units and off-site warehouses within a group, comprising: Step S1: Based on dynamic ETL, extract and transform the inventory details, equipment status, and QR code scanning data of the production unit and the remote warehouse at the field level to generate a standardized data stream; Step S2: Define at least two objective functions based on the preset management strategy, and integrate all objective functions to obtain a multi-objective optimization scheduling model; Step S3: Input the standardized data stream into the multi-objective optimization scheduling model, and solve the optimal management scheme of the multi-objective optimization scheduling model using a non-dominated sorting genetic algorithm; Step S4: Calculate safety stock and inventory age warning line based on spatiotemporal dimension, attribute dimension, and inventory age dimension using a dynamic threshold algorithm; Step S5: Monitor the delivery plan using the safety stock and the inventory age warning line, and predict the delivery progress completion time of the delivery plan using a linear regression model; Step S6: Obtain abnormal orders in the shipping plan that deviate from the shipping progress completion time, and classify the abnormal orders according to the abnormal reasons using a decision tree model to obtain a list of abnormal orders.

[0007] Beneficial effects of steps S1 to S6: This is reflected in the systematic solution of core problems pointed out in the background technology, such as data silos, extensive decision-making, and delayed response, by constructing a full-link, automated data-driven decision-making system. Specifically, step S1 uses dynamic ETL technology to achieve real-time field-level extraction and standardization transformation of heterogeneous multi-source data, generating a unified and clean standardized data stream. This breaks down data barriers between production units and remote warehouses, laying the foundation for group-level data fusion and eliminating time delays and errors caused by inconsistent data standards and manual aggregation and verification. Step S2, based on the standardized data stream, constructs a multi-objective optimization scheduling model by integrating multiple key performance indicators. It incorporates multiple factors such as equipment utilization, inventory turnover, and transportation costs into a unified mathematical framework for quantitative weighting, transforming scheduling decisions from an extensive model relying on local experience to a scientific model based on global optimization. Step S3 applies a non-dominated sorting genetic algorithm to efficiently solve the above model, automatically outputting the results among multiple competing objectives. Achieving a balanced optimal management solution enables intelligent collaboration and dynamic optimal allocation of production and logistics resources within the group. Step S4 dynamically calculates safety stock and inventory age warning lines based on multi-dimensional data, transforming inventory control thresholds from static experience values ​​to dynamic adjustments based on time, space, attributes, and inventory age, thus improving the accuracy and adaptability of inventory risk warnings. Step S5 utilizes a linear regression model for real-time monitoring and progress prediction of delivery plans, achieving quantitative tracking and forward-looking forecasting of the execution process, significantly enhancing the timeliness and controllability of the management process. Step S6 uses a decision tree model to automatically identify and classify deviating orders, generating a structured list of abnormal orders, shifting problem handling from reactive post-event response to proactive pre-event identification and attribution, improving the intelligence level of the management closed loop. In summary, the steps are interconnected, forming a complete management closed loop from data fusion, intelligent decision-making, dynamic early warning to anomaly self-healing, effectively supporting the group-level large-scale and refined management needs of cigarette packaging.

[0008] As a further improvement to this application, step S1 involves extracting and transforming the inventory details, equipment status, and QR code scanning data of the production unit and the remote warehouse at the field level based on dynamic ETL, generating a standardized data stream, including: Step S1.1: Configure the JDBC connection parameters of the production unit database and the RESTful API interface address of the remote warehouse, and set the MQTT topic of the QR code scanning device to establish a multi-source data connection; Step S1.2: Based on the preset field-level whitelist rules, the inventory details table, device status log, and QR code scanning records are extracted in parallel through multi-source data connection to obtain the raw data stream; Step S1.3: Clean the original data stream by removing invalid characters and filling in empty values ​​using regular expressions to obtain a clean data stream; Step S1.4: Standardize and transform the clean data stream based on a preset mapping rule base to obtain a unified encoded data stream; Step S1.5: Merge the unified encoded data stream into micro-batches according to time windows, and assign a unified timestamp and data source identifier to obtain several time-aligned data blocks; Step S1.6: Publish all time-aligned data blocks to a distributed message queue and write them to a group-level data pool to obtain a standardized data stream.

[0009] Beneficial effects of steps S1.1 to S1.6: An automated multi-source data integration pipeline was constructed to fundamentally address the data silos and integration challenges caused by independent systems and inconsistent standards. Specifically, step S1.1, with its unified configuration of connection parameters for heterogeneous data sources, laid the technical foundation for subsequent data extraction; step S1.2, based on preset rules, performed field-level extraction, effectively avoiding irrelevant or sensitive data and achieving accurate acquisition of core management data; step S1.3, through automated data cleaning and verification rules, significantly improved the quality and integrity of the original data; step S1.4, leveraging a configurable mapping rule base, converted the cleaned heterogeneous data into the group's standard format, resolving the fundamental problem of inconsistent data standards in the background technology; step S1.5, through time window alignment and identifier addition, ensured the temporal consistency of data from different sources, enabling subsequent correlation analysis; finally, step S1.6, utilizing message queues and data pooling technologies, buffered and integrated the processed data stream, generating a stable and reliable standardized data stream, providing a high-quality data foundation for the entire management method, thereby eliminating the need for manual aggregation and verification and providing an accurate data foundation for group-level real-time decision-making.

[0010] As a further improvement to this application, step S2 defines at least two objective functions based on a preset management strategy, and integrates all objective functions to obtain a multi-objective optimization scheduling model, including: Step S2.1: Extract historical operating time of equipment, fault records, time series data of inventory quantity changes, and transportation cost rates from the standardized data stream, and integrate them to obtain the model parameter set; Step S2.2: Based on the model parameter set, calculate the weight coefficients of equipment utilization rate, inventory turnover rate, and transportation cost using the entropy weight method to obtain the target weight vector; Step S2.3: Using the model parameter set and the target weight vector, define a first objective function that aims to maximize equipment utilization, a second objective function that aims to maximize inventory turnover, and a third objective function that aims to minimize transportation costs. Step S2.4: Integrate the first objective function, the second objective function, and the third objective function using a linear weighting method to form a comprehensive objective function; Step S2.5: Parse the real-time inventory limit and the maximum daily production capacity of each production unit from the standardized data stream and use them as constraints. Step S2.6: Integrate the comprehensive objective function and the constraints to obtain a multi-objective optimization scheduling model.

[0011] Beneficial effects of steps S2.1 to S2.6: This approach transforms dispersed management elements into a structured mathematical optimization problem to address the issue of inefficient scheduling decisions that rely on single pieces of information and lack overall coordination. Step S2.1 automatically extracts key parameters from the standardized data stream, providing an objective data foundation for model construction. Step S2.2 applies the entropy weight method to automatically calculate the weights of each management objective, avoiding biases caused by subjective weight setting and making the model objectives more reflective of actual business priorities. Step S2.3 clearly defines multiple key performance objective functions based on these weights, simultaneously incorporating equipment utilization, inventory turnover, and transportation costs into decision-making considerations, overcoming the limitations of relying on single inventory data. Step S2.4 integrates multiple objectives into a solvable comprehensive objective function using a linear weighting method, achieving synergy and trade-offs among different objectives. Step S2.5 automatically parses hard constraints such as inventory and capacity from the data, ensuring the practical feasibility of subsequent solution schemes. Finally, Step S2.6 completes the construction of the multi-objective optimization scheduling model, forming a decision framework that integrates data, objectives, and constraints, laying a core foundation for subsequent intelligent optimization scheduling, and shifting scheduling decisions from experience-driven to data-driven based on multi-factor global optimization.

[0012] As a further improvement to this application, step S3 involves inputting the standardized data stream into the multi-objective optimization scheduling model and solving for the optimal management scheme of the multi-objective optimization scheduling model using a non-dominated sorting genetic algorithm, including: Step S3.1: Define the chromosome length according to the number of production units, and randomly generate an initial population that satisfies the constraints by real number encoding; Step S3.2: Decode each chromosome in the initial population into a specific contract allocation scheme, and substitute it into the comprehensive objective function of the multi-objective optimization scheduling model to calculate the objective function value of each individual; Step S3.3: Based on the objective function value of each individual, the population is stratified using the fast non-dominated sorting algorithm to determine the frontier level of each individual; Step S3.4: Within the same frontier level, calculate the crowding distance for each individual; Step S3.5: Based on the frontier level and crowding distance of each individual, the parent population is selected by using a binary tournament selection operator. Step S3.6: Perform simulated binary crossover and polynomial mutation operations sequentially on all parent individuals in the parent population to obtain a new generation of offspring population; Step S3.7: Merge the parent population with the new generation offspring population, and repeat steps S3.3 to S3.7 with the merged population as the subject of execution. After each iteration, select a new generation population through an elite retention strategy. Repeat the iteration until the termination condition is met, and use the Pareto optimal solution set as the optimal management scheme.

[0013] Beneficial effects of steps S3.1 to S3.7: This paper proposes a systematic automatic optimization mechanism to transform a multi-objective optimization scheduling model into an executable optimal management scheme, thereby addressing the problem of low equipment utilization caused by reliance on manual experience in resource allocation and difficulty in coordinating multiple objectives in the background technology. Specifically, step S3.1 mathematically encodes the solution space and initializes the population, laying the foundation for subsequent searches; step S3.2 calculates the objective function value of each solution to quantitatively evaluate different allocation schemes; step S3.3 applies fast non-dominated sorting to stratify the schemes and identify the Pareto solution set that performs well among multiple competing objectives; step S3.4 maintains the diversity of solutions by calculating crowding distance, avoiding premature convergence to local optima; steps S3.5 and S3.6 simulate the natural evolutionary process through selection, crossover, and mutation operators, driving the population to iterate towards a better direction; and step S3.7 ensures that the algorithm efficiently and stably outputs a high-quality non-dominated solution set through an elite retention strategy and an iteration termination mechanism. The steps in this section together constitute a complete automated decision engine. Its effect is to autonomously find the best scheduling solution among multiple objectives such as equipment utilization, inventory turnover and transportation costs from a massive number of possible solutions, thereby improving decision-making from subjective and extensive experience-based judgment to objective and systematic optimization calculation.

[0014] As a further improvement to this application, step S4, based on the spatiotemporal dimension, attribute dimension, and inventory age dimension, calculates the safety stock and inventory age warning line using a dynamic threshold algorithm, including: Step S4.1: Based on the optimal contract allocation scheme, extract the associated inventory details, inbound and outbound documents, and QR code data from the standardized data stream; Step S4.2: Aggregate and group the inventory details, inbound and outbound documents, and QR code data in a spatiotemporal dimension according to a preset time period and a preset geographical level, and calculate the total inventory and change trend under each spatiotemporal unit. Step S4.3: Aggregate and group the inventory details, inbound and outbound documents, and QR code data according to the attribute dimensions based on product specifications and inventory status, and calculate the quantity and proportion of inventory in each group; Step S4.4: Calculate the inventory age of each cigarette box based on the coding timestamp in the QR code data, and associate the inventory age data with the spatiotemporal dimension aggregation grouping results and the attribute dimension aggregation grouping results to obtain a comprehensive inventory dataset including spatiotemporal dimension, attribute dimension and inventory age dimension. Step S4.5: Calculate the average sales volume and standard deviation of sales volume of the comprehensive inventory dataset, as well as the safety stock threshold for each product specification in the spatiotemporal dimension. Step S4.6: Identify inventory categories with different sales popularity based on the inventory age dimension of the comprehensive inventory dataset, and determine the inventory age coefficient for each inventory category; Step S4.7: Calculate the inventory age warning line for each inventory category based on the inventory age coefficient and the average inventory age of each inventory category.

[0015] Beneficial effects of steps S4.1 to S4.7: By establishing a multi-dimensional and dynamic inventory analysis and early warning mechanism, the problems of coarse granularity and delayed early warning in the background technology of inventory management are effectively addressed. Specifically, step S4.1 extracts correlations from multi-source inventory data, laying the data foundation for subsequent analysis; steps S4.2 and S4.3 aggregate and group data from spatiotemporal and attribute dimensions, achieving refined insights into the total inventory, distribution, and composition, changing the original extensive model that could only perform total quantity statistics; step S4.4 introduces the inventory age dimension and correlates it with preceding dimensions to construct a comprehensive inventory dataset that reflects multiple characteristics such as the spatiotemporal distribution, status attributes, and storage duration of inventory, enabling multi-angle penetrating analysis of inventory health status; step S4.5 calculates dynamically changing safety stock thresholds, allowing the setting of inventory protection levels to adapt to sales fluctuations and improving the sensitivity and accuracy of stockout early warnings; steps S4.6 and S4.7 further distinguish the turnover characteristics of different inventories and set differentiated inventory age early warning lines, achieving accurate identification and early warning of backlog risks. This section transforms static, passive inventory monitoring into a proactive early warning system based on multi-dimensional data fusion and capable of dynamically adjusting thresholds, thereby improving the precision and foresight of inventory management.

[0016] As a further improvement to this application, step S5 involves monitoring the delivery plan using the safety stock and the inventory aging warning line, and predicting the delivery completion time of the delivery plan using a linear regression model, including: Step S5.1: Obtain the total number of planned shipments, the quantity shipped, and the shipment timestamp from the standardized data stream, and associate the safety stock with the inventory age warning line to obtain the shipment plan monitoring dataset; Step S5.2: Calculate the current delivery progress execution rate based on the delivery plan monitoring dataset; Step S5.3: Using time as the independent variable and cumulative shipment volume as the dependent variable, the model is trained on the shipment plan monitoring dataset using a multiple linear regression algorithm to obtain a linear regression model that reflects the relationship between shipment rate and time. Step S5.4: Input the current time, the quantity shipped, and the current shipment progress execution rate into the linear regression model to predict the cumulative shipment volume at future time points and obtain the shipment progress prediction curve; Step S5.5: Obtain the time point corresponding to when the cumulative shipment volume reaches the shipment plan on the shipment progress prediction curve, and define the corresponding time point as the shipment progress completion time.

[0017] Beneficial effects of steps S5.1 to S5.5: A data-driven quantitative monitoring and prediction mechanism for delivery progress was constructed to directly address the shortcomings of weak delivery monitoring and lack of real-time quantitative guidance in the background technology. Specifically, step S5.1 integrates planned data and early warning thresholds to form a comprehensive monitoring dataset, providing an information foundation for real-time analysis; step S5.2 calculates the execution rate and compares it with inventory status to achieve immediate perception of delivery progress and inventory risk; step S5.3 uses historical data to train a linear regression model, transforming delivery progress prediction from empirical estimation to quantitative analysis based on statistical laws; step S5.4 applies this model to predict future delivery volumes and generate progress curves, enabling managers to proactively grasp plan execution trends; and step S5.5 objectively determines the estimated completion time of the plan from the prediction curve, providing a precise time benchmark for subsequent anomaly identification. This module elevates delivery plan management from a passive, lagging event recording state to a proactive, predictable, and dynamic process control state, enhancing the visibility and controllability of the execution process.

[0018] As a further improvement to this application, step S6 involves obtaining abnormal orders in the shipping plan that deviate from the completion time of the shipping schedule, and classifying the abnormal orders according to the cause of the abnormality using a decision tree model to obtain a list of abnormal orders, including: Step S6.1: Obtain the actual progress of the order in the real-time delivery plan and compare it with the delivery progress completion time to obtain the delivery progress deviation value; Step S6.2: Compare the shipping progress deviation value with a preset tolerance threshold, filter out orders whose shipping progress deviation value exceeds the preset tolerance threshold, and integrate them into a set of potential abnormal orders; Step S6.3: Extract real-time inventory, equipment operating status code, and transport vehicle GPS location data associated with the set of potential abnormal orders to construct an abnormal order feature vector; Step S6.4: Input the abnormal order feature vector into the pre-trained isolated forest model to calculate the abnormal score for each order; Step S6.5: Define orders with abnormal scores exceeding a preset abnormal score threshold as final abnormal orders, and extract the abnormal order feature vector corresponding to the final abnormal orders to obtain the confirmed abnormal order dataset; Step S6.6: Input the confirmed abnormal order dataset into the pre-trained decision tree classification model, and bind the abnormal cause labels output by the decision tree classification model with the corresponding final abnormal orders to obtain the abnormal order list.

[0019] Beneficial effects of steps S6.1 to S6.6: An automated closed-loop system for anomaly order identification and root cause analysis has been established, directly addressing the pain points of inefficient and inaccurate anomaly handling in the background technology, which relies on manual investigation. Specifically, step S6.1 quantifies and calculates delivery progress deviations, transforming anomaly judgment from subjective experience into objective indicators; step S6.2 performs preliminary screening based on preset thresholds, achieving rapid initial screening of anomaly orders; step S6.3 constructs feature vectors by associating multi-dimensional data, providing structured input for subsequent intelligent analysis; step S6.4 applies an isolated forest model to perform unsupervised anomaly detection on the feature vectors, effectively identifying hidden anomalies deviating from normal patterns; step S6.5 confirms the final anomaly orders based on anomaly score thresholds, improving the reliability of the detection results; finally, step S6.6 uses a decision tree model to automatically classify the confirmed anomalies, outputting clear anomaly cause labels. This module transforms the traditional anomaly handling process, which relies on manual experience, into an intelligent operation and maintenance mechanism based on data-driven and machine learning, achieving automation from anomaly perception to root cause location, significantly improving the efficiency and accuracy of problem handling.

[0020] To achieve the above objectives, this application also provides the following technical solutions: A group-level cigarette pack management device, wherein the group-level cigarette pack management device is applied to the group-level cigarette pack management method described above, and the group-level cigarette pack management device comprises: The standardized data stream acquisition module is used to extract and transform the inventory details, equipment status, and QR code scanning data of the production unit and the remote warehouse based on dynamic ETL, and generate a standardized data stream. The multi-objective optimization scheduling module is used to define at least two objective functions based on a preset management strategy and integrate all objective functions to obtain a multi-objective optimization scheduling model. The optimal management scheme calculation module is used to input the standardized data stream into the multi-objective optimization scheduling model and solve the optimal management scheme of the multi-objective optimization scheduling model through a non-dominated sorting genetic algorithm. The inventory and inventory age calculation module is used to calculate safety stock and inventory age warning line based on spatiotemporal dimensions, attribute dimensions, and inventory age dimensions using a dynamic threshold algorithm; The delivery progress completion time prediction module is used to monitor the delivery plan through the safety stock and the inventory age warning line, and to predict the delivery progress completion time of the delivery plan through a linear regression model. The abnormal order acquisition module is used to acquire abnormal orders in the shipping plan that deviate from the shipping progress completion time, and to classify the abnormal orders according to the abnormal reasons using a decision tree model to obtain an abnormal order list.

[0021] To achieve the above objectives, this application also provides the following technical solutions: An electronic device includes a processor and a memory coupled to the processor, the memory storing program instructions executable by the processor; when the processor executes the program instructions stored in the memory, it implements the group-level cigarette management method described above.

[0022] To achieve the above objectives, this application also provides the following technical solutions: A computer-readable storage medium storing program instructions, which, when executed by a processor, enable the implementation of the group-level cigarette management method described above. Attached Figure Description

[0023] Figure 1 This is a schematic flowchart illustrating the steps of one embodiment of a group-level cigarette pack management method according to this application; Figure 2 This is a schematic diagram of the functional modules of one embodiment of a group-level cigarette pack management device according to this application; Figure 3 This is a schematic diagram of the structure of an embodiment of the electronic device of this application; Figure 4 This is a schematic diagram of the structure of one embodiment of the storage medium of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0025] The terms "first," "second," and "third" in this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0026] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0027] like Figure 1 As shown, this embodiment provides an example of a group-level cigarette pack management method. In this embodiment, the group-level cigarette pack management method is applied to the group's production units and off-site warehouses.

[0028] Specifically, the management method for cigarettes at the group level includes the following steps: Step S1: Based on dynamic ETL, extract and transform the inventory details, equipment status, and QR code scanning data of production units and remote warehouses at the field level to generate a standardized data stream.

[0029] Further, step S1 involves extracting and transforming field-level data from inventory details, equipment status, and QR code scanning data of production units and remote warehouses based on dynamic ETL, generating a standardized data stream. This specifically includes the following steps: Step S1.1: Configure the JDBC connection parameters of the production unit database and the RESTful API interface address of the remote warehouse, and set the MQTT topic of the QR code scanning device to establish a multi-source data connection.

[0030] Preferably, the JDBC connection parameters are configured for each production unit's Oracle or MySQL database, and include the following: url:jdbc:mysql: / / [ip]:[port] / [database_name].

[0031] username: A dedicated account with read-only permissions.

[0032] password: The password for encrypted storage.

[0033] driver-class-name: com.mysql.cj.jdbc.Driver.

[0034] Preferably, the RESTful API configuration involves configuring the interface parameters for each remote warehouse, such as a third-party logistics cloud platform. endpoint: https: / / api.example.com / v1 / warehouse / plan.

[0035] Header: Contains authentication information, such as Authorization:Bearer[api_token].

[0036] Preferably, MQTT topic configuration: Set up subscription topics for different types of scanning devices, for example: Warehouse inbound scanning topic: / warehouse / inbound / scan Outbound scanning topic: / warehouse / outbound / scan After establishing the connection, continuously monitor the aforementioned data source to prepare for data extraction.

[0037] Step S1.2: Based on the preset field-level whitelist rules, the inventory details table, device status log, and QR code scanning records are extracted in parallel through multi-source data connection to obtain the raw data stream.

[0038] Preferably, the field-level whitelist predefines the specific fields that need to be extracted from each data source in the configuration file, avoiding full table scans and data redundancy. For example: Extract only item_id (material code), location (location number), batch_no (batch number), quantity (quantity), and update_time (update time) from the inventory details table.

[0039] Extract only device_id (device number), status (status code), and timestamp (time stamp) from the device status log.

[0040] Parallel extraction: Using multi-threading technology, data is simultaneously pulled from established JDBC connections, RESTful API endpoints, and MQTT topics.

[0041] Data encapsulation: Each extracted data item is encapsulated into a JSON object and tagged with a data source identifier (source_id).

[0042] Step S1.3: Clean the original data stream by removing invalid characters and filling in null values ​​using regular expressions to obtain a clean data stream.

[0043] Preferably, invalid character removal can be performed on string fields using predefined regular expression patterns. For example, to remove illegal characters from inventory location names, the regular expression [^a-zA-Z0-9\u4e00-\u9fa5] can be used to remove characters that are not letters, numbers, or Chinese characters.

[0044] Preferably, null values ​​can be standardized according to business rules. For example, numeric fields are filled with 0; string fields are filled with UNKNOWN; and timestamp fields are filled with the current device time from which the data was extracted.

[0045] Preferably, data integrity verification can also be performed by applying the CRC32 checksum algorithm to key fields (such as document number and QR code), calculating the checksum, and comparing it with the expected value. Data records that fail the verification will be marked and temporarily stored in the processing queue. This step ultimately yields a clean data stream that has undergone cleaning and preliminary verification.

[0046] Step S1.4: Standardize and transform the clean data stream based on the preset mapping rule base to obtain a unified encoded data stream.

[0047] Preferably, the mapping rule base is typically stored in a database or configuration file, defining the mapping relationship from source values ​​to group standard values. One commonly used mapping rule base is shown in Table 1 below: Table 1, Mapping Rule Base:

[0048] Step S1.5: Merge the unified coded data stream into micro-batches according to time windows, and assign a unified timestamp and data source identifier to obtain several time-aligned data blocks.

[0049] Preferably, the time window can be set to a fixed 5 seconds. At the end of the time window, all records cached within the window are merged into a time-aligned data block. At the same time, a unified batch timestamp (usually the window end time) and a data source type identifier are added to the data block to convert continuous streaming data into discrete batch data that is easy to process later, and to ensure that the data within the same batch is time-aligned.

[0050] Step S1.6: Publish all time-aligned data blocks to a distributed message queue and write them to a group-level data pool to obtain a standardized data stream.

[0051] Preferably, publishing to a message queue involves publishing each time-aligned data block as a message to a distributed message queue such as Apache Kafka, and the topic can be named standardized_data.

[0052] Preferably, the data pool is written to by downstream data consumption services that subscribe to and consume these messages from Kafka topics and persist them to a group-level central data pool, which can be HDFS, a data lake, or a relational database.

[0053] Beneficial effects of steps S1.1 to S1.6: An automated multi-source data integration pipeline was constructed to fundamentally address the data silos and integration challenges caused by independent systems and inconsistent standards. Specifically, step S1.1, with its unified configuration of connection parameters for heterogeneous data sources, laid the technical foundation for subsequent data extraction; step S1.2, based on preset rules, performed field-level extraction, effectively avoiding irrelevant or sensitive data and achieving accurate acquisition of core management data; step S1.3, through automated data cleaning and verification rules, significantly improved the quality and integrity of the original data; step S1.4, leveraging a configurable mapping rule base, converted the cleaned heterogeneous data into the group's standard format, resolving the fundamental problem of inconsistent data standards in the background technology; step S1.5, through time window alignment and identifier addition, ensured the temporal consistency of data from different sources, enabling subsequent correlation analysis; finally, step S1.6, utilizing message queues and data pooling technologies, buffered and integrated the processed data stream, generating a stable and reliable standardized data stream, providing a high-quality data foundation for the entire management method, thereby eliminating the need for manual aggregation and verification and providing an accurate data foundation for group-level real-time decision-making.

[0054] Step S2: Define at least two objective functions based on the preset management strategy, and integrate all objective functions to obtain a multi-objective optimization scheduling model.

[0055] Further, step S2 involves defining at least two objective functions based on a preset management strategy, and integrating all objective functions to obtain a multi-objective optimization scheduling model, specifically including the following steps: Step S2.1: Extract historical operating time of equipment, fault records, time series data of inventory quantity changes, and transportation cost rates from the standardized data stream, and integrate them to obtain the model parameter set.

[0056] Preferably, the equipment utilization parameters can be calculated for each production unit i by taking the average daily effective operating time T_ij_avg and the number of failures F_ij of its equipment j over the past 30 days; the maximum daily production capacity C_i is determined by combining the equipment nameplate parameters with the historical peak efficiency.

[0057] Preferably, the inventory turnover rate parameter can be obtained by querying the current inventory quantity S_ik for each product k of each unit i; the historical average daily outbound quantity D_ik_avg is calculated by statistically analyzing the outbound data of the past 90 days.

[0058] Preferably, the transportation cost-related parameters can be obtained from the master data, such as the average distance Dist_iz (km) from each unit i to each major customer region z, and the transportation cost rate P (yuan / piece·km) per unit distance per product.

[0059] In summary, all extracted and calculated parameters are integrated into a structured model parameter set Θ. This set can be represented as: Θ={C_i,S_ik,D_ik_avg,Dist_iz,P,...}foralli,k,z. This parameter set Θ serves as the foundational data for constructing the mathematical model.

[0060] Step S2.2: Based on the model parameter set, calculate the weight coefficients of equipment utilization rate, inventory turnover rate and transportation cost using the entropy weight method to obtain the target weight vector.

[0061] Preferably, the entropy weighting method is an objective weighting method that determines the weights based on the degree of variation in the data of each indicator, thus avoiding subjectivity. The specific steps are as follows: ① Construct the evaluation matrix: Assume there are m production units to be evaluated, and n = 3 evaluation indicators (equipment utilization rate, inventory turnover rate, and the reciprocal of transportation cost). Form an m×n matrix X. It is worth noting that transportation cost is a cost-based indicator and needs to be converted into a benefit-based indicator, usually by taking its reciprocal.

[0062] ②Standardized matrix: Standardize matrix X to obtain standardized matrix Y.

[0063] ③ Calculate the entropy value: Calculate the entropy value of the j-th index e_j=-kΣ(p_ij*ln(p_ij)), where p_ij=Y_ij / ΣY_ij,k=1 / ln(m).

[0064] ④ Calculate the weight: Calculate the weight of the j-th indicator w_j=(1-e_j) / Σ(1-e_j).

[0065] ⑤ Output weight vector: The final target weight vector is obtained as W=[w_utilization,w_turnover,w_cost].

[0066] Among them w_utilization+w_turnover+w_cost=1.

[0067] Step S2.3: Using the model parameter set and the target weight vector, define the first objective function with the goal of maximizing equipment utilization, the second objective function with the goal of maximizing inventory turnover, and the third objective function with the goal of minimizing transportation costs.

[0068] Preferably, the first objective function is to maximize equipment utilization, which aims to maximize the average equipment utilization of all production units within the group, as represented by the following formula: .

[0069] in, The total number of production units; This represents the total number of product specifications. Let k be the decision variable, representing the quantity of product k allocated to production unit i; The maximum daily production capacity of production unit i; The equipment utilization rate of production unit i.

[0070] Preferably, the second objective function is to maximize inventory turnover rate. This objective aims to optimize resource utilization efficiency by promoting rapid inventory turnover, as represented by the following formula: .

[0071] in, The current inventory of product k for production unit i; The ratio represents the inventory turnover rate of production unit i; the larger the ratio, the faster the turnover.

[0072] Preferably, the third objective function is to minimize transportation costs. This objective aims to minimize the total transportation costs incurred in completing the contract delivery task, as represented by the following formula: .

[0073] in, The total number of customers or delivery areas; The quantity of products shipped from production unit i to region z; Let i be the average transportation distance from production unit i to region z. This refers to the transportation cost rate per kilometer per unit of product.

[0074] Step S2.4: Integrate the first objective function, the second objective function, and the third objective function using a linear weighting method to form a comprehensive objective function.

[0075] Preferably, the linear weighted summation can be achieved by using the weight vector obtained in step S2.2 to integrate the three objective functions into a single comprehensive objective function F for solution.

[0076] F = Max[w_utilization*f_utilization+w_turnover*f_turnover+w_cost*(1-f_cost_normalized)]. It is worth noting that, to ensure uniformity of dimensions and that both functions aim to maximize performance, the cost function f_cost must first be normalized and transformed into a benefit-type indicator, such as 1-f_cost / f_cost_max.

[0077] Step S2.5: Parse the real-time inventory limit and the maximum daily production capacity of each production unit from the standardized data stream and use them as constraints.

[0078] Preferably, the constraints can be set as follows: Inventory constraint: For each unit i of each product k, the allocation quantity cannot exceed its current inventory. X_ik ≤ S_ik.

[0079] Capacity constraint: For each unit i, the total allocation cannot exceed its maximum daily capacity. Σ_kX_ik≤C_i.

[0080] Demand constraint: The total allocation of all units must equal the total order demand D_k. Σ_iX_ik=D_k.

[0081] Nonnegativity constraint: The allocation must be a nonnegative number. X_ik ≥ 0.

[0082] Step S2.6: Integrate the comprehensive objective function and constraints to obtain the multi-objective optimization scheduling model.

[0083] Beneficial effects of steps S2.1 to S2.6: This approach transforms dispersed management elements into a structured mathematical optimization problem to address the issue of inefficient scheduling decisions that rely on single pieces of information and lack overall coordination. Step S2.1 automatically extracts key parameters from the standardized data stream, providing an objective data foundation for model construction. Step S2.2 applies the entropy weight method to automatically calculate the weights of each management objective, avoiding biases caused by subjective weight setting and making the model objectives more reflective of actual business priorities. Step S2.3 clearly defines multiple key performance objective functions based on these weights, simultaneously incorporating equipment utilization, inventory turnover, and transportation costs into decision-making considerations, overcoming the limitations of relying on single inventory data. Step S2.4 integrates multiple objectives into a solvable comprehensive objective function using a linear weighting method, achieving synergy and trade-offs among different objectives. Step S2.5 automatically parses hard constraints such as inventory and capacity from the data, ensuring the practical feasibility of subsequent solution schemes. Finally, Step S2.6 completes the construction of the multi-objective optimization scheduling model, forming a decision framework that integrates data, objectives, and constraints, laying a core foundation for subsequent intelligent optimization scheduling, and shifting scheduling decisions from experience-driven to data-driven based on multi-factor global optimization.

[0084] Step S3: Input the standardized data stream into the multi-objective optimization scheduling model, and solve the optimal management scheme of the multi-objective optimization scheduling model through the non-dominated sorting genetic algorithm.

[0085] Further, step S3 involves inputting the standardized data stream into the multi-objective optimization scheduling model and solving for the optimal management scheme of the multi-objective optimization scheduling model using a non-dominated sorting genetic algorithm. This specifically includes the following steps: Step S3.1: Define the chromosome length based on the number of production units, and randomly generate an initial population that meets the constraints using real number encoding.

[0086] Preferably, for a scheduling problem with M production units and K types of products, a chromosome directly represents a possible scheduling scheme. Real number encoding is used, and the chromosome length is M×K. Each gene position is represented by a real number representing the quantity Xik of a specific product k allocated to a specific production unit i. For example (M=2, K=1): [X11, X21].

[0087] Preferably, for the initial population generation, N individuals (chromosomes) can be randomly generated to form the initial population. Each individual satisfies the constraints defined in step S2.5.

[0088] Preferably, the population size N is typically set to 50 to 100.

[0089] Step S3.2: Decode each chromosome in the initial population into a specific contract allocation scheme, and substitute it into the comprehensive objective function of the multi-objective optimization scheduling model to calculate the objective function value of each individual.

[0090] Preferably, the decoding process can directly map chromosome gene values ​​to allocation amounts Xik.

[0091] Preferably, the objective function value can be calculated by substituting Xik into the three objective functions f1 (equipment utilization rate), f2 (inventory turnover rate), and f3 (transportation cost) defined in step S2.3 to obtain three objective values. Since NSGA-II directly processes multiple objectives, weighted synthesis is not required here. The objective value of each individual i can be represented as a vector Fi=(f1i,f2i,f3i).

[0092] Step S3.3: Based on the objective function value of each individual, the population is stratified using a fast non-dominated sorting algorithm to determine the frontier level of each individual.

[0093] Preferably, the fast non-dominated sort can be implemented through the following process: ① Domination relationship comparison: For each individual p in the population, calculate the set of individuals Sp that it dominates and the number of individuals np that dominate it.

[0094] ② Definition: Individual p dominates individual q if and only if p is no worse than q in all objectives and is strictly better than q in at least one objective.

[0095] ③ Determine the first frontier: All individuals with np=0 are classified as the first non-dominated frontier (Front1) and assigned a level of 1.

[0096] ④ Determine the next frontier: For each individual p in Front1, examine each individual q in the set of individuals Sp that it dominates, and perform nq = nq−1. If nq decreases to 0, then q is assigned to the next frontier (Front2). Repeat this process until all individuals are stratified. The smaller the frontier rank number, the better the solution.

[0097] Step S3.4: Within the same frontier level, calculate the crowding distance for each individual.

[0098] Preferably, for each individual within the frontier, they are sorted according to each objective function value. The crowding distance Im(i) of individual i on objective m is the normalized absolute value of the difference between the function values ​​of its two neighboring individuals (i-1 and i+1) on that objective.

[0099] Preferably, the total crowding distance of individual i is the sum of the crowding distances on all its targets. The crowding distance is used to measure the density of an individual on its frontier; the larger the distance, the sparser the solutions around the individual and the better the diversity.

[0100] Step S3.5: Based on the frontier level and crowding distance of each individual, the parent population is selected by using a binary tournament selection operator.

[0101] Preferably, the specific implementation process of the binary tournament is as follows: ① Randomly select two individuals from the population.

[0102] ② Compare two individuals: If they are at different frontier levels, choose the individual with the smaller level number (better frontier); if they are at the same frontier level, choose the individual with a larger crowding distance.

[0103] ③ Repeat steps ① to ② until a parent population of size N is selected.

[0104] Step S3.6: Perform simulated binary crossover and polynomial mutation operations on all parent individuals in the parent population in sequence to obtain the new generation of offspring population.

[0105] Preferably, simulated binary crossover is used for real number encoding. For a selected pair of parent numbers p1, p2, crossover is performed according to the crossover probability p. c (Usually set to 0.8-0.9) Perform crossover to produce two offspring c1 and c2: c1=0.5*[(1+β)p1+(1−β)p2].

[0106] c2=0.5*[(1−β)p1+(1+β)p2].

[0107] Where β is a distribution index η c Related random variables. SBX can produce offspring similar to its parents and has good convergence.

[0108] Preferably, polynomial mutation can be performed on the offspring individuals after crossover according to the mutation probability p. m To mutate, p m It is usually set to 1 / chromosome length to introduce new genes.

[0109] Where c′=c+δ∗(upper_bound−lower_bound), δ is a distribution index η. m Related small perturbations. Polynomial mutations help explore new regions of the search space.

[0110] Step S3.7: Merge the parent population with the new generation of offspring population, and repeat steps S3.3 to S3.7 with the merged population as the subject of execution. After each iteration, select the new generation of population through the elite retention strategy. Repeat the iteration until the termination condition is met, and use the Pareto optimal solution set as the optimal management scheme.

[0111] Preferably, merging and elite selection can merge the parent population (size N) and the offspring population (size N) into a mixed population of size 2N.

[0112] Preferably, the new generation population selection can re-execute the fast non-dominated sorting (step S3.3) and crowding calculation (step S3.4) on the merged 2N individuals. Then, individuals with the best frontier are preferentially selected to enter the new population. If the individuals at a certain frontier cannot be fully accommodated, individuals within that frontier are selected from largest to smallest based on crowding distance, until the new generation population of size N is filled. This is the core of the elite preservation strategy of NSGA-II.

[0113] Preferably, the termination condition is typically set to a maximum number of iterations, such as 500 generations, or to stop when the optimal frontier has not improved significantly for several consecutive generations.

[0114] Preferably, when the algorithm terminates, the first non-dominated frontier (Pareto optimal solution set) is the set of optimal management schemes sought.

[0115] Beneficial effects of steps S3.1 to S3.7: This paper proposes a systematic automatic optimization mechanism to transform a multi-objective optimization scheduling model into an executable optimal management scheme, thereby addressing the problem of low equipment utilization caused by reliance on manual experience in resource allocation and difficulty in coordinating multiple objectives in the background technology. Specifically, step S3.1 mathematically encodes the solution space and initializes the population, laying the foundation for subsequent searches; step S3.2 calculates the objective function value of each solution to quantitatively evaluate different allocation schemes; step S3.3 applies fast non-dominated sorting to stratify the schemes and identify the Pareto solution set that performs well among multiple competing objectives; step S3.4 maintains the diversity of solutions by calculating crowding distance, avoiding premature convergence to local optima; steps S3.5 and S3.6 simulate the natural evolutionary process through selection, crossover, and mutation operators, driving the population to iterate towards a better direction; and step S3.7 ensures that the algorithm efficiently and stably outputs a high-quality non-dominated solution set through an elite retention strategy and an iteration termination mechanism. The steps in this section together constitute a complete automated decision engine. Its effect is to autonomously find the scheduling scheme that achieves the best balance among multiple objectives such as equipment utilization, inventory turnover and transportation costs from a massive number of possible solutions, thereby elevating decision-making from subjective and extensive experience-based judgment to objective and systematic optimization calculation.

[0116] Step S4: Calculate safety stock and inventory age warning line based on spatiotemporal dimension, attribute dimension, and inventory age dimension using a dynamic threshold algorithm.

[0117] Further, step S4, based on the spatiotemporal dimension, attribute dimension, and inventory age dimension, calculates the safety stock and inventory age warning line using a dynamic threshold algorithm, specifically including the following steps: Step S4.1: Extract associated inventory details, inbound and outbound documents, and QR code data from the standardized data stream based on the optimal contract allocation scheme.

[0118] Preferably, based on the optimal contract allocation scheme obtained in step S3, the relevant production units, products, and warehouses can be located. Subsequently, all data records related to these entities are extracted from the standardized data stream. For example, the following SQL query statement: SELECT inventory_detail.*, in_out_order.*, qr_data.* FROM standardized_data_stream WHERE (production_unit IN (list of units involved in the plan)) AND (product_sku IN (list of products involved in the solution)) AND (warehouse_id IN (list of warehouses involved in the solution)) AND (data_type IN ('inventory','order','qr_code')) Preferably, the inventory details include material code, location, batch, current inventory quantity, and inventory status (e.g., qualified, frozen); the inbound / outbound documents include document number, operation time, product, quantity, and related business order number (e.g., sales order, transfer order); the QR code data includes the unique code of each cigarette, coding time, associated material code, most recent scan timestamp, and location.

[0119] Step S4.2: Aggregate and group inventory details, inbound and outbound documents, and QR code data in a spatiotemporal dimension according to the preset time period and preset geographical level, and calculate the total inventory and change trend under each spatiotemporal unit.

[0120] Preferably, the spatiotemporal dimension is defined as follows: Time period: Configurable, such as by day, week, month, quarter, year. Supports sliding time window calculation, such as "inventory changes in the last 7 days".

[0121] Geographic hierarchy: configurable, such as by group, subsidiary, specific production unit, warehouse, storage area, or even storage location.

[0122] Preferably, the basic dataset can be aggregated by time and geographic hierarchy using SQL's GROUP BY and OLAP functions. The aggregation metric is the total inventory: SUM(current_quantity); the inventory change trend is calculated by month-on-month and year-on-year comparisons, or by using simple linear regression to fit the slope of the inventory change over time within this spatiotemporal unit; the output is an inventory snapshot and trend metric for each spatiotemporal unit (e.g., "Warehouse A - Month Y, 20XX").

[0123] Step S4.3: Aggregate and group inventory details, inbound and outbound documents, and QR code data by attribute dimension according to product specifications and inventory status, and calculate the quantity and proportion of inventory in each group.

[0124] Preferably, the attribute dimensions are defined as follows: Product specifications: Grouped by product attributes such as brand, specifications, and model.

[0125] Inventory status: Grouped by status such as "Pending Inspection", "Qualified", "Unqualified", and "Frozen".

[0126] Preferably, the GROUP BY method can also be used to aggregate the basic dataset by product specifications and inventory status. The aggregation metric is the inventory quantity within the group: SUM(current_quantity); the percentage within the group: (inventory quantity within the group / total inventory) * 100%; output the inventory quantity and percentage for each attribute group (e.g., "brand-specification-qualified product").

[0127] Step S4.4: Calculate the inventory age of each cigarette box based on the coding timestamp in the QR code data, and associate the inventory age data with the spatiotemporal dimension aggregation grouping results and the attribute dimension aggregation grouping results to obtain a comprehensive inventory dataset including spatiotemporal dimension, attribute dimension and inventory age dimension.

[0128] Preferably, the inventory age = current device time - cigarette barcode printing time (parsed from QR code data). The inventory age unit can be days.

[0129] For example, age_in_days=DATEDIFF(CURRENT_DATE,qr_code.print_time).

[0130] Preferably, data association can be achieved by using key fields such as material code, storage location / warehouse, and batch number to associate the inventory age information at the unit cigarette level with the spatiotemporal aggregation data in step S4.2 and the attribute aggregation data in step S4.3 (JOIN). After association, each inventory unit (such as a product in a certain batch at a certain storage location) will have spatiotemporal, attribute, and inventory age information.

[0131] Step S4.5: Calculate the average sales volume and standard deviation of sales volume of the comprehensive inventory dataset, as well as the safety stock threshold for each product specification in the spatiotemporal dimension.

[0132] Preferably, dynamic safety stock calculation can be performed through... Calculate, where, This is the safety stock threshold; The average demand is usually replaced by the average daily outbound volume, which is calculated based on historical outbound data. The standard deviation of daily outbound volume; The lead time is usually considered a constant, such as 1 day; The safety factor is related to the expected service level. The coefficient k can be dynamically adjusted according to seasonality or sales forecasts. For example, during the peak sales season, k=1.65 (corresponding to approximately 95% service level), and during the off-season, k=1.28 (corresponding to approximately 90% service level).

[0133] Preferably, a dynamic safety stock threshold can be calculated for each product specification at a specific time period and geographical level (e.g., "Warehouse A - Product X - Month").

[0134] Step S4.6: Identify inventory categories with different sales popularity based on the inventory age dimension of the comprehensive inventory dataset, and determine the inventory age coefficient for each inventory category.

[0135] Preferably, sales popularity is categorized as follows: Feature extraction: For each product specification, calculate its recent (e.g., within 30 days) sales frequency and sales volume.

[0136] Cluster analysis: Using the K-means clustering algorithm, based on sales frequency and quantity, the inventory of all product specifications is automatically divided into several categories, such as "high-demand, fast-turnover", "medium-demand, average-selling", and "low-demand, slow-turnover".

[0137] Preferably, the storage age coefficient can be determined by assigning a storage age coefficient R to each cluster. For example, R = 0.5 for high-popularity clusters (tolerating shorter storage age) and R = 0.2 for low-popularity clusters (tolerating longer storage age). The coefficient can be calibrated based on historical turnover data.

[0138] Step S4.7: Calculate the inventory age warning line for each inventory category based on the inventory age coefficient and the average inventory age of each inventory category.

[0139] Preferably, the dynamic inventory age warning line can be calculated through... The calculation yielded the following result. This is the warehouse age warning line; This represents the historical average inventory age of products under this inventory category.

[0140] Beneficial effects of steps S4.1 to S4.7: By establishing a multi-dimensional and dynamic inventory analysis and early warning mechanism, the problems of coarse granularity and delayed early warning in the background technology of inventory management are effectively addressed. Specifically, step S4.1 extracts correlations from multi-source inventory data, laying the data foundation for subsequent analysis; steps S4.2 and S4.3 aggregate and group data from spatiotemporal and attribute dimensions, achieving refined insights into the total inventory, distribution, and composition, changing the original extensive model that could only perform total quantity statistics; step S4.4 introduces the inventory age dimension and correlates it with preceding dimensions to construct a comprehensive inventory dataset that reflects multiple characteristics such as the spatiotemporal distribution, status attributes, and storage duration of inventory, enabling multi-angle penetrating analysis of inventory health status; step S4.5 calculates dynamically changing safety stock thresholds, allowing the setting of inventory protection levels to adapt to sales fluctuations and improving the sensitivity and accuracy of stockout early warnings; steps S4.6 and S4.7 further distinguish the turnover characteristics of different inventories and set differentiated inventory age early warning lines, achieving accurate identification and early warning of backlog risks. This section transforms static, passive inventory monitoring into a proactive early warning system based on multi-dimensional data fusion and capable of dynamically adjusting thresholds, thereby improving the precision and foresight of inventory management.

[0141] Step S5 involves monitoring the delivery plan using safety stock and inventory age warning lines, and predicting the delivery completion time of the delivery plan using a linear regression model.

[0142] Further, step S5 involves monitoring the delivery plan through safety stock and inventory aging warning lines, and predicting the delivery completion time of the delivery plan using a linear regression model. This specifically includes the following steps: Step S5.1: Obtain the total number of planned shipments, the quantity shipped, and the shipment timestamp from the standardized data stream, and associate them with safety stock and inventory age warning lines to obtain the shipment plan monitoring dataset.

[0143] Preferably, data extraction and correlation can be achieved by continuously subscribing to a standardized data stream and filtering out data records related to the shipping plan. For example: Plan Information Extraction: Parse the shipping plan document to obtain the plan's unique identifier (plan_id), product specification (sku), planned total shipment volume (Q_total), and plan start and end times.

[0144] Real-time aggregation of progress information: Based on the QR code data of outbound shipments, grouped by plan_id and sku, calculate the number of shipments Q_shipped in real time, and record the timestamp of the most recent shipment last_shipment_time.

[0145] Inventory status association: Using the SKU and warehouse number, associate the real-time safety stock threshold SS and the inventory age warning line L_w calculated in step S4. Simultaneously, query the current real-time inventory quantity (inventory_current) and average inventory age (inventory_age_avg) for that SKU.

[0146] Preferably, the dataset can be constructed by integrating the above information into time-series records to form a shipping plan monitoring dataset. Each record contains fields such as: [timestamp, plan_id, sku, Q_total, Q_shipped, last_shipment_time, SS, L_w, inventory_current, inventory_age_avg]. This dataset serves as the basis for subsequent analysis and prediction.

[0147] Step S5.2: Calculate the current delivery progress execution rate based on the delivery plan monitoring dataset.

[0148] Preferably, for each shipping plan (plan_id), at a specific time point t, its current execution rate R_t is calculated. The formula is R_t=(Q_shipped_t / Q_total)*100%. The calculation frequency can be triggered (e.g., whenever new goods are shipped out) or performed at fixed time intervals (e.g., every 15 minutes).

[0149] Preferably, the status flag can simultaneously perform real-time risk assessment based on associated inventory data: If inventory_current ≤ SS, mark the "Inventory Warning" status.

[0150] If inventory_age_avg ≥ L_w, mark the status as "backlog warning".

[0151] Obtain the progress index R_t and inventory status marker for each plan at time point t.

[0152] Step S5.3: Using time as the independent variable and cumulative shipment volume as the dependent variable, train the model on the shipment plan monitoring dataset using a multiple linear regression algorithm to obtain a linear regression model that reflects the relationship between shipment rate and time.

[0153] Preferably, a multiple linear regression model can be used, with the independent variables being: time_index: Sequential time units (such as hours or days) starting from the start of the plan.

[0154] inventory_status: Inventory status (e.g., Normal = 0, Warning = 1).

[0155] equipment_efficiency: The current effective operating rate of the equipment (from equipment status data).

[0156] The dependent variable is: cumulative shipments Q_shipped.

[0157] Preferably, the model training process is as follows: Data preparation: Use historically completed, similar shipping plan monitoring data as the training set.

[0158] Model fitting: The regression coefficients were solved by the least squares method to obtain the linear regression equation Q_shipped_predicted=β0+β1*time_index+β2*inventory_status+β3*equipment_efficiency.

[0159] Step S5.4: Input the current time, the quantity shipped, and the current shipment progress execution rate into the linear regression model to predict the cumulative shipment volume at future time points and obtain the shipment progress prediction curve.

[0160] Preferably, the feature values ​​at the current time point t can be input into the trained model. Where time_index_t = current time - planned start time; inventory_status_t = current inventory status (numerical); equipment_efficiency_t = current equipment efficiency.

[0161] Preferably, in the prediction generation stage, the model outputs the predicted cumulative shipment volume Q_shipped_predicted for multiple future time points (such as t+1, t+2, ...).

[0162] Preferably, during the curve plotting process, the aforementioned prediction points are connected to form a delivery progress prediction curve extending from the current state to the future. This curve visually illustrates the expected path for completing the plan under current conditions.

[0163] Step S5.5: Obtain the time point corresponding to when the cumulative shipment volume reaches the shipment plan on the shipment progress prediction curve, and define the corresponding time point as the shipment progress completion time.

[0164] Preferably, the time point T_finish on the prediction curve can be found where Q_shipped_predicted = Q_total.

[0165] Beneficial effects of steps S5.1 to S5.5: A data-driven quantitative monitoring and prediction mechanism for delivery progress was constructed to directly address the shortcomings of weak delivery monitoring and lack of real-time quantitative guidance in the background technology. Specifically, step S5.1 integrates planned data and early warning thresholds to form a comprehensive monitoring dataset, providing an information foundation for real-time analysis; step S5.2 calculates the execution rate and compares it with inventory status to achieve immediate perception of delivery progress and inventory risk; step S5.3 uses historical data to train a linear regression model, transforming delivery progress prediction from empirical estimation to quantitative analysis based on statistical laws; step S5.4 applies this model to predict future delivery volumes and generate progress curves, enabling managers to proactively grasp plan execution trends; and step S5.5 objectively determines the estimated completion time of the plan from the prediction curve, providing a precise time benchmark for subsequent anomaly identification. This module elevates delivery plan management from a passive, lagging event recording state to a proactive, predictable, and dynamic process control state, enhancing the visibility and controllability of the execution process.

[0166] Step S6: Obtain abnormal orders in the shipping plan that deviate from the shipping progress completion time, and classify the abnormal orders according to the abnormal reasons using a decision tree model to obtain a list of abnormal orders.

[0167] Further, step S6 involves obtaining abnormal orders in the shipping plan that deviate from the delivery schedule completion time, and classifying the abnormal orders according to the cause of the abnormality using a decision tree model to obtain a list of abnormal orders. This specifically includes the following steps: Step S6.1: Obtain the actual progress of the order in the real-time delivery plan and compare it with the delivery progress completion time to obtain the delivery progress deviation value.

[0168] Preferably, during data synchronization and comparison, the actual execution progress of each order (or delivery plan sub-task) can be obtained in real time through the system.

[0169] Preferably, the actual progress calculation can be based on the barcode scanning data to calculate the cumulative shipment volume Q_actual for each order.

[0170] Preferably, the planned progress calculation can be based on the predicted delivery completion time T_predicted in step S5.5 and the linear regression model to infer the planned cumulative delivery volume Q_planned_at_t at the current time point t.

[0171] Preferably, the deviation calculation involves calculating both absolute and relative deviations. The absolute deviation ΔQ = Q_actual - Q_planned_at_t; the relative deviation (schedule deviation) Deviation = (Q_actual - Q_planned_at_t) / Q_planned_at_t * 100%. This step calculates a quantified delivery schedule deviation value for each order, with positive values ​​indicating ahead of schedule and negative values ​​indicating behind schedule.

[0172] Step S6.2: Compare the shipping progress deviation value with the preset tolerance threshold, filter out orders whose shipping progress deviation value exceeds the preset tolerance threshold, and integrate them into a set of potential abnormal orders.

[0173] Preferably, upper and lower tolerance thresholds can be set.

[0174] For example, lower_threshold=-15%, upper_threshold=+10%. This means that a delay of more than 15% or a progress of more than 10% is considered a significant deviation.

[0175] Step S6.3: Extract real-time inventory, equipment operating status codes, and transport vehicle GPS location data associated with the set of potential abnormal orders to construct an abnormal order feature vector.

[0176] Preferably, multi-source feature association can extract multi-dimensional features from each associated link for each order in the potentially abnormal order set. Among these, inventory features include current_inventory_level (real-time inventory level) and inventory_status (whether it is below safety stock); equipment features include equipment_status_code (running / failure / standby code), downtime_duration (duration of the current failure), and historical_failure_rate (historical failure rate); transportation features include vehicle_gps_latest_timestamp (latest GPS time), distance_from_destination (distance from destination), and estimated_delay (delay based on GPS trajectory prediction); and order-specific features include deviation_value (deviation value itself) and order_priority (order priority).

[0177] Preferably, vectorization can be achieved by normalizing all feature values ​​of each order and combining them into a fixed-dimensional numerical array, i.e., the abnormal order feature vector. For example: [deviation, inventory_level, downtime_duration, distance, ...].

[0178] Step S6.4: Input the abnormal order feature vector into the pre-trained isolated forest model to calculate the abnormal score for each order.

[0179] Preferably, the feature vector is input into the pre-trained isolated forest model. The pre-training of the isolated forest model can also be performed using the same type of historical data. The model calculates an anomaly score for each order. The anomaly score is usually between 0 and 1. The closer the score is to 1, the greater the probability that the order is an anomaly.

[0180] Preferably, the judgment logic can set an anomaly score threshold, such as 0.7. If the anomaly score > 0.7, the order is judged as statistically abnormal.

[0181] Step S6.5: Define orders with abnormal scores exceeding a preset abnormal score threshold as final abnormal orders, and extract the abnormal order feature vectors corresponding to the final abnormal orders to obtain the confirmed abnormal order dataset.

[0182] Preferably, the final judgment is based on the objective score of the isolated forest to avoid misjudgment based on a single bias value. For example, the Python statement `confirmed_anomalies = [order for order in potential_anomalies iforder.anomaly_score>threshold]`.

[0183] Step S6.6: Input the confirmed abnormal order dataset into the pre-trained decision tree classification model, and bind the abnormal reason labels output by the decision tree classification model with the corresponding final abnormal orders to obtain the abnormal order list.

[0184] Preferably, the classification of anomaly causes can be performed using the C4.5 decision tree classification model for cause analysis. This C4.5 decision tree classification model has also been trained with historical data (feature vectors + manually labeled root causes) and is able to learn the mapping rules between features and causes.

[0185] Preferably, the classification process involves inputting the feature vectors from the confirmed abnormal order dataset into a decision tree model. The model predicts the most likely cause of the abnormality for each order based on internal splitting rules (e.g., classifying an order as "equipment failure" if downtime_duration > 2 hours and inventory_level > safety stock). Common labels include: equipment failure, insufficient inventory, shipping delay, data asynchrony, etc.

[0186] Preferably, the device binds basic order information (ID, product, quantity), deviation details, anomaly scores, and predicted root cause tags. This ultimately generates a structured, sortable, and filterable list of abnormal orders. This list can be directly pushed to the operations or management platform to guide subsequent processing priorities and measures.

[0187] Beneficial effects of steps S6.1 to S6.6: An automated closed-loop system for anomaly order identification and root cause analysis has been established, directly addressing the pain points of inefficient and inaccurate anomaly handling in the background technology, which relies on manual investigation. Specifically, step S6.1 quantifies and calculates delivery progress deviations, transforming anomaly judgment from subjective experience into objective indicators; step S6.2 performs preliminary screening based on preset thresholds, achieving rapid initial screening of anomaly orders; step S6.3 constructs feature vectors by associating multi-dimensional data, providing structured input for subsequent intelligent analysis; step S6.4 applies an isolated forest model to perform unsupervised anomaly detection on the feature vectors, effectively identifying hidden anomalies deviating from normal patterns; step S6.5 confirms the final anomaly orders based on anomaly score thresholds, improving the reliability of the detection results; finally, step S6.6 uses a decision tree model to automatically classify the confirmed anomalies, outputting clear anomaly cause labels. This module transforms the traditional anomaly handling process, which relies on manual experience, into an intelligent operation and maintenance mechanism based on data-driven and machine learning, achieving automation from anomaly perception to root cause location, significantly improving the efficiency and accuracy of problem handling.

[0188] Beneficial effects of steps S1 to S6: This is reflected in the systematic solution of core problems pointed out in the background technology, such as data silos, extensive decision-making, and delayed response, by constructing a full-link, automated data-driven decision-making system. Specifically, step S1 uses dynamic ETL technology to achieve real-time field-level extraction and standardization transformation of heterogeneous multi-source data, generating a unified and clean standardized data stream. This breaks down data barriers between production units and remote warehouses, laying the foundation for group-level data fusion and eliminating time delays and errors caused by inconsistent data standards and manual aggregation and verification. Step S2, based on the standardized data stream, constructs a multi-objective optimization scheduling model by integrating multiple key performance indicators. It incorporates multiple factors such as equipment utilization, inventory turnover, and transportation costs into a unified mathematical framework for quantitative weighting, transforming scheduling decisions from an extensive model relying on local experience to a scientific model based on global optimization. Step S3 applies a non-dominated sorting genetic algorithm to efficiently solve the above model, automatically outputting the results among multiple competing objectives. Achieving a balanced optimal management solution enables intelligent collaboration and dynamic optimal allocation of production and logistics resources within the group. Step S4 dynamically calculates safety stock and inventory age warning lines based on multi-dimensional data, transforming inventory control thresholds from static experience values ​​to dynamic adjustments based on time, space, attributes, and inventory age, thus improving the accuracy and adaptability of inventory risk warnings. Step S5 utilizes a linear regression model for real-time monitoring and progress prediction of delivery plans, achieving quantitative tracking and forward-looking forecasting of the execution process, significantly enhancing the timeliness and controllability of the management process. Step S6 uses a decision tree model to automatically identify and classify deviating orders, generating a structured list of abnormal orders, shifting problem handling from reactive post-event response to proactive pre-event identification and attribution, improving the intelligence level of the management closed loop. In summary, the steps are interconnected, forming a complete management closed loop from data fusion, intelligent decision-making, dynamic early warning to anomaly self-healing, effectively supporting the group-level large-scale and refined management needs of cigarette packaging.

[0189] like Figure 2 As shown, this embodiment provides an example of a group-level cigarette pack management device. In this embodiment, the group-level cigarette pack management device is applied to the group-level cigarette pack management method as described in the above embodiment.

[0190] Specifically, the group-level cigarette management device includes a standardized data stream acquisition module 1, a multi-objective optimization scheduling module 2, an optimal management scheme calculation module 3, an inventory and inventory age calculation module 4, a delivery progress completion time prediction module 5, and an abnormal order acquisition module 6, which are connected by electrical or signal connections in sequence.

[0191] The system comprises the following modules: Standardized Data Stream Acquisition Module 1, which extracts and transforms field-level data from inventory details, equipment status, and QR code scanning data of production units and remote warehouses based on dynamic ETL, generating a standardized data stream; Multi-Objective Optimization Scheduling Module 2, which defines at least two objective functions based on a preset management strategy and integrates all objective functions to obtain a multi-objective optimization scheduling model; Optimal Management Scheme Calculation Module 3, which inputs the standardized data stream into the multi-objective optimization scheduling model and solves for the optimal management scheme of the multi-objective optimization scheduling model using a non-dominated sorting genetic algorithm; Inventory and Inventory Age Calculation Module 4, which calculates safety stock and inventory age warning lines based on spatiotemporal, attribute, and inventory age dimensions using a dynamic threshold algorithm; Delivery Progress Completion Time Prediction Module 5, which monitors the delivery plan using safety stock and inventory age warning lines and predicts the delivery progress completion time of the delivery plan using a linear regression model; and Abnormal Order Acquisition Module 6, which acquires abnormal orders that deviate from the delivery progress completion time in the delivery plan and classifies the abnormal orders according to the cause of the abnormality using a decision tree model to obtain a list of abnormal orders.

[0192] It should be noted that this embodiment is a functional module embodiment based on the above method embodiment. For additional content such as extensions, optimizations, limitations, examples, principle explanations, and beneficial effects of this embodiment, please refer to the above embodiments. This embodiment will not repeat them here.

[0193] Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Figure 3 As shown, the electronic device 7 includes a processor 71 and a memory 72 coupled to the processor 71.

[0194] The memory 72 stores program instructions for implementing the federated learning-based collaborative energy-saving method for government data clusters in any of the above embodiments.

[0195] The processor 71 is used to execute program instructions stored in the memory 72 for collaborative energy saving of government data clusters based on federated learning.

[0196] The processor 71 can also be referred to as a CPU (Central Processing Unit). The processor 71 may be an integrated circuit chip with signal processing capabilities. The processor 71 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor.

[0197] Furthermore, Figure 4 This is a schematic diagram of the structure of a storage medium according to an embodiment of this application. See also: Figure 4 The storage medium 8 in this embodiment stores program instructions 81 capable of implementing all the above methods. These program instructions 81 can be stored in the storage medium as a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods in each embodiment of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, or terminal devices such as computers, servers, mobile phones, and tablets.

[0198] In the several embodiments provided in this application, it should be understood that the disclosed apparatus, devices, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, signal, or other forms.

[0199] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A group-level management method for cigarette cartons, wherein the group-level management method for cigarette cartons is applied to production units and off-site warehouses under a group, characterized in that, The management method for group-level cigarettes includes: Step S1: Based on dynamic ETL, extract and transform the inventory details, equipment status, and QR code scanning data of the production unit and the remote warehouse at the field level to generate a standardized data stream; Step S2: Define at least two objective functions based on the preset management strategy, and integrate all objective functions to obtain a multi-objective optimization scheduling model; Step S3: Input the standardized data stream into the multi-objective optimization scheduling model, and solve the optimal management scheme of the multi-objective optimization scheduling model using a non-dominated sorting genetic algorithm; Step S4: Calculate safety stock and inventory age warning line based on spatiotemporal dimension, attribute dimension, and inventory age dimension using a dynamic threshold algorithm; Step S5: Monitor the delivery plan using the safety stock and the inventory age warning line, and predict the delivery progress completion time of the delivery plan using a linear regression model; Step S6: Obtain abnormal orders in the shipping plan that deviate from the shipping progress completion time, and classify the abnormal orders according to the abnormal reasons using a decision tree model to obtain a list of abnormal orders.

2. The group-level cigarette pack management method according to claim 1, characterized in that, Step S1: Based on dynamic ETL, perform field-level extraction and transformation of the inventory details, equipment status, and QR code scanning data of the production unit and the remote warehouse to generate a standardized data stream, including: Step S1.1: Configure the JDBC connection parameters of the production unit database and the RESTful API interface address of the remote warehouse, and set the MQTT topic of the QR code scanning device to establish a multi-source data connection; Step S1.2: Based on the preset field-level whitelist rules, the inventory details table, device status log, and QR code scanning records are extracted in parallel through multi-source data connection to obtain the raw data stream; Step S1.3: Clean the original data stream by removing invalid characters and filling in empty values ​​using regular expressions to obtain a clean data stream; Step S1.4: Standardize and transform the clean data stream based on a preset mapping rule base to obtain a unified encoded data stream; Step S1.5: Merge the unified encoded data stream into micro-batches according to time windows, and assign a unified timestamp and data source identifier to obtain several time-aligned data blocks; Step S1.6: Publish all time-aligned data blocks to a distributed message queue and write them to a group-level data pool to obtain a standardized data stream.

3. The group-level cigarette pack management method according to claim 2, characterized in that, Step S2: Define at least two objective functions based on a preset management strategy, and integrate all objective functions to obtain a multi-objective optimization scheduling model, including: Step S2.1: Extract historical operating time of equipment, fault records, time series data of inventory quantity changes, and transportation cost rates from the standardized data stream, and integrate them to obtain the model parameter set; Step S2.2: Based on the model parameter set, calculate the weight coefficients of equipment utilization rate, inventory turnover rate, and transportation cost using the entropy weight method to obtain the target weight vector; Step S2.3: Using the model parameter set and the target weight vector, define a first objective function that aims to maximize equipment utilization, a second objective function that aims to maximize inventory turnover, and a third objective function that aims to minimize transportation costs. Step S2.4: Integrate the first objective function, the second objective function, and the third objective function using a linear weighting method to form a comprehensive objective function; Step S2.5: Parse the real-time inventory limit and the maximum daily production capacity of each production unit from the standardized data stream and use them as constraints. Step S2.6: Integrate the comprehensive objective function and the constraints to obtain a multi-objective optimization scheduling model.

4. The group-level cigarette pack management method according to claim 3, characterized in that, Step S3 involves inputting the standardized data stream into the multi-objective optimization scheduling model and solving for the optimal management scheme of the multi-objective optimization scheduling model using a non-dominated sorting genetic algorithm, including: Step S3.1: Define the chromosome length according to the number of production units, and randomly generate an initial population that satisfies the constraints by real number encoding; Step S3.2: Decode each chromosome in the initial population into a specific contract allocation scheme, and substitute it into the comprehensive objective function of the multi-objective optimization scheduling model to calculate the objective function value of each individual; Step S3.3: Based on the objective function value of each individual, the population is stratified using the fast non-dominated sorting algorithm to determine the frontier level of each individual; Step S3.4: Within the same frontier level, calculate the crowding distance for each individual; Step S3.5: Based on the frontier level and crowding distance of each individual, the parent population is selected by using a binary tournament selection operator. Step S3.6: Perform simulated binary crossover and polynomial mutation operations sequentially on all parent individuals in the parent population to obtain a new generation of offspring population; Step S3.7: Merge the parent population with the new generation offspring population, and repeat steps S3.3 to S3.7 with the merged population as the subject of execution. After each iteration, select a new generation population through an elite retention strategy. Repeat the iteration until the termination condition is met, and use the Pareto optimal solution set as the optimal management scheme.

5. The group-level cigarette pack management method according to claim 1, characterized in that, Step S4 involves calculating safety stock and inventory age warning lines using a dynamic threshold algorithm based on spatiotemporal, attribute, and inventory age dimensions. This includes: Step S4.1: Based on the optimal contract allocation scheme, extract the associated inventory details, inbound and outbound documents, and QR code data from the standardized data stream; Step S4.2: Aggregate and group the inventory details, inbound and outbound documents, and QR code data in a spatiotemporal dimension according to a preset time period and a preset geographical level, and calculate the total inventory and change trend under each spatiotemporal unit. Step S4.3: Aggregate and group the inventory details, inbound and outbound documents, and QR code data according to the attribute dimensions based on product specifications and inventory status, and calculate the quantity and proportion of inventory in each group; Step S4.4: Calculate the inventory age of each cigarette box based on the coding timestamp in the QR code data, and associate the inventory age data with the spatiotemporal dimension aggregation grouping results and the attribute dimension aggregation grouping results to obtain a comprehensive inventory dataset including spatiotemporal dimension, attribute dimension and inventory age dimension. Step S4.5: Calculate the average sales volume and standard deviation of sales volume of the comprehensive inventory dataset, as well as the safety stock threshold for each product specification in the spatiotemporal dimension. Step S4.6: Identify inventory categories with different sales popularity based on the inventory age dimension of the comprehensive inventory dataset, and determine the inventory age coefficient for each inventory category; Step S4.7: Calculate the inventory age warning line for each inventory category based on the inventory age coefficient and the average inventory age of each inventory category.

6. The group-level cigarette pack management method according to claim 1, characterized in that, Step S5 involves monitoring the delivery plan using the safety stock and the inventory aging warning line, and predicting the delivery completion time of the delivery plan using a linear regression model, including: Step S5.1: Obtain the total number of planned shipments, the quantity shipped, and the shipment timestamp from the standardized data stream, and associate the safety stock with the inventory age warning line to obtain the shipment plan monitoring dataset; Step S5.2: Calculate the current delivery progress execution rate based on the delivery plan monitoring dataset; Step S5.3: Using time as the independent variable and cumulative shipment volume as the dependent variable, the model is trained on the shipment plan monitoring dataset using a multiple linear regression algorithm to obtain a linear regression model that reflects the relationship between shipment rate and time. Step S5.4: Input the current time, the quantity shipped, and the current shipment progress execution rate into the linear regression model to predict the cumulative shipment volume at future time points and obtain the shipment progress prediction curve; Step S5.5: Obtain the time point corresponding to when the cumulative shipment volume reaches the shipment plan on the shipment progress prediction curve, and define the corresponding time point as the shipment progress completion time.

7. The group-level cigarette pack management method according to claim 1, characterized in that, Step S6: Obtain abnormal orders in the shipping plan that deviate from the shipping progress completion time, and classify the abnormal orders according to the cause of the abnormality using a decision tree model to obtain a list of abnormal orders, including: Step S6.1: Obtain the actual progress of the order in the real-time delivery plan and compare it with the delivery progress completion time to obtain the delivery progress deviation value; Step S6.2: Compare the shipping progress deviation value with a preset tolerance threshold, filter out orders whose shipping progress deviation value exceeds the preset tolerance threshold, and integrate them into a set of potential abnormal orders; Step S6.3: Extract real-time inventory, equipment operating status code, and transport vehicle GPS location data associated with the set of potential abnormal orders to construct an abnormal order feature vector; Step S6.4: Input the abnormal order feature vector into the pre-trained isolated forest model to calculate the abnormal score for each order; Step S6.5: Define orders with abnormal scores exceeding a preset abnormal score threshold as final abnormal orders, and extract the abnormal order feature vector corresponding to the final abnormal orders to obtain the confirmed abnormal order dataset; Step S6.6: Input the confirmed abnormal order dataset into the pre-trained decision tree classification model, and bind the abnormal cause labels output by the decision tree classification model with the corresponding final abnormal orders to obtain the abnormal order list.

8. A management device for group-level cigarette cartons, wherein the management device for group-level cigarette cartons is applied to the management method for group-level cigarette cartons as described in any one of claims 1 to 7, characterized in that, The group-level cigarette management device includes: The standardized data stream acquisition module is used to extract and transform the inventory details, equipment status, and QR code scanning data of the production unit and the remote warehouse based on dynamic ETL, and generate a standardized data stream. The multi-objective optimization scheduling module is used to define at least two objective functions based on a preset management strategy and integrate all objective functions to obtain a multi-objective optimization scheduling model. The optimal management scheme calculation module is used to input the standardized data stream into the multi-objective optimization scheduling model and solve the optimal management scheme of the multi-objective optimization scheduling model through a non-dominated sorting genetic algorithm. The inventory and inventory age calculation module is used to calculate safety stock and inventory age warning line based on spatiotemporal dimensions, attribute dimensions, and inventory age dimensions using a dynamic threshold algorithm; The delivery progress completion time prediction module is used to monitor the delivery plan through the safety stock and the inventory age warning line, and to predict the delivery progress completion time of the delivery plan through a linear regression model. The abnormal order acquisition module is used to acquire abnormal orders in the shipping plan that deviate from the shipping progress completion time, and to classify the abnormal orders according to the abnormal reasons using a decision tree model to obtain an abnormal order list.

9. An electronic device, characterized in that, The method includes a processor and a memory coupled to the processor, the memory storing program instructions executable by the processor; when the processor executes the program instructions stored in the memory, it implements the group-level cigarette management method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions that, when executed by a processor, enable the management method for group-level cigarette packs as described in any one of claims 1 to 7.

Images