Big data-based shoe industry ERP data management and control method and system

Abstract

This invention discloses a data management method and system for footwear ERP based on big data. It acquires multi-source data from sales terminal data, production equipment information, and external market dynamics. Addressing the heterogeneity of this multi-source data, it standardizes the format using pre-established mapping rules to obtain a standardized dataset. To address information silos within the standardized dataset, it employs distributed storage technology for centralized management and links dispersed data sources through cross-channel data association algorithms to establish a unified global data view. Based on this global data view, it extracts feature fields from historical data analysis. For missing and outlier values ​​in these feature fields, it applies interpolation techniques and anomaly detection mechanisms to clean the data, resulting in a complete historical data feature set. This invention achieves intelligent processing and analysis of multi-source heterogeneous data, improving the accuracy of demand forecasting and the rationality of resource allocation, thus providing strong support for enterprise production decisions.

Description

Technical Field

[0001] This invention relates to the field of ERP data management technology, and in particular discloses a method and system for managing ERP data in the footwear industry based on big data. Background Technology

[0002] In modern manufacturing, the footwear industry, as an sector closely linked to consumer demand, relies heavily on operational efficiency and market responsiveness to enhance its competitiveness. Optimizing business processes through data-driven approaches has become a core issue for the industry's development. Particularly in resource allocation and market forecasting, the scientific management of data resources is considered a key path for enterprises to achieve rapid response and accurate decision-making.

[0003] However, many footwear companies still face significant limitations in data management. While many have begun to collect sales and inventory information, these efforts often remain superficial, lacking in-depth integration and collaborative analysis of multi-source data. Information silos are prevalent, especially when dealing with data from different channels, making it difficult for companies to form comprehensive market insights and effectively respond to rapidly changing consumer trends. This limitation often puts companies in a passive position when formulating production plans and inventory strategies.

[0004] A deeper challenge lies in the connection between data integration and demand forecasting. First, the diversity and heterogeneity of data sources make unified storage and processing exceptionally complex. Information from sales terminals, production equipment, and external markets often comes in different formats and update frequencies, directly impacting the accuracy of data analysis. When data integration issues are not properly resolved, demand forecasting suffers from insufficient accuracy because historical data cannot be effectively combined with real-time market dynamics, making forecasting models ill-suited to changing consumer preferences and seasonal fluctuations. These two technical factors interact, jointly restricting companies' ability to respond quickly to market demands.

[0005] Therefore, how to build an efficient data integration framework and achieve accurate demand forecasting on this basis has become a key issue in improving the operational efficiency of footwear companies. Summary of the Invention

[0006] This invention provides a data management method and system for footwear ERP based on big data, aiming to solve at least one of the defects existing in the prior art.

[0007] One aspect of this invention relates to a data management method for footwear ERP systems based on big data, comprising the following steps:

[0008] Data from multiple sources, including sales terminal data, production equipment information, and external market dynamics, is obtained. Given the heterogeneous nature of the multi-source data, it is format-standardized using pre-established mapping rules to obtain a standardized dataset.

[0009] To address the information silos in standardized datasets, distributed storage technology is used for centralized management. Cross-channel data association algorithms are used to link dispersed data sources and establish a unified global data view.

[0010] Based on the global data view, feature fields for historical data analysis are extracted. For missing and outlier values ​​in the feature fields, interpolation techniques and anomaly detection mechanisms are applied to clean them, resulting in a complete set of historical data features.

[0011] Time-related information is obtained from the complete historical data feature set and combined with the real-time data update stream. If the real-time data update frequency is lower than the preset threshold, the supplementary recording mechanism is triggered to obtain a synchronized comprehensive dataset.

[0012] For the synchronized comprehensive dataset, we analyze the characteristics of changes in consumer preferences and seasonal fluctuations, perform trend decomposition using time series analysis methods, determine the distribution patterns of short-term fluctuations and long-term trends, and obtain the trend decomposition result set.

[0013] Based on the trend decomposition result set, key indicators of changes in consumption preferences are extracted, and machine learning clustering methods are used to stratify user groups and determine the demand preference feature sets of different groups.

[0014] Core variables are obtained from the set of demand preference features, input features are used to construct a prediction model, and demand prediction output results are obtained through iterative calculation of the model. Resource allocation suggestion data is then generated and transmitted to the production scheduling system.

[0015] Furthermore, the steps involved in obtaining multi-source data from sales terminal data, production equipment information, and external market dynamics, and then standardizing the format of this data using pre-established mapping rules to address its heterogeneous nature, resulting in a standardized dataset, include:

[0016] Data from multiple sources is obtained from sales terminals, production equipment information, and external market dynamics. Given the heterogeneous nature of the multi-source data, a pre-established mapping rule is used for preliminary cleaning. By comparing the fields of the multi-source data with the preset format template, a preliminary data set is obtained.

[0017] For the initially organized dataset, a format standardization tool is used to uniformly convert the fields. If the field values ​​in the dataset do not conform to the preset threshold range, they are marked as an anomaly and completed to obtain an intermediate dataset with a consistent format.

[0018] Based on the intermediate dataset, data integration tools are used to match the associated fields of sales terminal and equipment information, and combined with external market dynamic data, the integrated dataset is determined by timestamp alignment.

[0019] For the comprehensive dataset, a data validation tool is used to check its completeness. If there are missing items, they are filled in using historical data to obtain the final normalized dataset.

[0020] Furthermore, to address the information silos in standardized datasets, a distributed storage technology is used for centralized management. The steps to establish a unified global data view by linking dispersed data sources through cross-channel data association algorithms include:

[0021] Based on the data silo phenomenon in the standardized dataset, and considering the differences in scattered data and channels, distributed storage tools are used to centrally archive data from different sources. Data from different channels are integrated through pre-established field mapping rules to obtain a preliminary unified storage set.

[0022] For the initially unified storage set, obtain the related field information, and cross-compare the data in the storage distribution using data linking tools. If inconsistent field values ​​are found, they are corrected according to the time alignment rules to determine the corrected intermediate data set.

[0023] Based on the intermediate dataset, data integration tools are used to check data integrity. If a field is found to be missing, the relevant content is filled in through historical records to obtain a categorized and integrated data combination.

[0024] For the categorized and integrated data sets, a data view generation tool is used to finally link the information from the scattered data sources, determine whether they meet the cross-channel linking standards, and obtain the final unified view result.

[0025] Furthermore, based on the global data view, feature fields for historical data analysis are extracted. For missing and outlier values ​​in these feature fields, interpolation techniques and anomaly detection mechanisms are applied for cleaning to obtain a complete set of historical data features. The steps include:

[0026] Based on the global data view, time series data related to the feature fields are obtained from the historical records. For the missing data, a pre-established interpolation method is used for preliminary processing to obtain a preliminary completed data set.

[0027] For the initially completed dataset, the distribution of outlier values ​​is obtained. The data distribution is compared using a detection mechanism. If outlier values ​​exceed a preset threshold, they are corrected according to the cleaning rules, and the corrected intermediate data combination is determined.

[0028] Based on the corrected intermediate data combination, obtain the field association information, use business logic to perform consistency checks on the time series data, and if the check results do not meet the preset standards, trace the relevant content through historical records to obtain the sorted data set after the check.

[0029] For the cleaned data set after detection, the complete dataset is processed using cleaning rules to determine whether it meets the requirements of business logic and obtain the final combination of historical data features that meet the standards.

[0030] Furthermore, the steps to obtain time-related information from the complete historical data feature set, combined with the real-time data update stream, and triggering a supplementary recording mechanism if the real-time data update frequency is lower than a preset threshold, to obtain a synchronized comprehensive dataset include:

[0031] Time-related information is obtained from the complete historical data feature set. For the dynamic changes in the real-time data flow, a pre-established monitoring tool is used to track the update frequency. If the update frequency is lower than the preset threshold, the subsequent process is triggered to obtain a preliminary judgment result of the frequency anomaly.

[0032] Based on the initial judgment of frequency anomalies, obtain the time-series correlation information between real-time data and historical data, and compare the business timeliness of the two types of data, real-time data and historical data, through data integration tools. If the comparison results show that the timeliness deviation exceeds the preset range, the supplementary recording mechanism is activated to determine the range of data required for supplementary recording.

[0033] To determine the data range required for supplementary recording, a dynamic adjustment tool is used to extract the corresponding time dimension information from the historical data feature set. By matching it with the frequency threshold of real-time data, a temporary dataset after supplementary recording is obtained.

[0034] Based on the supplemented temporary dataset, complete data verification information is obtained. A data synchronization tool is used to integrate the temporary dataset with the real-time data flow to determine whether the requirements for synchronizing the dataset are met, thus obtaining the final comprehensive dataset.

[0035] Furthermore, for the synchronized comprehensive dataset, the steps to analyze changes in consumer preferences and seasonal fluctuations, perform trend decomposition using time series analysis methods, determine the distribution patterns of short-term fluctuations and long-term trends, and obtain the trend decomposition result set include:

[0036] Information on consumption preferences and changes in preferences is obtained from the synchronized comprehensive dataset. Based on the distribution characteristics of the comprehensive dataset in the time dimension, a pre-established time series decomposition tool is used to perform preliminary processing on the comprehensive dataset to obtain an initial feature set containing seasonal fluctuations.

[0037] Based on the initial feature set, relevant information on seasonal and short-term fluctuations is extracted. Combined with the data characteristics of the time series, the fluctuation characteristics are judged by data comparison tools to determine whether they meet the preset threshold range. If they exceed the threshold range, the initial feature set is adjusted to determine the adjusted feature set.

[0038] For the adjusted feature set, relevant information on long-term trends and business patterns is obtained. Data integration tools are used to compare the feature set with historical data to determine whether there is a trend deviation, and a matching trend distribution set is obtained.

[0039] By combining the matched trend distribution set with the business requirements of dataset integration, a dynamic mapping tool is used to extract the correlation information between consumer preferences and long-term trends. The correlation information is then validated for regularity to obtain the final trend decomposition result set.

[0040] Furthermore, based on the trend decomposition result set, key indicators of changes in consumer preferences are extracted, and machine learning clustering methods are used to stratify user groups. The steps to determine the demand preference feature sets of different groups include:

[0041] Based on the trend decomposition result set, purchase frequency, purchase amount and category preference data are obtained. User behavior records under time distribution are classified and organized, and compared with a pre-established database to determine the initial set of consumption preferences for each user.

[0042] Clustering tools are used to process the initial set of consumption preferences. Combined with user age, income level and regional differences, if the consumption frequency of a certain user group is higher than the preset threshold, it is classified into the high-frequency consumption category, and the stratified user group segmentation results are obtained.

[0043] Based on the user group segmentation results, we obtain data on each group's purchase channels, promotion sensitivity, and brand loyalty. We then use data comparison tools to analyze the differences in payment methods and return rates to determine the combination of demand and preference characteristics of different groups.

[0044] Based on the combination of demand and preference characteristics and the time distribution attributes, data visualization tools are used to generate charts showing the changing trends of consumption preferences for each group, thus determining the final distribution of group characteristics.

[0045] Furthermore, the steps of obtaining core variables from the demand preference feature set, constructing the input features of the prediction model, obtaining the demand prediction output through model iterative calculation, and generating resource allocation suggestion data to be transmitted to the production scheduling system include:

[0046] Core variables are obtained from the set of demand preference features. Pre-established screening rules are used to sort and filter at least one key indicator. By comparing with preset thresholds, a set of variables that meet the conditions is determined.

[0047] Construct the prediction input based on the set of variables, obtain the data fields related to the set of variables, and transform the data fields into a standardized input format through a logical mapping table to obtain a structured input dataset;

[0048] The calculation process is performed on the input dataset. General data processing tools are used for batch calculation. If the calculation result deviates from the preset threshold range, the input parameters are adjusted and the calculation is repeated to obtain the predicted output value that meets the standard.

[0049] Based on the predicted output values, resource allocation suggestions are generated and sent to the production scheduling system via a data transmission interface. If the transmission is interrupted, a retransmission mechanism is triggered to determine whether the data has been received completely.

[0050] Another aspect of the present invention relates to a big data-based footwear ERP data management and control system, used to execute the aforementioned big data-based footwear ERP data management and control method. The big data-based footwear ERP data management and control system includes:

[0051] The normalized dataset acquisition module is used to acquire multi-source data from sales terminal data, production equipment information and external market dynamics. In view of the heterogeneous characteristics of multi-source data, it performs format standardization processing through pre-established mapping rules to obtain a normalized dataset.

[0052] The global data view determination module is used to address the information silos in standardized datasets by using distributed storage technology for centralized management and linking dispersed data sources through cross-channel data association algorithms to determine a unified global data view.

[0053] The historical data feature set acquisition module is used to extract feature fields for historical data analysis based on the global data view. For missing values ​​and outliers in the feature fields, interpolation techniques and anomaly detection mechanisms are applied to clean them to obtain a complete historical data feature set.

[0054] The comprehensive dataset acquisition module is used to obtain time-related information from the complete historical data feature set. Combined with the real-time data update stream, if the real-time data update frequency is lower than the preset threshold, a supplementary recording mechanism is triggered to obtain a synchronized comprehensive dataset.

[0055] The trend decomposition result set acquisition module is used to analyze changes in consumer preferences and seasonal fluctuations in a synchronized comprehensive dataset. It performs trend decomposition using time series analysis methods to determine the distribution patterns of short-term fluctuations and long-term trends, and obtains the trend decomposition result set.

[0056] The demand preference feature set determination module is used to extract key indicators of changes in consumption preferences based on the trend decomposition result set, and to use machine learning clustering methods to stratify user groups and determine the demand preference feature sets of different groups.

[0057] The resource allocation suggestion generation module is used to generate core variables obtained from the demand preference feature set, construct the input features of the prediction model, obtain the demand prediction output results through model iterative calculation, and generate resource allocation suggestion data to be transmitted to the production scheduling system.

[0058] Furthermore, the normalized dataset acquisition module includes:

[0059] The data set acquisition unit is used to acquire multi-source data from sales terminals, production equipment information and external market dynamics. In view of the heterogeneous characteristics of multi-source data, it uses pre-established mapping rules to perform preliminary cleaning. By comparing the multi-source data fields with the preset format template, a preliminary data set is obtained.

[0060] The intermediate dataset acquisition unit is used to perform field uniform conversion using a format standardization tool on the initially organized dataset. If the field values ​​in the dataset do not conform to the preset threshold range, they are marked as abnormal and completed to obtain an intermediate dataset with a consistent format.

[0061] The integrated dataset acquisition unit is used to match the associated fields of sales terminal and equipment information with data integration tools based on the intermediate dataset, and combine them with external market dynamic data to determine the integrated dataset through timestamp alignment.

[0062] The normalized dataset acquisition unit is used to check the completeness of the comprehensive dataset using data validation tools. If there are missing items, they are filled in with historical data to obtain the final normalized dataset.

[0063] The beneficial effects achieved by this invention are as follows:

[0064] This invention provides a data management method and system for footwear ERP based on big data. By standardizing and centrally managing multi-source heterogeneous data from sales terminals, production equipment, and external markets, it solves the problem of information silos. A cross-channel data association algorithm is used to construct a global data view, and historical data is feature extracted and cleaned, combined with real-time data updates to form a synchronized comprehensive dataset. Time series analysis is used to decompose changes in consumer preferences and seasonal fluctuations, and machine learning clustering methods are used to stratify user groups. Finally, a predictive model is constructed to output demand forecast results and resource allocation suggestions. This invention achieves intelligent processing and analysis of multi-source heterogeneous data, improves the accuracy of demand forecasting and the rationality of resource allocation, and provides strong support for enterprise production decisions. Attached Figure Description

[0065] Figure 1 This is a flowchart illustrating an embodiment of the big data-based ERP data management method for the footwear industry according to the present invention.

[0066] Figure 2 This is a functional block diagram of an embodiment of the big data-based footwear ERP data management system of the present invention.

[0067] Explanation of icon numbers:

[0068] 10. Comprehensive dataset acquisition module; 20. Global data view determination module; 30. Historical data feature set acquisition module; 40. Comprehensive dataset acquisition module; 50. Trend decomposition result set acquisition module; 60. Demand preference feature set determination module; 70. Resource allocation suggestion generation module. Detailed Implementation

[0069] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0070] like Figure 1 As shown, the first embodiment of the present invention proposes a data management method for footwear ERP based on big data, including the following steps:

[0071] Step S100: Obtain multi-source data from sales terminal data, production equipment information and external market dynamics. In view of the heterogeneous characteristics of multi-source data, perform format standardization processing through pre-established mapping rules to obtain a standardized dataset.

[0072] A normalized dataset refers to a collection of data with a unified format, fields, and logic, formed by eliminating differences in data format, structure, and semantics from heterogeneous data sources such as sales terminal data, production equipment information, and external market dynamics through pre-defined mapping rules. Its core objective is to solve the "data silo" problem and provide a consistent data foundation for subsequent data analysis (such as market forecasting and production scheduling).

[0073] Step S200: To address the information silo phenomenon in standardized datasets, distributed storage technology is used for centralized management. A cross-channel data association algorithm is used to link dispersed data sources and determine a unified global data view.

[0074] A global data view refers to a unified data presentation and access layer that covers the entire business chain and supports multi-dimensional correlation analysis. This is achieved by using distributed storage technology to centralize the management of multi-source data on a standardized dataset and by breaking down "information silos" (i.e., the lack of correlation between different data sources due to physical isolation and logical separation) through cross-channel data association algorithms. Its core feature is "physically distributed storage and logically centralized association," allowing users to query and access related data scattered across different nodes through a single entry point.

[0075] Step S300: Based on the global data view, extract the feature fields for historical data analysis. For missing values ​​and outliers in the feature fields, apply interpolation techniques and anomaly detection mechanisms to clean them, and obtain a complete set of historical data features.

[0076] A complete historical data feature set refers to a structured feature set that is "free of critical missing values, free of significant anomalies, and can be directly used for modeling and analysis" formed by extracting historical data analysis feature fields related to business objectives (such as "monthly growth rate" in sales data and "equipment failure interval" in production data) from the global data view, and by cleaning the data using interpolation techniques (filling in missing values) and anomaly detection mechanisms (correcting or removing anomalies) to address missing values ​​(such as unrecorded terminal sales on some dates) and outliers (such as abnormally high production temperatures due to sensor failure).

[0077] Step S400: Obtain time-related information from the complete historical data feature set, and combine it with the real-time data update stream. If the real-time data update frequency is lower than the preset threshold, trigger the supplementary recording mechanism to obtain a synchronized comprehensive dataset.

[0078] A synchronized comprehensive dataset refers to a dataset that integrates time-related information with real-time data updates based on a complete set of historical data features, and ensures data timeliness and completeness through dynamic monitoring and supplementation mechanisms. Its core is to achieve spatiotemporal consistency between historical and real-time data, eliminate data gaps caused by real-time update delays, and provide continuous and reliable data support for subsequent analysis and decision-making.

[0079] Step S500: For the synchronized comprehensive dataset, analyze the characteristics of changes in consumer preferences and seasonal fluctuations, perform trend decomposition using time series analysis methods, determine the distribution patterns of short-term fluctuations and long-term trends, and obtain the trend decomposition result set.

[0080] Trend decomposition result set refers to a structured dataset formed by breaking down changes in consumer preferences and seasonal fluctuations based on a synchronized comprehensive dataset using time series analysis methods, separating short-term fluctuations, long-term trends, and other potential patterns. Its core is to quantify and decompose the multi-dimensional patterns of change (such as long-term development trends, periodic fluctuations, and random disturbances) mixed in the original data, providing a clear analytical basis for understanding the inherent logic of the data and predicting future changes.

[0081] Step S600: Based on the trend decomposition result set, extract key indicators of changes in consumer preferences, use machine learning clustering methods to stratify user groups, and determine the demand preference feature sets of different groups.

[0082] The demand preference feature set of different groups refers to the unique set of characteristics extracted from the key indicators of changes in consumption preferences in the trend decomposition result set after segmenting user groups through machine learning clustering methods, revealing the consumption habits, demand tendencies, and behavioral patterns of each segmented group. Its core purpose is to achieve accurate user profiling, reveal the differentiated needs of different groups, and provide a basis for personalized strategy formulation.

[0083] Step S700: Obtain core variables from the demand preference feature set, construct the input features of the prediction model, obtain the demand prediction output results through model iterative calculation, and generate resource allocation suggestion data to be transmitted to the production scheduling system.

[0084] Based on core variables in the demand preference feature set, an input feature system for the prediction model is constructed. The model iteratively calculates and outputs demand prediction results, and generates resource allocation suggestions that can be directly used for production scheduling based on the prediction conclusions. Ultimately, this achieves end-to-end data-driven operation from user demand analysis to production execution. Its core is to transform abstract user preferences into concrete production resource allocation schemes, supporting a dynamic balance between supply and demand.

[0085] Furthermore, the big data-based footwear ERP data management method provided in this embodiment includes step S100 as follows:

[0086] Step S110: Obtain multi-source data from sales terminals, production equipment information and external market dynamics. In view of the heterogeneous characteristics of multi-source data, perform preliminary cleaning using pre-established mapping rules. By comparing the multi-source data fields with the preset format template, obtain a preliminary data set.

[0087] When processing multi-source data from sales terminals, production equipment, and external market dynamics, we begin by addressing the heterogeneity of the data. Heterogeneous data often manifests as inconsistencies in data format, field naming, and data structure across different sources. For example, suppose sales terminal data records daily sales revenue in a table format with a field named "Sale_Amt," while production equipment data records equipment operating status in a log format with a field named "Equip_Status." In this case, pre-established mapping rules can uniformly map "Sale_Amt" to "Sales Revenue" and "Equip_Status" to "Equipment Status," thus completing initial data cleaning. By comparing against preset format templates, such as requiring field names to be in Chinese and no more than 10 characters long, we obtain a pre-organized dataset. This approach effectively reduces the complexity of subsequent processing and improves data consistency.

[0088] Step S120: For the initially organized data set, use a format standardization tool to uniformly convert the fields. If the field values ​​in the data set do not conform to the preset threshold range, they are marked as abnormal and completed to obtain an intermediate dataset with a consistent format.

[0089] For the initially compiled dataset, a format standardization tool can further unify the field formats. For example, if some values ​​in the "Sales Amount" field are in "yuan" and others in "ten thousand yuan," the standardization tool will convert them all to "yuan." If a field value, such as a negative sales amount, does not fall within the preset threshold range of 0 to 1,000,000, it is marked as an anomaly and padded, for example, by using the average value from similar terminals to padded it to a reasonable value, ultimately resulting in an intermediate dataset with a consistent format. This method ensures the readability and usability of the data, laying the foundation for subsequent analysis.

[0090] Step S130: Based on the intermediate dataset, use the data integration tool to match the associated fields of sales terminal and equipment information, combine with external market dynamic data, and determine the integrated dataset by aligning with timestamps.

[0091] During the data integration phase, data integration tools are used to match the relevant fields between sales terminals and equipment information, such as linking the two through the "Terminal ID" field. Simultaneously, external market dynamic data, such as market price fluctuations over a certain period, is combined with timestamp alignment to merge the three into a comprehensive dataset. For example, assuming a terminal's sales revenue on October 1, 2023, was 50,000 yuan, its equipment status was "normal," and the market price increased by 5%, timestamp alignment clearly reflects the relationship between the three. This integration method helps to gain multi-dimensional insights into business conditions and improve the comprehensiveness of decision-making.

[0092] Step S140: For the comprehensive dataset, use a data validation tool to check its completeness. If there are missing items, fill them in with historical data to obtain the final normalized dataset.

[0093] For example, data validation tools can check the completeness of comprehensive datasets. If sales data for a particular terminal is missing on a certain day, it can be filled in using historical data, such as the average of the same terminal from the previous week, resulting in a final normalized dataset. This method ensures data integrity, avoids analytical biases due to missing data, and ultimately provides reliable support for business analysis. Through these multi-stage processing steps, not only is data quality improved, but a solid foundation is also laid for subsequent sales forecasting, equipment maintenance and optimization, significantly enhancing the ability to drive data-driven decision-making.

[0094] Furthermore, the big data-based footwear ERP data management method provided in this embodiment includes step S200 as follows:

[0095] Step S210: Based on the data silo phenomenon in the normalized dataset, and considering the differences in scattered data and channels, a distributed storage tool is used to centrally archive data from different sources. Data from different channels is integrated through pre-established field mapping rules to obtain a preliminary unified storage set.

[0096] When addressing data silos within standardized datasets, one can start by tackling the issues of fragmented data and channel differences, using distributed storage tools for centralized archiving of data from different sources. The core of distributed storage tools lies in unifying data scattered across various terminals and systems onto a single accessible platform. For example, sales terminal data might be stored on local servers, production equipment information in the cloud, and external market dynamics data from third-party interfaces. Distributed storage tools allow these data to be categorized and archived according to their source, forming a unified storage pool. For instance, daily transaction records from sales terminals can be archived by date and terminal number, while production equipment information can be stored by equipment number and time period. This approach facilitates subsequent data retrieval and integration.

[0097] Step S220: For the initially unified storage set, obtain the associated field information, and cross-compare the data in the storage distribution using a data linking tool. If inconsistent field values ​​are found, correct them according to the time alignment rules to determine the corrected intermediate data set.

[0098] For the initially unified storage set, obtaining the relevant field information and cross-referencing it using a data linking tool is a crucial step. Suppose there's a correspondence between the terminal number field in the sales terminal data and the equipment number field in the production equipment information, but due to different entry times, the values ​​may differ. The data linking tool can compare these two fields; if inconsistencies are found, it can correct them according to time alignment rules, such as using the most recent update as the standard, and generate an intermediate data set.

[0099] Specifically, if the sales data for a terminal with the serial number T001 was recorded on October 1, 2023, while the device information was recorded on September 30, 2023, the tool will prioritize the newer sales data as the basis for correction.

[0100] Step S230: Based on the intermediate data set, use a data integration tool to check the data integrity. If a field is found to be missing, fill in the relevant content through historical records to obtain a categorized and integrated data combination.

[0101] Building upon the intermediate dataset, it is crucial to employ data integration tools to check for completeness and fill in missing fields. For example, if a terminal's sales record for a particular day is missing an amount field in the integrated data, the tool can fill in the missing field using historical records, such as the average of the previous 7 days. Let's say the filled value is 50,000 yuan, forming a categorized and integrated data set. This completion method ensures data continuity, providing a complete foundation for subsequent analysis.

[0102] Step S240: For the categorized and integrated data combination, the scattered data source information is finally associated through the data view generation tool to determine whether it meets the cross-channel linking standard and obtain the final unified view result.

[0103] For categorized and integrated data sets, utilizing data view generation tools for final correlation and determining whether cross-channel linking standards are met is crucial for creating a unified view. Data view generation tools can link sales, device, and market dynamics data through common fields such as timestamps or terminal IDs. For example, assuming a terminal's sales amounted to 50,000 yuan on October 1, 2023, the device status was normal, and market dynamics showed a 3% price increase, the tool would generate a comprehensive view clearly showing the relationship between the three and determining whether cross-channel linking standards are met, such as data coverage exceeding 90%. This view facilitates observation of business status from multiple perspectives, improving the practicality of data integration.

[0104] Furthermore, the big data-based footwear ERP data management method provided in this embodiment includes step S300 as follows:

[0105] Step S310: Based on the global data view, retrieve time series data related to the feature fields from the historical records. For the missing data, use a pre-established interpolation method for preliminary processing to obtain a preliminary completed data set.

[0106] When processing time-series data in the global data view, the first step is to extract data related to the feature field from the historical records. Assuming the feature field is the daily sales volume of a certain terminal, the historical records show some missing data from the past 30 days. For the missing portions, a pre-established interpolation method can estimate the missing data based on the trend of the preceding and following data, resulting in a preliminary completed dataset.

[0107] Specifically, if a terminal is missing data on days 5 and 10, it can be supplemented by averaging the data from days 4 and 6, and days 9 and 11, forming a preliminary data set to lay the foundation for subsequent analysis.

[0108] Step S320: For the initially completed data set, obtain the distribution of abnormal values, compare the data distribution through the detection mechanism, and if abnormal values ​​are found to exceed the preset threshold, correct them according to the cleaning rules to determine the corrected intermediate data combination.

[0109] For the initially completed dataset, detecting the distribution of outliers is a crucial step. Suppose a sales volume figure for a particular day is 5000, while the average for the surrounding days is only 2000, significantly exceeding a preset threshold. After comparing the data distribution using a detection mechanism, this outlier can be corrected according to cleaning rules, for example, adjusting it to the average of the surrounding three days (2100) to form an intermediate data set. This method effectively smooths out abnormal fluctuations and improves data reliability.

[0110] Step S330: Based on the corrected intermediate data combination, obtain the field association information, and use business logic to perform consistency detection on the time series data. If the detection result does not meet the preset standard, trace the relevant content through historical records to obtain the sorted data set after detection.

[0111] Field association information is calculated using the following formula:

[0112] (1)

[0113] In formula (1), Representation field and fields The strength of the correlation between them This represents the total number of correlation metrics. Show the first The weights of each metric function Indicates the first A metric function for the field and The correlation calculation results. Formula (1) is used to quantify the correlation between different fields in the corrected intermediate data.

[0114] In the corrected intermediate data set, obtaining field correlation information and performing consistency checks is also crucial. For example, suppose there's a logical relationship between sales volume and inventory in the time series data; business logic requires that a decrease in inventory should correspond to an increase in sales volume. If the check finds that sales volume increased by 500 on a certain day, but inventory did not decrease, it's necessary to trace back through historical records to confirm whether there were any data entry errors, correct the data, and obtain a cleaned-up dataset. This traceability mechanism helps ensure logical consistency between data.

[0115] Step S340: For the cleaned data set after detection, the complete dataset is finally processed according to the cleaning rules to determine whether it meets the requirements of the business logic and obtain the final combination of historical data features that meet the standards.

[0116] For the organized dataset, cleaning rules are used to perform final processing on the complete dataset to ensure it meets business logic requirements. For example, if some data fields on a certain terminal still contain null values ​​or have inconsistent formats, cleaning rules can be used to standardize the format and fill in null values. For instance, null value fields can be set to the default value of 0 according to business rules, ultimately forming a standardized combination of historical data features. This comprehensive processing provides high-quality data support for subsequent business analysis, reducing decision-making biases caused by data issues.

[0117] Furthermore, the big data-based footwear ERP data management method provided in this embodiment includes step S400 as follows:

[0118] Step S410: Obtain time-related information from the complete historical data feature set. For the dynamic changes in the real-time data flow, use a pre-established monitoring tool to track the update frequency. If the update frequency is lower than the preset threshold, trigger the subsequent process to obtain a preliminary judgment result of frequency anomaly.

[0119] When processing complete historical data feature sets, relevant information is extracted based on the time dimension. Focusing on daily sales records for a specific terminal, this is combined with monitoring of dynamic changes in real-time data flow. Assuming the monitoring tool is set to an update frequency threshold of once per hour, if the data update frequency drops to once every two hours within a certain period, significantly lower than the preset value, subsequent processes are triggered, initially identifying an anomaly in frequency. This approach helps to promptly identify potential problems in the data flow.

[0120] Step S420: Based on the preliminary judgment of frequency anomalies, obtain the time-series correlation information of real-time data and historical data, and compare the business timeliness of the two types of data, real-time data and historical data, through data integration tools. If the comparison results show that the timeliness deviation exceeds the preset range, the supplementary recording mechanism is activated to determine the data range required for supplementary recording.

[0121] The frequency anomaly metric is calculated using the following formula:

[0122] (2)

[0123] In formula (2), Indicates the frequency anomaly metric. Indicates the length of the observation time window. Indicates time The frequency value, Indicates the historical frequency mean. Indicates the time decay coefficient. The formula (2) is used to quantify the degree to which the frequency deviates from the normal range and takes into account the time weight, indicating the current time.

[0124] Based on the initial assessment of frequency anomalies, when obtaining the time-series correlation information between real-time and historical data, a data integration tool is used to compare the timeliness of the two. For example, if historical data shows that sales data for a certain terminal is typically updated at the end of each day, while real-time data shows an update delayed until the next day, with the deviation exceeding a preset 24-hour range, a supplementary data entry mechanism is activated to determine the range of data requiring supplementation: missing records from the past 48 hours. This comparison mechanism effectively identifies timeliness issues and ensures the continuity of data processing.

[0125] Step S430: For the data range required for supplementary recording, a dynamic adjustment tool is used to extract the corresponding time dimension information from the historical data feature set, and the information is matched with the frequency threshold of the real-time data to obtain the temporary dataset after supplementary recording.

[0126] The data integrity level of the temporary dataset after supplementary data entry is quantitatively evaluated using the following formula:

[0127] (3)

[0128] In formula (3), Indicates data integrity verification metrics. This indicates the total number of fields in the temporary dataset. Indicates the first The weight coefficients of each field, Indicates the first Integrity scoring function for each field.

[0129] After determining the required data range for supplementation, a dynamic adjustment tool is used to extract corresponding time dimension information from the historical data feature set for matching. Assuming the real-time data frequency threshold is updated hourly, and some time periods in the historical data are missing records, the tool will extract data from adjacent time periods within the past 7 days for frequency matching, generating a temporary dataset after supplementation. This dynamic adjustment method can quickly fill data gaps and improve the completeness of the dataset.

[0130] Step S440: Based on the supplemented temporary dataset, obtain complete data verification information, use a data synchronization tool to integrate the temporary dataset with the real-time data flow, determine whether it meets the requirements of the synchronized dataset, and obtain the final comprehensive dataset.

[0131] The synchronization effect of integrating temporary datasets with real-time data streams is evaluated using the following formula:

[0132] (4)

[0133] In formula (4), This indicates the data synchronization quality assessment value. Indicators representing the reliability of temporary datasets Indicators representing the reliability of real-time data streams Indicates the data transmission delay factor. , , These are the corresponding weight parameters.

[0134] The final comprehensive dataset consists of a set of data records that meet the quality requirements.

[0135] (5)

[0136] In formula (5), This represents the final composite dataset. This represents the total number of all candidate data records. Show the first 1 data record Represents data records Quality assessment function value, This represents the minimum threshold for data quality.

[0137] For the temporary dataset after supplementary data entry, when obtaining complete data verification information and integrating real-time data flow, a data synchronization tool is used to determine whether the synchronization requirements are met. For example, if the temporary dataset covers missing sales records, but inventory data from some time points is still not synchronized, the tool will further integrate inventory update information from the real-time data flow to ultimately form a comprehensive dataset. This synchronization process ensures the consistency of multi-dimensional data, providing a reliable foundation for subsequent business analysis.

[0138] Furthermore, the big data-based footwear ERP data management method provided in this embodiment includes step S500:

[0139] Step S510: Obtain relevant information on consumption preferences and preference changes from the synchronized comprehensive dataset. Based on the distribution characteristics of the comprehensive dataset in the time dimension, use a pre-established time series decomposition tool to perform preliminary processing on the comprehensive dataset to obtain an initial feature set containing seasonal fluctuations.

[0140] The obtained consumption preferences are calculated using the following formula:

[0141] (6)

[0142] In formula (6), Indicates the first The weight of consumption preferences at time t. This indicates the total number of consumption categories. Indicates the first Weighting coefficients for different types of consumption. Indicates the first The consumer on the first Product category in time Consumption volume Indicates the total number of consumers. Indicates the first A consumer in time Total consumption Indicates the first Adjustment parameters for class preferences.

[0143] When extracting information on consumer preferences and changes in preferences from synchronized comprehensive datasets, we should first focus on the distribution characteristics over time. Suppose a terminal's sales data covers records from the past 12 months. Analysis reveals that consumers purchase significantly more in specific months than in other months, which may be related to holiday promotions. Based on this distribution characteristic, a pre-established time-series decomposition tool can be used for preliminary processing, splitting the data into three parts: seasonal fluctuations, short-term fluctuations, and long-term trends, resulting in an initial feature set. This decomposition method helps to more clearly identify consumption patterns over different time periods.

[0144] Step S520: Based on the initial feature set, extract relevant information on seasonal fluctuations and short-term fluctuations. Combine the data characteristics of the time series and use data comparison tools to determine whether the fluctuation characteristics meet the preset threshold range. If they exceed the threshold range, adjust the initial feature set and determine the adjusted feature set.

[0145] Extracting seasonal fluctuation patterns from time series using a weighted average method:

[0146] (7)

[0147] In formula (7), Indicates time Seasonal fluctuation characteristic values, This indicates the number of observation points within a seasonal cycle. Indicates lag The raw data values ​​for each time unit. Indicates the length of the seasonal cycle. Indicates the first The weighting coefficients for each observation point.

[0148] When extracting information related to seasonal and short-term fluctuations, data comparison tools can be used to determine whether the fluctuation characteristics meet the preset threshold range.

[0149] Assuming the preset threshold for seasonal fluctuations is a monthly sales fluctuation of no more than 20%, and if a certain month's sales growth reaches 30%, significantly exceeding this range, then the initial feature set needs adjustment. This adjustment can be achieved by incorporating the average value of historical data from the same period for smoothing, resulting in an adjusted feature set. This method avoids the interference of abnormal fluctuations on subsequent analysis.

[0150] Step S530: For the adjusted feature set, obtain relevant information on long-term trends and business patterns, use data integration tools to compare the feature set with historical data, determine whether there is a trend deviation, and obtain the matched trend distribution set.

[0151] The degree of trend deviation between the feature set and historical data is quantified by the following formula:

[0152] (8)

[0153] In formula (8), This represents a measure of trend deviation. Indicates the length of the time series. This indicates the adjusted feature set at time [time]. The value, This represents the corresponding value of historical data at time t. The standard deviation of the feature set, The standard deviation of historical data.

[0154] The degree of matching between the feature set and historical data is calculated using the following formula:

[0155] (9)

[0156] In formula (9), Indicates the matching score. Indicates the number of feature dimensions. Indicates the first The weight coefficients of each feature Represents the current feature distribution's th One portion, The first digit representing the historical baseline distribution One portion, This represents the matching sensitivity parameter.

[0157] The model for the matched trend distribution set is as follows:

[0158] (10)

[0159] In formula (10), Represents the set of trend distributions. Indicates the number of distributed components. Indicates the first The mixed weights of the components, Indicates the first A vector of mean values ​​from a Gaussian distribution. Let represent the covariance matrix of the i-th Gaussian distribution.

[0160] When extracting long-term trend and business pattern information from the adjusted feature set, data integration tools are used to compare it with historical data to determine if there is a trend deviation. For example, if historical data shows that the sales of a certain terminal have been steadily growing at a rate of 5% per year, while the current feature set shows a growth rate of only 2%, a preliminary judgment can be made that a trend deviation exists. Further analysis suggests that this may be due to recent changes in the market environment. By matching historical trends with current data, a trend distribution set that more closely reflects the actual situation can be obtained. This comparison helps ensure that the analysis results are close to business reality.

[0161] Step S540: Based on the matched trend distribution set and the business requirements for dataset integration, a dynamic mapping tool is used to extract the correlation information between consumer preferences and long-term trends. The correlation information is then validated for regularity to obtain the final trend decomposition result set.

[0162] When extracting the correlation between consumer preferences and long-term trends from the matched trend distribution set, a dynamic mapping tool is used. For example, if it is found that consumer preference for a certain type of product has gradually increased over the past three years and is highly correlated with long-term sales growth trends, then the stability of this correlation can be confirmed through regularity checks. If the check results show that the correlation holds true across multiple time periods, it is included in the final trend decomposition result set. This mapping and verification method provides a more reliable basis for business decisions.

[0163] Furthermore, the big data-based footwear ERP data management method provided in this embodiment includes step S600 as follows:

[0164] Step S610: Based on the trend decomposition result set, obtain consumption frequency, consumption amount and category preference data, classify and organize the user's behavior records under the time distribution, and compare them with the pre-established database to determine the initial set of consumption preferences for each user.

[0165] When analyzing user consumption behavior, the first step is to categorize and organize data based on purchase frequency, spending amount, and category preference. Let's assume that database records show a user makes an average of 5 purchases per month, with an average purchase amount of 200 yuan per transaction, and a preference for electronics and clothing. Based on time distribution, the user's behavior over the past 6 months can be divided into weekday and holiday purchases. It's found that holiday purchases are higher, and the user's category preference is more heavily weighted towards electronics. By comparing this with a pre-established database, the initial set of the user's consumption preferences is preliminarily determined to be "high-frequency electronics consumer." This approach helps to accurately depict the user profile, laying the foundation for subsequent personalized recommendations.

[0166] Step S620: Use clustering tools to process the initial set of consumption preferences. Combine user age, income level and regional differences. If the consumption frequency of a certain user group is higher than the preset threshold, it will be classified into the high-frequency consumption category to obtain the stratified user group division results.

[0167] The similarity distance between users is calculated using the following formula:

[0168] (11)

[0169] In formula (11), Indicates user and users Cluster distance between them Indicates the total number of attribute dimensions. Indicates the first The weight coefficients of each attribute, Indicates user In the Standardized values ​​for each attribute Indicates user In the The standardized values ​​for each attribute. Formula (11) is used to calculate the similarity distance between users considering multidimensional attributes such as age, income level, and regional differences.

[0170] When using clustering tools to process the initial set of consumer preferences, the analysis incorporates user age, income level, and regional differences. Assuming age is divided into ranges such as 20-30 and 30-40, income levels are categorized into low, middle, and high, and regional differences are considered based on city tier. The clustering results show that users aged 20-30, with middle income and residing in first-tier cities, have a consumption frequency exceeding 8 times per month, surpassing the preset threshold of 6 times; therefore, they are classified as high-frequency consumers. This stratified user segmentation effectively distinguishes differences in consumer behavior, providing data support for targeted marketing.

[0171] Step S630: Based on the user group segmentation results, obtain the purchase channels, promotion sensitivity and brand loyalty data for each group, and analyze the differences in payment methods and return rates through data comparison tools to determine the combination of demand preference characteristics of different groups.

[0172] The preference intensity of different user groups on various purchasing channels is calculated using the following formula:

[0173] (12)

[0174] In formula (12), This indicates that group g passes through a channel. The probability of purchase, Representing a group Through channels Number of purchases Representing a group Total number of purchases Representing a group Promotional sensitivity weight, Representing a group Brand loyalty coefficient.

[0175] The degree of difference in payment method choices among different groups is quantified using the following formula:

[0176] (13)

[0177] In formula (13), This represents the index indicating differences in payment methods. This indicates the total number of payment methods. Representing a group Use the The number of times each payment method is used. Representing a group Total number of transactions Representing a group Use the The number of times each payment method is used. Representing a group Total number of transactions.

[0178] Considering the influence of promotional sensitivity and brand loyalty on group return behavior, the following characteristics are calculated using a formula:

[0179] (14)

[0180] In formula (14), This indicates the adjusted return rate. This represents the number of returned orders in group n. Representing a group Total number of orders Representing a group Promotional sensitivity coefficient Representing a group Brand loyalty coefficient.

[0181] Based on the user segmentation results, data on purchase channels, promotion sensitivity, and brand loyalty were analyzed. It was assumed that high-frequency consumers primarily purchase through online e-commerce platforms, exhibiting high promotion sensitivity and low brand loyalty, while low-frequency consumers prefer offline brick-and-mortar stores, showing less reaction to promotions but higher brand loyalty. Data comparison tools revealed that high-frequency consumers primarily use electronic payments, with a return rate of 10%, while low-frequency consumers primarily use cash, with a return rate of only 3%. This difference analysis helps determine the combination of demand and preference characteristics of different groups; for example, high-frequency consumers prioritize price discounts, while low-frequency consumers value product quality and brand trust.

[0182] Step S640: Based on the combination of demand preference characteristics and time distribution attributes, use data visualization tools to generate consumption preference trend charts for each group to determine the final distribution of group characteristics.

[0183] When generating charts depicting changing consumer preferences, data visualization tools are used to showcase the characteristics of each group, taking into account time distribution attributes. Assuming a monthly basis, a curve showing the spending changes of high-frequency consumers over the past 12 months is plotted. This reveals a surge in spending during holidays such as Singles' Day (November 11th), while spending remains relatively stable at other times. Such charts intuitively reflect the cyclical patterns of consumer behavior, helping businesses to rationally schedule promotional activities and improve resource allocation efficiency. Through this multi-dimensional analysis, from initial data processing to final trend presentation, a complete user behavior profile is formed, significantly improving the targeting and effectiveness of marketing strategies.

[0184] Furthermore, the big data-based footwear ERP data management method provided in this embodiment includes step S700 as follows:

[0185] Step S710: Obtain core variables from the set of demand preference features, use pre-established screening rules to sort and filter at least one key indicator, and determine the set of variables that meet the conditions by comparing with preset thresholds.

[0186] When analyzing the cyclical patterns of consumer behavior, we start with the behavioral characteristics of high-frequency and low-frequency consumer groups to construct a set of core variables. For the high-frequency group, data shows they prefer online e-commerce platforms, are highly sensitive to promotions, and have lower brand loyalty. For example, data from a certain e-commerce platform shows that high-frequency users increased their purchase frequency by 30% during promotional events, while their brand switching rate reached 40%. In contrast, the low-frequency group prefers offline brick-and-mortar stores, responds less enthusiastically to promotions, but has higher brand loyalty. For instance, in statistics from a certain offline chain store, the brand repurchase rate for low-frequency users reached 70%. Using these core variables, we sort and filter key indicators such as purchase frequency and brand loyalty, setting thresholds such as a purchase frequency greater than 5 times / month for high frequency and a brand repurchase rate greater than 50% for high loyalty, thus selecting a set of variables that meet these criteria.

[0187] Step S720: Construct the prediction input based on the variable set, obtain the data fields related to the variable set, and transform the data fields into a standardized input format through a logical mapping table to obtain a structured input dataset.

[0188] The prediction input constructed based on the set of variables is:

[0189] (15)

[0190] In formula (15), Represents the set of inputs to be predicted. Represents a set of variables. Represents a collection of data fields. Indicates the first One data field, Indicates the first One variable, This function represents the relationship between a variable and a data field; a value of 1 indicates that a relationship exists. This represents a mapping function used to retrieve relevant data fields based on a set of variables.

[0191] The final structured input dataset is:

[0192] (16)

[0193] In formula (16), This represents the final structured input dataset. Indicates the total number of data records processed. Indicates the first A standardized data record, Indicates the first One original data record, This represents a structured processing function that obtains a complete structured input dataset through a union operation on all standardized data records.

[0194] When constructing the predictive input, relevant data fields, such as payment method and return rate, are obtained based on the aforementioned set of variables. High-frequency users primarily use electronic payments, with a return rate of 10%, while low-frequency users primarily use cash payments, with a return rate of only 3%. These fields are transformed into a standardized input format using a logical mapping table; for example, payment method is mapped to "electronic payment = 1, cash payment = 0," forming a structured input dataset. This standardization process helps improve the accuracy and consistency of subsequent calculations.

[0195] Step S730: Perform the calculation process on the input dataset. Use general data processing tools for batch calculation. If the calculation result deviates from the preset threshold range, adjust the input parameters and recalculate to obtain a predicted output value that meets the standard.

[0196] The output of the batch operation is as follows:

[0197] (17)

[0198] In formula (17), Indicates the first Batch processing output results for each data sample This indicates the total number of general-purpose data processing tools. Indicates the first Calculation functions of a data processing tool Indicates the first One input data sample, Indicates the first The parameter configuration of each processing tool. Formula (17) describes the process of performing batch calculations and taking the average value of a single data sample using multiple general processing tools.

[0199] For the calculation process of the input dataset, general data processing tools are used for batch operations. For example, a data analysis platform is used to predict and analyze the promotion sensitivity of high-frequency groups and the brand loyalty of low-frequency groups. If the calculation results deviate from the preset threshold range, such as the predicted promotion response rate being lower than the expected value of 20%, the input parameters are adjusted, such as increasing the weight of promotion intensity, and the calculation is repeated until the output value meets the standard. This iterative adjustment can effectively improve the reliability of the prediction.

[0200] Step S740: Generate suggested resource configuration data based on the predicted output value, and send the suggested data to the production scheduling system through the data transmission interface. If the transmission is interrupted, trigger the retransmission mechanism to determine whether the data has been received completely.

[0201] When generating resource allocation recommendations, based on predicted output values, it suggests increasing online promotional resource allocation for high-frequency groups, such as allocating 60% of the promotional budget to e-commerce platforms; for low-frequency groups, it suggests increasing investment in offline brand activities. The recommendations are sent to the production scheduling system via a data transmission interface. If transmission is interrupted, a retransmission mechanism is triggered to ensure complete data reception. For example, if the system detects a packet loss rate exceeding 5% during a transmission, it automatically initiates retransmission until data integrity is confirmed. This mechanism ensures the timeliness and accuracy of resource allocation recommendations.

[0202] Please see Figure 2This embodiment provides a big data-based footwear ERP data management and control system for executing the aforementioned big data-based footwear ERP data management and control method. The big data-based footwear ERP data management and control system includes a standardized dataset acquisition module 10, a global data view determination module 20, a historical data feature set acquisition module 30, a comprehensive dataset acquisition module 40, a trend decomposition result set acquisition module 50, a demand preference feature set determination module 60, and a resource allocation suggestion generation module 70. The standardized dataset acquisition module 10 acquires multi-source data from sales terminal data, production equipment information, and external market dynamics. Addressing the heterogeneous nature of the multi-source data, it performs format standardization processing using pre-established mapping rules to obtain a standardized dataset. The global data view determination module 20 addresses the information silos in the standardized dataset by using distributed storage technology for centralized management. It links dispersed data sources through cross-channel data association algorithms to determine a unified global data view. The historical data feature set acquisition module 30 extracts feature fields from historical data analysis based on the global data view, targeting specific features... Missing and outlier values ​​in the feature fields are cleaned using interpolation techniques and anomaly detection mechanisms to obtain a complete historical data feature set. The comprehensive dataset acquisition module 40 extracts time-related information from the complete historical data feature set and, combined with real-time data update streams, triggers a supplementary recording mechanism if the real-time data update frequency is below a preset threshold, resulting in a synchronized comprehensive dataset. The trend decomposition result set acquisition module 50 analyzes changes in consumer preferences and seasonal fluctuations in the synchronized comprehensive dataset, performs trend decomposition using time series analysis methods, determines short-term fluctuations and long-term trend distribution patterns, and obtains a trend decomposition result set. The demand preference feature set determination module 60 extracts key indicators of changes in consumer preferences based on the trend decomposition result set, uses machine learning clustering methods to stratify user groups, and determines the demand preference feature sets of different groups. The resource allocation suggestion generation module 70 generates core variables from the demand preference feature set, constructs input features for the prediction model, iteratively calculates the demand prediction output, and generates resource allocation suggestion data which is then transmitted to the production scheduling system.

[0203] Furthermore, this embodiment provides a big data-based footwear ERP data management and control system. The standardized dataset acquisition module 10 includes a data set acquisition unit, an intermediate dataset acquisition unit, a comprehensive dataset acquisition unit, and a standardized dataset acquisition unit. The data set acquisition unit acquires multi-source data from sales terminals, production equipment information, and external market dynamics. Addressing the heterogeneity of the multi-source data, it performs preliminary cleaning using pre-established mapping rules. By comparing the fields of the multi-source data with preset format templates, a preliminary data set is obtained. The intermediate dataset acquisition unit performs unified field conversion on the preliminary data set using a format standardization tool. If field values ​​in the data set do not conform to a preset threshold range, they are marked as abnormal and completed to obtain a consistent intermediate dataset. The comprehensive dataset acquisition unit matches the associated fields of sales terminals and equipment information using a data integration tool based on the intermediate dataset. Combined with external market dynamics data, it aligns the data with timestamps to determine the fused comprehensive dataset. The standardized dataset acquisition unit checks the completeness of the comprehensive dataset using a data verification tool. If missing items exist, they are filled in using historical data to obtain the final standardized dataset.

[0204] The beneficial effects of the big data-based ERP data management and control method and system for the footwear industry provided in this embodiment are as follows:

[0205] I. Improving Data Quality and Management Efficiency

[0206] 1. Eliminate information silos and achieve unified data management.

[0207] By leveraging distributed storage technology and cross-channel data association algorithms, heterogeneous data sources that were originally scattered across sales terminals, production equipment, and external markets are integrated into a global data view. This breaks down information barriers between various departments within the enterprise (such as production, sales, and procurement) and upstream and downstream of the supply chain, avoiding issues such as duplicate data storage and inconsistent data definitions. It improves data consistency and availability, providing a unified data foundation for decision-making at all stages.

[0208] 2. Data cleaning and dynamic synchronization ensure data accuracy and timeliness. Cleaning and processing of missing and outlier values ​​reduces the interference of "dirty data" on analysis results. Through real-time data update stream monitoring and supplementation mechanisms, the synchronization of data in the time dimension is ensured (such as real-time matching of sales data, inventory data and production data), avoiding decision-making biases caused by data lag and improving the credibility of ERP system data.

[0209] II. Demand Forecasting and Production Efficiency Optimization

[0210] 1. Accurately capture market trends and improve the accuracy of demand forecasting.

[0211] By combining time series analysis to decompose changes in consumer preferences and seasonal fluctuations into trends, distinguishing between short-term fluctuations (such as the impact of promotional activities) and long-term trends (such as the fashion cycle of styles), and then using machine learning clustering to segment user groups, the demand characteristics of different groups can be accurately identified (such as the functional preferences of young people in a certain region for athletic shoes, and the demand of middle-aged and elderly people for comfortable shoes). Based on this, the predictive model can output more accurate demand forecast results, reducing the occurrence of "overproduction" or "stockouts and supply disruptions".

[0212] 2. Optimize resource allocation and reduce production costs.

[0213] Demand forecasting results are directly converted into resource allocation recommendations and transmitted to the production scheduling system, enabling dynamic adjustments to production plans.

[0214] 1) Prioritize the allocation of equipment, raw materials, and manpower for high-demand styles to shorten the production cycle;

[0215] 2) Reduce production capacity for styles with low demand or seasonal decline to avoid inventory backlog; ultimately improve the utilization rate of production equipment, reduce raw material waste, and lower unit production costs.

[0216] III. Collaborative Optimization of Inventory and Supply Chain

[0217] 1. Dynamically match inventory with market demand to reduce capital occupation.

[0218] Based on a synchronized comprehensive dataset (including real-time sales, inventory, and forecast data), refined inventory management can be achieved: best-selling items maintain reasonable inventory levels, while slow-moving items trigger promotions or production cuts through early warnings, avoiding the problems of "overstocking leading to capital occupation" or "insufficient inventory leading to order loss" in traditional inventory management, and improving inventory turnover.

[0219] 2. Promote supply chain collaboration and shorten response cycles.

[0220] Once the global data view is opened to upstream and downstream suppliers (such as suppliers and distributors), suppliers can prepare raw materials in advance based on production demand forecasts, and distributors can adjust order quantities based on sales trends, realizing a collaborative model of "production based on sales and supply based on production", shortening the overall response time of the supply chain (such as the cycle from order placement to finished product delivery) and improving supply chain resilience.

[0221] IV. Enhanced Market Competitiveness and Scientific Decision-Making

[0222] 1. Quickly respond to changes in consumer demand and enhance product competitiveness.

[0223] By segmenting user groups and extracting demand and preference characteristics, companies can accurately capture the needs of niche markets (such as the preferences of specific age groups for footwear materials and designs), guiding product development and style iteration (such as launching non-slip shoes for outdoor enthusiasts and lightweight and comfortable models for commuters), making products more aligned with market demands and enhancing brand competitiveness.

[0224] Data-driven decision-making reduces the blind spots in decision-making. Global data views and visualized analysis results (such as trend decomposition and user segmentation characteristics) provide management with intuitive decision-making basis, replacing traditional experience-based "gut feeling" decisions. For example, adjusting offline store inventory strategies based on regional sales trends and planning peak season production in advance based on seasonal fluctuations improves the scientific and forward-looking nature of decision-making. V. Business Process Automation and Intelligent Upgrades

[0225] By standardizing processing, automatically filling in data, and iteratively calculating models, the number of manual data processing steps (such as manually entering sales data and manually calculating inventory) is reduced, thereby lowering labor costs and human error. At the same time, the ERP system is upgraded from "passively recording data" to "actively outputting decision-making suggestions" (such as resource allocation plans), driving footwear companies to transform from traditional manufacturing models to "data-driven intelligent production" and improving overall operational efficiency.

[0226] In summary, the big data-based ERP data management and control method and system for the footwear industry provided in this embodiment achieves a full-chain upgrade from data management and control to business optimization and from passive response to proactive prediction through the deep integration of big data technology and ERP system. Ultimately, it brings comprehensive benefits to footwear enterprises, including cost reduction, efficiency improvement, quality enhancement, and enhanced market competitiveness.

[0227] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A big data-based ERP data management and control method for the footwear industry, applied to the resource planning data management and control of footwear enterprises, wherein the data source includes production equipment information, characterized in that... Includes the following steps: Data from multiple sources, including sales terminal data, production equipment information, and external market dynamics, is obtained. Given the heterogeneous nature of this multi-source data, a standardized dataset is generated by standardizing the format using pre-established mapping rules. To address the information silos in the standardized dataset, distributed storage technology is used for centralized management, and a cross-channel data association algorithm is used to link dispersed data sources to determine a unified global data view. Based on the global data view, feature fields for historical data analysis are extracted. For missing and outlier values ​​in the feature fields, interpolation techniques and anomaly detection mechanisms are applied to clean them, resulting in a complete set of historical data features. Time-related information is obtained from the complete historical data feature set and combined with the real-time data update stream. If the real-time data update frequency is lower than the preset threshold, the supplementary recording mechanism is triggered to obtain a synchronized comprehensive dataset. For the synchronized comprehensive dataset, we analyze the characteristics of changes in consumer preferences and seasonal fluctuations, perform trend decomposition using time series analysis methods, determine the distribution patterns of short-term fluctuations and long-term trends, and obtain the trend decomposition result set. Based on the trend decomposition result set, key indicators of changes in consumption preferences are extracted, and machine learning clustering methods are used to stratify user groups and determine the demand preference feature sets of different groups. Core variables are obtained from the set of demand preference features, input features for the prediction model are constructed, and the demand prediction output is obtained through iterative calculation of the model. Resource allocation suggestion data is then generated and transmitted to the production scheduling system. Specifically, this includes: Core variables are obtained from the set of demand preference features. Using pre-established screening rules, at least one key indicator is sorted and filtered. By comparing with preset thresholds, a set of variables that meet the conditions is determined. Based on the set of variables, construct the prediction input, obtain the data fields related to the set of variables, and transform the data fields into a standardized input format through a logical mapping table to obtain a structured input dataset; The calculation process is performed on the input dataset. A general data processing tool is used for batch calculation. If the calculation result deviates from the preset threshold range, the input parameters are adjusted and the calculation is repeated to obtain a predicted output value that meets the standard. Based on the predicted output value, resource configuration suggestion data is generated and sent to the production scheduling system through the data transmission interface. If the transmission is interrupted, a retransmission mechanism is triggered to determine whether the data has been received completely.

2. The big data-based ERP data management and control method for the footwear industry as described in claim 1, characterized in that, The steps of acquiring multi-source data from sales terminal data, production equipment information, and external market dynamics, and then standardizing the format of the multi-source data using pre-established mapping rules to obtain a normalized dataset, include: Data from multiple sources is obtained from sales terminals, production equipment information, and external market dynamics. Based on the heterogeneous characteristics of the multi-source data, a pre-established mapping rule is used for preliminary cleaning. By comparing the fields of the multi-source data with the preset format template, a preliminary data set is obtained. For the initially organized dataset, a format standardization tool is used to uniformly convert the fields. If the field values ​​in the dataset do not conform to the preset threshold range, they are marked as abnormal and completed to obtain an intermediate dataset with a consistent format. Based on the intermediate dataset, the data integration tool is used to match the associated fields of the sales terminal and equipment information, and combined with the external market dynamic data, the integrated dataset is determined by timestamp alignment. For the comprehensive dataset, a data validation tool is used to check its completeness. If there are missing items, they are filled in with historical data to obtain the final normalized dataset.

3. The big data-based ERP data management and control method for the footwear industry as described in claim 1, characterized in that, To address the information silos within the standardized dataset, a distributed storage technology is employed for centralized management. This involves linking disparate data sources using cross-channel data association algorithms to determine a unified global data view. The steps include: Based on the data silo phenomenon in the standardized dataset, and considering the differences in scattered data and channels, a distributed storage tool is used to centrally archive data from different sources. Data from different channels is integrated through pre-established field mapping rules to obtain a preliminary unified storage set. For the initially unified storage set, obtain the related field information, and cross-compare the data in the storage distribution using data linking tools. If inconsistent field values ​​are found, they are corrected according to the time alignment rules to determine the corrected intermediate data set. Based on the intermediate data set, a data integration tool is used to check the data integrity. If a field is found to be missing, the relevant content is filled in through historical records to obtain a categorized and integrated data combination. For the categorized and integrated data sets, a data view generation tool is used to finally link the information from the scattered data sources, determine whether they meet the cross-channel linking standards, and obtain the final unified view result.

4. The big data-based ERP data management and control method for the footwear industry as described in claim 1, characterized in that, Based on the global data view, the steps of extracting feature fields for historical data analysis, and cleaning the feature fields by applying interpolation techniques and anomaly detection mechanisms to obtain a complete set of historical data features include: Based on the global data view, time series data related to the feature field are obtained from the historical records. For the missing data, a pre-established interpolation method is used for preliminary processing to obtain a preliminary completed data set. For the initially completed dataset, the distribution of outlier values ​​is obtained. The data distribution is compared using a detection mechanism. If outlier values ​​exceed a preset threshold, they are corrected according to the cleaning rules, and the corrected intermediate data combination is determined. Based on the corrected intermediate data combination, obtain the field association information, use business logic to perform consistency checks on the time series data, and if the check results do not meet the preset standards, trace the relevant content through historical records to obtain the sorted data set after the check. For the cleaned data set after detection, the complete dataset is processed using cleaning rules to determine whether it meets the requirements of business logic, and finally obtains a set of historical data features that meet the standards.

5. The big data-based ERP data management and control method for the footwear industry as described in claim 1, characterized in that, The steps of obtaining time-related information from a complete historical data feature set, combining it with real-time data update streams, and triggering a supplementary recording mechanism if the real-time data update frequency is lower than a preset threshold to obtain a synchronized comprehensive dataset include: Time-related information is obtained from the complete historical data feature set. For the dynamic changes in the real-time data flow, a pre-established monitoring tool is used to track the update frequency. If the update frequency is lower than the preset threshold, the subsequent process is triggered to obtain a preliminary judgment result of the frequency anomaly. Based on the preliminary judgment of frequency anomalies, the time-series correlation information of real-time data and historical data is obtained. The business timeliness of the two types of data, real-time data and historical data, is compared by data integration tools. If the comparison results show that the timeliness deviation exceeds the preset range, the supplementary recording mechanism is activated to determine the range of data required for supplementary recording. To determine the data range required for supplementary recording, a dynamic adjustment tool is used to extract the corresponding time dimension information from the historical data feature set. This information is then matched with the frequency threshold of the real-time data to obtain a temporary dataset after supplementary recording. Based on the supplemented temporary dataset, complete data verification information is obtained. The temporary dataset is then integrated with the real-time data flow using a data synchronization tool to determine whether it meets the requirements for a synchronized dataset, thus obtaining the final comprehensive dataset.

6. The big data-based ERP data management and control method for the footwear industry as described in claim 1, characterized in that, The steps involved in analyzing changes in consumer preferences and seasonal fluctuations in a synchronized comprehensive dataset, performing trend decomposition using time series analysis methods, determining the distribution patterns of short-term fluctuations and long-term trends, and obtaining the trend decomposition result set include: Relevant information on consumption preferences and preference changes is obtained from the synchronized comprehensive dataset. Based on the distribution characteristics of the comprehensive dataset in the time dimension, a pre-established time series decomposition tool is used to perform preliminary processing on the comprehensive dataset to obtain an initial feature set containing seasonal fluctuations. Based on the initial feature set, relevant information on seasonal and short-term fluctuations is extracted. Combined with the data characteristics of the time series, a data comparison tool is used to determine whether the fluctuation characteristics meet the preset threshold range. If they exceed the threshold range, the initial feature set is adjusted to determine the adjusted feature set. For the adjusted feature set, relevant information on long-term trends and business patterns is obtained. Data integration tools are used to compare the feature set with historical data to determine whether there is a trend deviation, and a matching trend distribution set is obtained. By combining the matched trend distribution set with the business requirements of dataset integration, a dynamic mapping tool is used to extract the correlation information between consumer preferences and long-term trends. The correlation information is then verified for regularity to obtain the final trend decomposition result set.

7. The big data-based ERP data management and control method for the footwear industry as described in claim 1, characterized in that, Based on the trend decomposition result set, the steps of extracting key indicators of changes in consumer preferences, using machine learning clustering methods to stratify user groups, and determining the demand preference feature sets of different groups include: Based on the trend decomposition result set, purchase frequency, purchase amount and category preference data are obtained. User behavior records under time distribution are classified and organized, and compared with a pre-established database to determine the initial set of purchase preferences for each user. The initial set of consumption preferences is processed using clustering tools. Combining user age, income level, and regional differences, if the consumption frequency of a certain user group is higher than a preset threshold, it is classified into the high-frequency consumption category, thus obtaining the stratified user group segmentation results. Based on the user group segmentation results, purchase channels, promotion sensitivity, and brand loyalty data for each group are obtained. Data comparison tools are used to analyze the differences in payment methods and return rates to determine the combination of demand and preference characteristics of different groups. Based on the combination of demand and preference characteristics and the time distribution attributes, data visualization tools are used to generate trend charts of consumption preference changes for each group, thus determining the final distribution of group characteristics.

8. A big data-based footwear ERP data management and control system, used to execute the big data-based footwear ERP data management and control method as described in any one of claims 1 to 7, characterized in that, The big data-based footwear ERP data management system includes: The standardized dataset acquisition module is used to acquire multi-source data from sales terminal data, production equipment information and external market dynamics. In view of the heterogeneous characteristics of the multi-source data, the module performs format standardization processing through pre-established mapping rules to obtain a standardized dataset. The global data view determination module is used to address the information silos in the standardized dataset by using distributed storage technology for centralized management and linking dispersed data sources through cross-channel data association algorithms to determine a unified global data view. The historical data feature set acquisition module is used to extract feature fields for historical data analysis based on the global data view, and to clean the missing and outlier values ​​in the feature fields by applying interpolation techniques and anomaly detection mechanisms to obtain a complete historical data feature set. The comprehensive dataset acquisition module is used to obtain time-related information from the complete historical data feature set. Combined with the real-time data update stream, if the real-time data update frequency is lower than the preset threshold, a supplementary recording mechanism is triggered to obtain a synchronized comprehensive dataset. The trend decomposition result set acquisition module is used to analyze changes in consumer preferences and seasonal fluctuations in a synchronized comprehensive dataset. It performs trend decomposition using time series analysis methods to determine the distribution patterns of short-term fluctuations and long-term trends, and obtains the trend decomposition result set. The demand preference feature set determination module is used to extract key indicators of changes in consumption preferences based on the trend decomposition result set, and to use machine learning clustering methods to stratify user groups and determine the demand preference feature sets of different groups. The resource allocation suggestion generation module is used to obtain core variables from the demand preference feature set, construct the input features of the prediction model, obtain the demand prediction output results through model iterative calculation, and generate resource allocation suggestions, which are then transmitted to the production scheduling system.

9. The footwear ERP data management and control system based on big data as described in claim 8, characterized in that, The normalized dataset acquisition module includes: The data set acquisition unit is used to acquire multi-source data from sales terminals, production equipment information and external market dynamics. In view of the heterogeneous characteristics of the multi-source data, it performs preliminary cleaning using pre-established mapping rules. By comparing the multi-source data fields with the preset format template, a preliminary data set is obtained. The intermediate dataset acquisition unit is used to perform field uniform conversion using a format standardization tool on the initially organized dataset. If the field values ​​in the dataset do not conform to the preset threshold range, they are marked as abnormal and completed to obtain an intermediate dataset with a consistent format. The comprehensive dataset acquisition unit is used to determine the integrated comprehensive dataset by matching the associated fields of the sales terminal and equipment information with the data integration tool based on the intermediate dataset, and by aligning the data with the external market dynamics data through timestamps. The normalized dataset acquisition unit is used to check the completeness of the comprehensive dataset using data verification tools. If there are missing items, they are filled in with historical data to obtain the final normalized dataset.

Images