Intelligent stationery supply chain optimization system based on big data analysis

Abstract

The invention discloses an intelligent stationery supply chain optimization system based on big data analysis, belongs to the field of big data analysis, and aims to solve the problem of how to construct a high-precision demand prediction model through real-time and accurate data acquisition and processing to quickly identify and deal with abnormal conditions. The system comprises a data acquisition module, a storage management module, an intelligent prediction module, a mining analysis module and a supply chain cooperation module. The Internet of Things automatically collects supply chain product data, uploads the data to the cloud platform, stores the data in a database after cleaning, and classifies and adaptively compresses the data according to access evaluation indexes. A demand prediction model is constructed based on historical data, periodical and triggered updating is carried out, and the order demand of a product is predicted. According to the method, historical data with frequent access is screened for abnormal mode identification and reason classification, dynamic adjustment and collaboration of supply chains are realized through an information sharing platform, inventory purchase is optimized, and decision support is provided for abnormal modes.

Description

Technical Field

[0001] The present invention relates to the field of big data analysis, and more specifically to an intelligent stationery supply chain optimization system based on big data analysis. Background Art

[0002] With the deepening of global economic integration, market competition in the stationery industry is becoming increasingly fierce. Consumer demand for stationery products is no longer limited to basic writing functions but is shifting towards diversified requirements such as personalization, high quality, and environmental friendliness. This shift in demand has shortened the lifecycle of stationery products and accelerated the pace of replacement, placing higher demands on supply chain flexibility and responsiveness. Furthermore, the stationery industry's supply chain involves multiple links, including raw material procurement, manufacturing, logistics and distribution, and point-of-sale (POS), each of which presents potential risks and uncertainties. For example, issues such as raw material price fluctuations, production equipment failures, and logistics and distribution delays can impact the stability and efficiency of the entire supply chain.

[0003] Against this backdrop, traditional supply chain management models have struggled to adapt to market changes and business needs. Traditional models often rely on manual experience and fixed processes, lacking real-time and dynamic capabilities. This results in companies being slow to respond to emergencies and unable to adjust their strategies promptly. Furthermore, information silos within traditional models severely hinder collaboration and optimization across supply chain links. Therefore, leveraging big data analytics to optimize the stationery supply chain has become an inevitable trend in the industry. Big data analytics can help companies achieve real-time monitoring and data analysis across all supply chain links, enabling timely identification of issues and appropriate adjustments. By analyzing historical data, companies can predict future market demand trends, providing a scientific basis for production planning and inventory management. Furthermore, big data analytics can help companies optimize logistics and delivery routes, reduce costs, and improve delivery efficiency.

[0004] The invention patent with publication number CN118505112A discloses a method, system, equipment and medium for intelligent supply chain prediction based on big data. It belongs to the field of supply chain management and big data analysis technology. The technical problem it solves is how to help enterprises better manage the supply chain, optimize operational efficiency and reduce costs. The technical solution is: data collection and preprocessing: obtain supply chain data by accessing multiple data sources, and use big data technology to preprocess the supply chain data; establish a supply chain prediction model: establish a supply chain prediction model based on historical data and machine learning algorithms, and analyze and predict key indicators of demand and inventory levels in the supply chain through the supply chain prediction model; real-time monitoring and risk warning: real-time monitoring of key indicators in the supply chain, and provide warning information of potential risks based on the supply chain prediction model and anomaly detection algorithm; intelligent decision-making; security and privacy protection.

[0005] While existing technologies have addressed the problem of how to better manage supply chains, optimize operational efficiency, and reduce costs, they still haven't solved the problem of how to build high-precision demand forecasting models through real-time, accurate data collection and processing, and quickly identify and respond to abnormal situations. Therefore, to overcome these limitations, this paper proposes an intelligent stationery supply chain optimization system based on big data analysis. Summary of the Invention

[0006] In response to the shortcomings of the existing technology, the present invention aims to provide an intelligent stationery supply chain optimization system based on big data analysis. This system solves the problem of how to build a high-precision demand forecasting model through real-time and accurate data collection and processing, and quickly identify and respond to abnormal situations. Through the Internet of Things, automated data collection is performed to collect product data from all links in the supply chain and upload it to a cloud platform. The cloud platform data is cleaned and stored in a database, and an adaptive compression algorithm is used to reduce storage space. Historical product data is used to build a demand forecasting model to predict future order demand, and the model is updated periodically and triggered to improve accuracy. In addition, based on the classification results, frequently accessed historical data can be filtered out for abnormal pattern recognition and cause analysis, thereby providing more comprehensive and timely support for supply chain management.

[0007] To achieve the above object, the present invention provides the following technical solutions: An intelligent stationery supply chain optimization system based on big data analysis, including data collection module, storage management module, intelligent prediction module, mining analysis module and supply chain collaboration module; The data collection module is used to automatically collect data through the Internet of Things, collect product data in the supply chain, and upload it to the cloud platform; The storage management module is used to clean the product data on the cloud platform, store the product data in a database, classify the product data by accessing the evaluation index values, and adaptively delete and compress the product data based on the classification results; The intelligent forecasting module is used to obtain product sales trends based on historical product data, build a demand forecasting model, predict product order demand, and automatically update the demand forecasting model through periodicity and triggering; The mining and analysis module is used to filter out frequently accessed historical product data from the database based on the classification results of product data, identify abnormal patterns in this historical product data through the demand forecast model, and classify the causes of the identified abnormal patterns; The supply chain collaboration module is used to achieve dynamic adjustment and collaborative work among various links in the supply chain through an information sharing platform, optimize inventory and procurement through demand forecasting models, and provide decision support for abnormal patterns.

[0008] Specifically, the storage management module includes a data cleaning unit and a data storage unit. The data storage unit is configured with a data compression strategy. The data compression strategy is used to adaptively adjust the storage granularity and retention period of the data according to the access frequency of the product data, and use a compression algorithm to compress and store historical product data.

[0009] Specifically, data compression strategies include: Configure an update threshold to trigger adaptive pruning and compression of product data based on storage space usage. When storage space usage exceeds the update threshold, access frequency information of product data in the database is retrieved. Set an observation period, filter the access frequency information of product data during the observation period, and construct an access evaluation index based on the access frequency information. The calculation formula for the access evaluation index is as follows:

[0010] in, To access the evaluation metric values, is the total number of days in the observation period, is the total number of visits, is the number of independent visiting users, is the standard deviation of the access frequency change, is the recent visit frequency, The product data during the observation period Number of visits per day, The product data during the observation period The number of independent visitors per day, For the observation period Number of visits to the day The change in the number of visits per day, is the total number of days in the recent period, 、 、 and is a non-negative weight coefficient; Configure classification thresholds, including high and low classification thresholds. If the access evaluation index of product data is greater than the high classification threshold, it is marked as hot data. If the access evaluation index of product data is less than the low classification threshold, it is marked as cold data. Otherwise, it is marked as warm data.

[0011] Specifically, the data compression strategy also includes: Filter out product data marked as warm data and reduce the storage granularity of product data, including time series data optimization, unstructured data streamlining, and database-level optimization; Filter out product data marked as cold data, compress the product data using a compression algorithm, migrate the compressed cold data to a long-term storage solution, and create an index; Set a retention period. If the access evaluation index is higher than the low classification threshold within the retention period, the data will be reactivated and migrated back to the database. Otherwise, it will continue to be stored in the long-term storage solution and automatically deleted after the retention period expires.

[0012] Specifically, the intelligent prediction module includes a model building unit and a model updating unit. The model building unit is configured with a periodic prediction strategy, which is used to periodically predict future demand based on the demand prediction model. The model updating unit is configured with a trigger update strategy, which is used to update the demand prediction model when a trigger condition is detected and predict future order demand.

[0013] Specifically, the construction of the demand forecasting model includes: Configuring the Prediction Time Threshold , used to measure the time length of demand forecasting, obtain the current update time of the model, filter historical product data and historical external data within the forecast time threshold from the current update time from the database, and build a historical data set; Configure the sampling interval to measure the time interval for demand forecasting. The historical data set is aggregated based on the sampling interval and the historical product data of the sampling points is summarized to construct the forecast data set. Preprocess the prediction data set, including data cleaning, data conversion and data normalization; Perform feature extraction on the preprocessed prediction data set to extract key features, including historical order demand, inventory, and product quantitative sentiment value; The product sentiment quantification value is calculated by predicting the customer feedback information on various products in each region in the data set, and using natural language processing to perform sentiment analysis to obtain the sentiment tendency of customer feedback at each sampling point, thereby outputting the quantified sentiment value of each sampling point.

[0014] Specifically, the construction of the demand forecasting model includes: Based on the key features of the forecast data set, a basic demand forecast model is constructed, parameter estimation is performed, and model parameters are obtained. The expression of the demand forecast model is as follows:

[0015] in, It is The product demand forecast value of the sampling points, It is The actual order demand for products at each sampling point, is the prediction time threshold The number of sampling points used for product demand forecasting, yes The random error term of the day represents the part that the model fails to capture, It is The product inventory of each sampling point, It is The product of each sampling point quantifies the sentiment value. is the time threshold used to measure the product satisfaction impact day, and , and is the non-negative weight coefficient of the demand forecasting model for the autoregression of time series data, is the non-negative weight coefficient of product inventory, It is the non-negative weight coefficient of the product's quantitative sentiment value; The demand forecasting model is fitted according to the forecast data set of each product in each region, and the forecast demand of each product in each region is obtained based on the fitted demand forecasting model, thereby calculating the total forecast demand of each product.

[0016] Specifically, the steps to trigger the update strategy include: Based on the output of the demand forecast model and the actual order demand, calculate the forecast deviation and cumulative deviation of the demand forecast model, and standardize the forecast deviation; Perform trend analysis on the forecast deviation and calculate the change rate and acceleration of the forecast deviation, that is, calculate the first-order difference and second-order difference of the forecast deviation; Based on the immediate impact of forecast deviation, the long-term cumulative effect, and the speed and acceleration of trend changes, an updated evaluation index is constructed. The calculation formula for the updated evaluation index includes:

[0017] in, The current demand forecasting model is used to predict the The updated evaluation index value of the sampling point is It is The prediction deviation of the sampling points, It is The normalized prediction deviation of the sampling points is It is the The cumulative value of all prediction deviations at sampling points, It is The first-order difference of the prediction deviation of the sampling points, It is The second-order difference of the prediction deviation of the sampling points, 、 、 and are all non-negative weight parameters, It is The actual order demand of each sampling point, It is The output of the demand forecast model for each sampling point; Configure the trigger threshold. If the update evaluation index value is greater than the trigger threshold, the model update is triggered, the demand update model is refitted, and the model parameters are trained. Otherwise, the order sales volume for the next moment is predicted based on the current demand forecast model.

[0018] Specifically, the mining and analysis module includes an anomaly detection unit and a cause classification unit. The anomaly detection unit is configured with an anomaly identification strategy, which is used to automatically monitor the deviation between the predicted value of the demand forecasting model and the actual order demand, and identify abnormal situations of product orders. The cause classification unit is configured with an association analysis strategy, which is used to determine the type of abnormal situation through association analysis of the identified abnormal situation.

[0019] Specifically, the anomaly identification strategy includes: Calculate the deviation between the demand forecast model's predicted value and the actual sales value, and smooth the deviation using the weighted moving average method to identify long-term trends and cyclical fluctuations, that is:

[0020] in, It is The weighted shift value of the deviation of the sampling points, is the number of days used for weighted moving average, It is The deviation of the sampling points, It is The non-negative weight coefficient corresponding to the deviation of each sampling point; Configure the deviation threshold. When the weighted moving value of the deviation of the first sampling point is greater than the deviation threshold, the sampling points are marked as abnormal points, otherwise they are marked as normal points.

[0021] Beneficial effects of the present invention: 1. Automated data collection through IoT technology ensures real-time and accurate data. This reduces manual data entry errors and delays, improving data acquisition efficiency. Collected data is cleaned and classified, and adaptive data compression strategies are used to optimize storage space, saving storage costs and increasing data processing speed.

[0022] 2. Utilizing historical data to build demand forecasting models, we can accurately predict future sales volumes. Through periodic and triggered update mechanisms, forecasts are dynamically adjusted to address market changes. This enables more precise production and inventory planning, preventing oversupply or shortages. Anomaly detection promptly identifies sales anomalies and, through correlation analysis, identifies the causes, helping to implement corrective measures early on and mitigate losses.

[0023] 3. The mining and analysis module can automatically monitor and identify anomalies in sales and take timely countermeasures to quickly respond to market changes. The supply chain collaboration module provides decision support, allowing more informed strategic decisions to be made based on real-time data, thereby enhancing corporate competitiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a schematic diagram of the structure of the intelligent stationery supply chain optimization system based on big data analysis; Figure 2 A flowchart of the specific steps of the data compression strategy; Figure 3 A flowchart of the specific steps for building a demand forecasting model; Figure 4 A flowchart of the specific steps for triggering the update strategy; Figure 5 A flowchart of the specific steps of the anomaly identification strategy. DETAILED DESCRIPTION

[0025] See also Figure 1 ,This embodiment introduces an intelligent stationery supply chain optimization system based on big data analysis,,including: data acquisition module, storage management module, intelligent prediction module,,mining analysis module and supply chain collaboration module; Preferably, the data collection module is used to automatically collect data through the Internet of Things, collect product data from the supply chain, and upload it to the cloud platform. The supply chain includes production, warehousing, logistics, and sales. The product data includes product inventory levels, order status, shipping progress, and customer feedback. In this embodiment, the data acquisition module collects product data from various links in the smart stationery supply chain. This data collection is achieved through sensors, RFID tags, an IoT gateway, and mobile devices, covering production, warehousing, logistics, and sales. Sensors are deployed on the production line to monitor the status and output of production equipment in real time. RFID tags are attached to the packaging of smart stationery products to track inventory levels and shipping progress. The IoT gateway connects the sensors and RFID tags, uploading the collected data to a cloud platform. Mobile terminals are used by sales staff to record customer feedback. The supply chain includes production, warehousing, logistics, and sales. In the production stage, sensors are installed on the production line to monitor the operating status and output of production equipment in real time, transmitting this data to the cloud platform via the IoT gateway. In the warehousing stage, RFID tags are attached to the packaging of each stationery product. Readers in the warehouse regularly scan RFID tags, updating inventory level data, and uploading this data to the cloud platform via the IoT gateway. During the logistics process, RFID readers on logistics vehicles scan RFID tags on goods, track transportation progress in real time, and transmit the data to the cloud platform. During the sales process, sales staff use mobile terminals to input customer feedback, including product quality evaluations and market demand changes. This product data is then transmitted to the cloud platform via the mobile network. All collected product data is packaged in a standardized format and transmitted to the cloud platform using secure protocols to ensure data integrity and security.

[0026] Preferably, the storage management module is used to clean the product data of the cloud platform and then use the database to store the product data to eliminate invalid or abnormal records, ensure the accuracy and reliability of the data in the subsequent analysis process, and support the long-term preservation and fast query functions of historical data, and have the ability to adaptively delete and compress product data to optimize storage space utilization. In addition, it also has data backup and recovery functions.

[0027] In this embodiment, the storage management module is used to cleanse, store, and manage product data in the smart stationery supply chain, ensuring high data quality and reliability while optimizing storage space utilization. The cloud platform receives raw data from the data acquisition module. It performs a preliminary check to verify data integrity and consistency, identifying and marking possible errors or abnormal records. Data marked as abnormal is further verified to determine whether to correct, delete, or retain it. Data from different sources and formats is converted to a unified standard format to facilitate subsequent processing and analysis. A database system is used to store cleaned data, supporting high-concurrency access and horizontal scalability. Indexes are established for commonly used query fields to improve query efficiency. Data is partitioned based on timestamps or other attributes for easier management and faster retrieval. Data storage granularity and retention period are automatically adjusted based on data access frequency and importance. Historical data is compressed and stored using a compression algorithm to reduce storage space usage. Infrequently used historical data is regularly moved to low-cost storage media to free up primary storage space. Sensitive data is encrypted for storage to prevent unauthorized access. Regular full and incremental database backups are performed to ensure rapid recovery in the event of data loss or corruption. Establish a comprehensive disaster recovery plan, including data backup, failover, and business continuity strategies.

[0028] Preferably, the storage management module includes a data cleaning unit and a data storage unit. The data cleaning unit is configured with a data verification strategy, which is used to check the integrity and consistency of product data and identify and mark errors or abnormal records; the data storage unit is configured with a data compression strategy, which is used to adaptively adjust the storage granularity and retention period of product data according to the access frequency of product data, and use a compression algorithm to compress and store historical product data.

[0029] See also Figure 2 , preferably, the data compression strategy specifically includes: Configure an update threshold to trigger adaptive pruning and compression of product data based on storage space. When storage space usage exceeds the update threshold, access frequency information for product data in the database is retrieved. Access frequency information includes the total number of accesses, the most recent access time, and the number of accessing users. Adaptive pruning and compression dynamically adjusts storage strategies through machine learning algorithms, including training an LSTM model based on historical access frequency data to predict future data access trends. It also calculates data hot and cold migration thresholds in real time based on storage space usage. A sliding window mechanism is used to update storage granularity parameters every 24 hours. Set an observation period, filter the access frequency information of product data during the observation period, and construct an access evaluation index based on the access frequency information to determine the frequency of product data access. The calculation formula for the access evaluation index is as follows: in, To access the evaluation metric values, is the total number of days in the observation period, The total number of visits, that is, the sum of the number of visits during the observation period, is used to measure the average number of visits per unit time and reflect the access popularity of the data. is the number of independent visiting users, that is, the number of unique users who visited during the observation period. It measures the average number of independent visiting users per unit time and reflects the user coverage of the data. is the standard deviation of the access frequency change, which is used to measure the change in access frequency and reflect the stability and volatility of data access. It is used to measure the recent access frequency and reflect the access trend of data in the short term. The product data during the observation period Number of visits per day, The product data during the observation period The number of independent visitors per day, For the observation period Number of visits to the day The change in the number of visits per day, is the total number of days in the recent period, 、 、 and It is a non-negative weight coefficient used to adjust the contribution of different factors to the overall index.

[0030] Configure classification thresholds, including high classification thresholds and low classification thresholds, to distinguish the frequency of product data access. If the access evaluation index of product data is greater than the high classification threshold, it is marked as hot data. If the access evaluation index of product data is less than the low classification threshold, it is marked as cold data, otherwise it is marked as warm data. The high classification threshold and the low classification threshold are determined by the K-means clustering algorithm. The AEI values ​​of all data in the observation period are clustered into three categories according to the time series, namely hot, warm and cold. The high classification threshold is the center value of the hot data cluster plus the standard deviation, and the low classification threshold is the center value of the cold data cluster minus the standard deviation. The dimensionality reduction of warm data adopts the time series segmentation aggregation approximation method to compress the original data into a symbol sequence. The weight coefficient is obtained by collecting abnormal event samples with sudden changes in access frequency in historical data and optimizing them using genetic algorithms. 、 、 and , to maximize the sensitivity of AEI to abnormal events, set is the visit count weight, is the user number weight, is the volatility weight, is the recent trend weight; The data compression strategy is linked to AEI. When AEI exceeds the high classification threshold, the hot data caching strategy is triggered, and the LRU algorithm is used to retain the most recently accessed records. When AEI exceeds the low classification threshold, the cold data compression process is started, using GZIP+dictionary encoding compression. After the warm data is reduced in dimension, a data fingerprint is generated, and similarity queries can be performed in seconds.

[0031] Filter out product data marked as warm data and reduce the storage granularity of product data to save storage space, including time series data optimization, unstructured data streamlining, and database-level optimization. Time series data optimization includes using statistical sampling methods to store only data at key time points, such as daily start and end times. Unstructured data streamlining includes compressing and archiving generated log files, retaining only error logs and key operation logs, and cleaning up routine operation logs. For multimedia files such as product images and promotional videos, reduce the resolution or use more efficient encoding methods to reduce file size without affecting the user experience. Database-level optimization includes partitioning database tables based on product data creation time, migrating infrequently used historical data, and applying compression algorithms to text-type fields in database tables. Filter out product data marked as cold data and compress it using compression algorithms, including GZIP, BZIP2, and LZMA, to significantly reduce storage usage. Migrate the compressed cold data to long-term storage solutions, including object storage services and tape backups, to free up space in the primary database or file system while ensuring that the data remains retrievable when needed. Set a retention period. If the access evaluation index is higher than the low classification threshold within the retention period, the data will be reactivated and migrated back to the database. Otherwise, it will continue to be stored in the long-term storage solution. After the retention period, it will be automatically deleted to free up space.

[0032] Preferably, the intelligent forecasting module is used to obtain product sales trends based on historical product data, build a demand forecasting model, predict product order demand, and automatically update the demand forecasting model through periodicity and triggering; In this embodiment, the cleaned historical product data is obtained from the storage management module, including past demand records, inventory levels, order status, shipping progress, and customer feedback, and the collected data is subjected to feature extraction and conversion for use in model training. For example, time series data is converted into seasonal or trend features, or text data is converted into numerical features, and a prediction model is selected, including linear regression, decision tree, random forest, support vector machine, and neural network. The selected model is trained using historical product data to understand the trends and patterns of demand for smart stationery. The trained model is used to predict the demand for product orders in the future. It can be based on periodic forecasts, for example, weekly or monthly demand forecasts, or it can be based on forecasts triggered by specific events. In order to maintain the accuracy and adaptability of the model, the intelligent prediction module automatically updates the demand forecast model regularly or when a specific event occurs.

[0033] Preferably, the intelligent prediction module includes a periodic update unit and a trigger update unit. The periodic update unit is configured with a periodic prediction strategy, which is used to periodically predict future demand based on the demand prediction model. The trigger update unit is configured with a trigger update strategy, which is used to update the demand prediction model when a trigger condition is detected and predict future order demand.

[0034] Specifically, the construction of the demand forecasting model includes: Configuring the Prediction Time Threshold , used to measure the length of time used for demand forecasting, obtain the current update time of the model, filter historical product data and historical external data within the forecast time threshold from the current update time from the database, and build a historical data set; historical product data includes: order demand, inventory, and customer feedback information of various types of smart stationery; by setting an appropriate time threshold, effectively focus on the most recent historical data, ensure that the data used is most relevant to the current market environment, improve the accuracy of the forecast, extract historical product data that matches the current time point from the database, provide basic data for subsequent model construction, and ensure the timeliness and relevance of the data.

[0035] Configure the sampling interval to measure the time interval for demand forecasting. The historical data set is aggregated based on the sampling interval and the historical product data of the sampling points is summarized to build a forecast data set. By setting the sampling interval, the number of data points is reduced, and through data accumulation and aggregation, a streamlined historical data set is built. The historical data of the sampling points are summarized into a representative time period data, providing effective support for the subsequent construction of the forecast data set.

[0036] Preprocess the prediction data set, which includes data cleaning, data conversion and data normalization, to ensure the quality and consistency of the data. Through data cleaning, conversion and normalization, the consistency and quality of the input data are ensured, laying a good foundation for subsequent feature extraction and model training, and improving the learning effect of the model.

[0037] Feature extraction is performed on the preprocessed prediction data set to extract key features, including historical order demand, inventory and product quantitative sentiment value. The product quantitative sentiment value is calculated through customer feedback information on various products in each region in the prediction data set. Natural language processing is used to perform sentiment analysis to obtain the emotional tendency of customer feedback at each sampling point, and then output the quantitative sentiment value of each sampling point. By extracting key features, important input information is provided to the model, thereby improving the model's ability to capture demand changes.

[0038] Based on the key features of the forecast data set, a basic demand forecast model is constructed, parameter estimation is performed, and model parameters are obtained. The expression of the demand forecast model is as follows:

[0039] in, It is The product demand forecast value of the sampling points, It is The actual order demand for products at each sampling point, is the prediction time threshold The number of sampling points used for product demand forecasting, yes The random error term of the day represents the part that the model fails to capture, It is The product inventory of each sampling point, It is The product of each sampling point quantifies the sentiment value. is the time threshold used to measure the product satisfaction impact day, and , and is the non-negative weight coefficient of the demand forecasting model for the autoregression of time series data, is the non-negative weight coefficient of product inventory, It is the non-negative weight coefficient of the product's quantitative sentiment value; The product's quantitative sentiment value is calculated using the BERT model, and the Chinese stationery review dataset is used to fine-tune the BERT model. The sentiment score is calculated for each customer feedback, and the model parameters are trained using the Adam optimizer.

[0040] Fit the demand forecast model based on the forecast data set of each product in each region, and obtain the forecast demand of each product in each region based on the fitted demand forecast model, thereby calculating the total forecast demand of each product; Preferably, the specific steps of the regular forecasting strategy include: Configure a periodic update threshold to measure the time interval for demand forecasting model updates. When the model update interval is greater than the periodic update threshold, the demand forecasting model is updated to predict product demand changes. Otherwise, product demand changes are predicted according to the current demand forecasting model. By setting an appropriate time interval, the demand forecasting model is kept synchronized with market changes, reducing forecast errors caused by lags.

[0041] Continuously monitor the time interval since the last model update. When the time interval exceeds the set periodic update threshold, the model update process will be initiated and the model parameters will be re-estimated. Continuously monitor the time since the last update to ensure that the model update is triggered in a timely manner, thereby improving the response speed and enhancing the adaptability and real-time performance of the model.

[0042] After the periodic update conditions are met, the historical data set is re-screened and the demand forecast model is retrained using the historical data set; by regularly screening the latest historical data, it is ensured that the data on which the training model is based reflects current market trends, thereby improving the accuracy and reliability of the forecast.

[0043] After the demand forecasting model is updated, the updated demand forecasting model is used to forecast future demand, including generating forecast results, calculating confidence intervals, and identifying potential demand fluctuations. Using the updated model to forecast future demand can provide more accurate forecast results, help decision makers identify potential fluctuations, and thus optimize inventory management and production plans.

[0044] Record the performance data of the updated demand forecast model, including forecast accuracy and deviation analysis, to facilitate subsequent adjustments and optimizations. Recording the model's performance data provides a basis for subsequent adjustments and optimizations, ensuring continuous model improvement and maintaining high forecast accuracy and low deviations.

[0045] See also Figure 4 Preferably, the specific steps of triggering the update strategy include: Based on the output of the demand forecast model and the actual order demand, the forecast deviation and cumulative deviation of the demand forecast model are calculated to evaluate the accuracy and overall performance of the model. The forecast deviation is also standardized to make the forecast deviation at different time points comparable and eliminate the impact of dimension. Perform trend analysis on the forecast deviation and calculate the change speed and acceleration of the forecast deviation to capture the trend change, that is, calculate the first-order difference and second-order difference of the forecast deviation; Based on the immediate impact of forecast deviation, the long-term cumulative effect, and the speed and acceleration of trend changes, an update evaluation index is constructed to determine whether the Overture forecast model needs to be updated. The calculation formula for the update evaluation index includes:

[0046] in, The current demand forecasting model is used to predict the The updated evaluation index value of the sampling point is It is The prediction deviation of the sampling points, It is The normalized prediction deviation of the sampling points is It is the The cumulative value of all prediction deviations at sampling points, It is The first-order difference of the prediction deviation of the sampling points, It is The second-order difference of the prediction deviation of the sampling points, 、 、 and are all non-negative weight parameters used to balance the influence of each component. is the instantaneous bias weight, is the cumulative bias weight, is the deviation change speed weight, is the deviation acceleration weight, It is The actual order demand of each sampling point, It is The output of the demand forecast model for each sampling point; Configure a trigger threshold. If the update evaluation index value is greater than the trigger threshold, the model update is triggered, the demand update model is refitted, and the parameters are trained to improve the accuracy and reliability of the model prediction. Otherwise, the order demand quantity for the next moment is predicted based on the current demand forecast model.

[0047] Preferably, the mining and analysis module is used to filter out frequently accessed historical product data from the database based on the classification results of the product data, identify abnormal patterns of the historical product data through the demand forecasting model, and classify the causes of the identified abnormal patterns.

[0048] The mining and analysis module is used to filter frequently accessed historical product data from the database based on the product data classification results. The module then uses the demand forecasting model to identify abnormal patterns in this historical product data and classify the causes of these abnormal patterns. The demand forecasting model is then applied to analyze the historical data of frequently accessed products, predicting future demand trends based on historical access and order demand data. During the demand forecasting process, the module automatically monitors the deviation between the predicted and actual values ​​and identifies abnormal patterns, such as a sudden and significant increase or decrease in the actual number of visits to a particular product. For identified abnormal patterns, the module further analyzes the causes and generates a report categorizing the causes of these abnormal patterns, helping management understand the factors behind the fluctuations in access.

[0049] Preferably, the mining and analysis module includes an anomaly detection unit and a cause classification unit. The anomaly detection unit is configured with an anomaly identification strategy, which is used to automatically monitor the deviation between the predicted value of the demand forecasting model and the actual order demand, and identify abnormal situations of product orders. The cause classification unit is configured with an association analysis strategy, which is used to determine the type of abnormal situation through association analysis of the identified abnormal situation. See also Figure 5 ,Preferably, the anomaly identification strategy includes: Calculate the deviation between the demand forecast model's predicted value and the actual sales value, and smooth the deviation using the weighted moving average method to identify long-term trends and cyclical fluctuations, that is:

[0050] in, It is The weighted shift value of the deviation of the sampling points, is the number of days used for weighted moving average, It is The deviation of the sampling points, It is The non-negative weight coefficients corresponding to the deviations of each sampling point; smoothing the daily deviations through the weighted moving average method can eliminate the impact of short-term fluctuations and help identify long-term trends and cyclical fluctuations. This makes the analysis results more stable and predictable; Configure the deviation threshold. When the weighted moving value of the deviation of the first sampling point is greater than the deviation threshold, the Sampling points are marked as abnormal points, otherwise they are marked as normal points. After setting the deviation threshold, when the weighted moving value exceeds the threshold, it is marked as an abnormal day. This helps to quickly identify days that significantly deviate from normal sales performance and provides clear guidance for subsequent processing. Establish a real-time monitoring mechanism. When an abnormal day is detected, an alarm is immediately triggered and relevant information is recorded for subsequent analysis and processing, ensuring that relevant personnel can respond quickly and record important information for subsequent analysis. This enhances the ability to respond to abnormal situations in a timely manner; Regularly evaluate the results of the anomaly identification strategy and adjust the deviation threshold settings. Regularly evaluating and adjusting the deviation threshold can ensure the effectiveness and adaptability of the strategy and optimize the sensitivity and accuracy of anomaly identification.

[0051] Preferably, the association analysis strategy includes: Identify correlation patterns and potential correlations between different products. This includes analyzing the relationship between product sales data, inventory data, and shipping progress data to uncover potential causes of demand fluctuations, including: Frequent item set analysis identifies product combinations that frequently appear together within a specific time period to optimize product portfolios and manage inventory; Association rule mining uses Apriori and FP-growth algorithms to calculate association rules and reveal the demand correlation between different products, including: if the order demand for product A increases, then the order demand for product B may also increase; Identify abnormal patterns in association patterns and build anomaly detection models based on association rules and historical data sets to automatically detect demand patterns that are different from normal patterns; When an abnormal pattern is detected, the causes of the abnormal pattern are classified into different categories through correlation analysis, and the main influencing factors are identified, including Conduct causal inference based on the identified abnormal patterns and infer the possible causes of the abnormalities through correlation analysis, for example, inventory management issues, logistics delays, or changes in customer demand; Abnormal patterns are divided into different categories, including demand fluctuation anomalies, that is, drastic changes in product demand may be related to seasonal demand, promotional activities, etc.; inventory anomalies, that is, inventory is too low or too high, resulting in out-of-stock or inventory backlog; logistics anomalies, that is, transportation delays or errors, resulting in product failure to be delivered in time; customer feedback anomalies, that is, customer dissatisfaction with product quality or increased returns, indicating product quality problems or changes in market demand.

[0052] The identified abnormal patterns and their causes are classified to generate an abnormal pattern report, which includes an overview of the abnormal pattern, that is, describing the specific circumstances of the abnormal pattern, for example, abnormal fluctuations in product demand or abnormal product inventory; association rule analysis results, that is, displaying the association rules that lead to the abnormality, for example, if the inventory level of product A is low, the demand for product B may increase; abnormal cause classification, that is, listing the causes of various abnormalities and their possible influencing factors; recommended countermeasures, that is, based on the analysis results, proposing specific suggestions for optimizing inventory management, adjusting production plans or improving logistics distribution.

[0053] The supply chain collaboration module is optimized for coordinating and optimizing the operations of all supply chain links. It integrates real-time product data captured by the data acquisition module and leverages the data cleaning and classification capabilities of the storage management module to ensure data accuracy and timely updates. Simultaneously, it accesses demand forecasts from the intelligent forecasting module and adjusts production plans based on order demand trends to achieve a balance between supply and demand. When faced with sales anomalies identified by the mining and analysis module, the supply chain collaboration module responds quickly, adjusting production, logistics, or inventory strategies to mitigate the impact of the anomaly and developing improvement measures based on cause classification. Decision support is also provided to evaluate the pros and cons of different strategies and select the optimal solution.

[0054] In this embodiment, a smart stationery manufacturer deploys IoT sensors in key areas such as production lines, warehouses, and logistics to collect real-time product data, including production progress, inventory levels, and logistics status. This data is automatically uploaded to a cloud platform via a data collection module, allowing the supply chain collaboration module to access this data in real time. The storage management module cleans the collected raw data, removing errors and irrelevant data, correcting inconsistencies and omissions, and standardizing data formats from different sources. After cleaning, the data is classified according to pre-set rules, such as by product type, sales region, and time period. The intelligent forecasting module analyzes historical sales data and, based on factors such as historical order volume and inventory levels, constructs a demand forecasting model to predict product demand over a specific timeframe. The supply chain collaboration module automatically adjusts production plans based on the forecast results. This includes automatically increasing production for a particular stationery product if forecasts indicate increased demand. The mining and analysis module regularly analyzes sales data to identify unusual patterns, including sudden sales declines or increases. Once an anomaly is identified, the supply chain collaboration module initiates an emergency response mechanism. This includes, if a sudden drop in sales of a certain product is discovered, checking the production, logistics and inventory links, identifying the problem, and taking prompt measures, including adjusting production plans, speeding up logistics, or clearing excess inventory. The supply chain collaboration module has a built-in decision support system that provides a variety of supply chain optimization solutions, including changing suppliers, adjusting transportation routes, and adopting new technologies. Managers can choose the best solution by simulating the implementation effects of different solutions. This includes preparing inventory in advance before peak demand periods, or concentrating shipments when logistics costs are lower. Regularly generate supply chain performance reports, including key indicators such as inventory turnover, order fulfillment rate, and customer satisfaction. The supply chain collaboration module is highly flexible and adaptable, able to respond to market changes and emergencies, and ensure the stability and efficient operation of the supply chain by dynamically adjusting strategies.

[0055] Working principle and its effect: The intelligent stationery supply chain optimization system based on big data analysis involves the field of big data analysis, including data acquisition module, storage management module, intelligent prediction module, mining analysis module and supply chain collaboration module; The data acquisition module uses IoT technology to automatically collect product data from all links in the supply chain. This data includes product inventory levels, order status, shipping progress, and customer feedback. The storage management module receives raw data from the data acquisition module, performs preliminary inspection and cleaning, and removes invalid or abnormal records. It then converts the data into a unified standard format and stores it in the database. The storage management module also features an adaptive pruning and compression strategy, adjusting storage granularity and retention period based on data access frequency and importance. It also compresses historical data for optimal storage space utilization. It also provides data backup and recovery capabilities to ensure data security and reliability. The mining and analysis module, based on product data classification results, selects frequently accessed historical product data. Using a demand forecasting model, it identifies anomaly patterns in this data and further analyzes the causes of these anomalies. The anomaly detection unit uses a weighted moving average method to smooth deviations and identify long-term trends and cyclical fluctuations. The correlation analysis unit applies a clustering algorithm to group anomaly data, identify similar anomaly patterns, and verify causal relationships between different variables using causal inference. The intelligent prediction module draws on historical product data from the storage management module and constructs a time series analysis model to predict product order demand over a period of time. This module includes a periodic update unit and a triggered update unit, capable of automatically updating the demand forecast model based on set time intervals or specific trigger conditions to improve forecast accuracy and adaptability. The supply chain collaboration module integrates the data and analysis results of the aforementioned modules to coordinate the operations of all links in the supply chain. It adjusts production plans based on the demand forecast results of the intelligent forecast module to achieve supply and demand balance. It responds quickly to sales anomalies identified by the mining and analysis module, adjusting production, logistics, or inventory strategies to mitigate the impact of the anomalies. It also provides a decision support system to help managers evaluate the pros and cons of different strategies and select the optimal solution.

[0056] The intelligent stationery supply chain optimization system, based on big data analysis, achieves comprehensive monitoring and optimization of all supply chain links through modules such as real-time data collection, storage management, mining and analysis, intelligent forecasting, and supply chain collaboration. This system improves supply chain efficiency, reduces inventory costs, enhances market competitiveness, improves decision-making quality, and effectively reduces operational risks. Through precise demand forecasting and anomaly detection mechanisms, companies can better seize market opportunities, quickly respond to market changes, and ensure business continuity and data security. Overall, it provides companies with an efficient and intelligent supply chain management solution, helping them gain an advantage in the fiercely competitive market.

[0057] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. Intelligent stationery supply chain optimization system based on big data analysis, characterized by: It includes data collection module, storage management module, intelligent prediction module, mining and analysis module and supply chain collaboration module; The data acquisition module is used to automatically collect data through the Internet of Things, collect product data in the supply chain, and upload it to the cloud platform; The storage management module is used to clean the product data of the cloud platform, store the product data in a database, classify the product data by accessing the evaluation index values, and adaptively delete and compress the product data based on the classification results; The intelligent forecasting module is used to obtain product sales trends based on historical product data, build a demand forecasting model, predict product order demand, and automatically update the demand forecasting model through periodicity and triggering; The mining and analysis module is used to filter out frequently accessed historical product data from the database based on the classification results of the product data, identify abnormal patterns in the historical product data using the demand forecast model, and classify the causes of the identified abnormal patterns; The supply chain collaboration module is used to achieve dynamic adjustment and collaborative work between various links in the supply chain through an information sharing platform, optimize inventory and procurement through a demand forecasting model, and provide decision support for abnormal patterns.

2. The intelligent stationery supply chain optimization system based on big data analysis according to claim 1 is characterized in that: The storage management module includes a data cleaning unit and a data storage unit. The data storage unit is configured with a data compression strategy. The data compression strategy is used to adaptively adjust the storage granularity and retention period of the data according to the access frequency of the product data, and use a compression algorithm to compress and store historical product data.

3. The intelligent stationery supply chain optimization system based on big data analysis according to claim 2, characterized in that: The data compression strategy includes: Configure an update threshold to trigger adaptive pruning and compression of product data based on storage space usage. When storage space usage exceeds the update threshold, access frequency information of product data in the database is retrieved. Set an observation period, filter the access frequency information of product data during the observation period, and construct access evaluation indicators based on the access frequency information; Configure classification thresholds, including high and low classification thresholds. If the access evaluation index of product data is greater than the high classification threshold, it is marked as hot data. If the access evaluation index of product data is less than the low classification threshold, it is marked as cold data. Otherwise, it is marked as warm data.

4. The intelligent stationery supply chain optimization system based on big data analysis according to claim 3 is characterized in that: The data compression strategy also includes: Filter out product data marked as warm data and reduce the storage granularity of product data, including time series data optimization, unstructured data streamlining, and database-level optimization; Filter out product data marked as cold data, compress the product data using a compression algorithm, migrate the compressed cold data to a long-term storage solution, and create an index; Set a retention period. If the access evaluation index is higher than the low classification threshold within the retention period, the data will be reactivated and migrated back to the database. Otherwise, it will continue to be stored in the long-term storage solution and automatically deleted after the retention period expires.

5. The intelligent stationery supply chain optimization system based on big data analysis according to claim 1, characterized in that: The intelligent prediction module includes a model building unit and a model updating unit. The model building unit is configured with a periodic prediction strategy, which is used to periodically predict future demand based on the demand prediction model. The model updating unit is configured with a trigger update strategy, which is used to update the demand prediction model when a trigger condition is detected and predict future order demand.

6. The intelligent stationery supply chain optimization system based on big data analysis according to claim 5 is characterized in that: The construction of the demand forecasting model includes: Configuring the Prediction Time Threshold , used to measure the time length of demand forecasting, obtain the current update time of the model, filter historical product data and historical external data within the forecast time threshold from the current update time from the database, and build a historical data set; Configure the sampling interval to measure the time interval for demand forecasting. The historical data set is aggregated based on the sampling interval and the historical product data of the sampling points is summarized to construct the forecast data set. Preprocess the prediction data set, including data cleaning, data conversion and data normalization; Perform feature extraction on the preprocessed prediction data set to extract key features, including historical order demand, inventory, and product quantitative sentiment value; The product sentiment quantification value is calculated by predicting the customer feedback information on various products in each region in the data set, and using natural language processing to perform sentiment analysis to obtain the sentiment tendency of customer feedback at each sampling point, thereby outputting the quantified sentiment value of each sampling point.

7. The intelligent stationery supply chain optimization system based on big data analysis according to claim 6, characterized in that: The construction of the demand forecasting model includes: Based on the key features of the forecast data set, a basic demand forecast model is constructed, and parameter estimation is performed to obtain model parameters; The demand forecasting model is fitted according to the forecast data set of each product in each region, and the forecast demand of each product in each region is obtained based on the fitted demand forecasting model, thereby calculating the total forecast demand of each product.

8. The intelligent stationery supply chain optimization system based on big data analysis according to claim 5, characterized in that: The specific steps of triggering the update strategy include: Based on the output of the demand forecast model and the actual order demand, calculate the forecast deviation and cumulative deviation of the demand forecast model, and standardize the forecast deviation; Perform trend analysis on the forecast deviation and calculate the change rate and acceleration of the forecast deviation, that is, calculate the first-order difference and second-order difference of the forecast deviation; Construct updated evaluation indicators based on the immediate impact of forecast deviations, the long-term cumulative effects, and the speed and acceleration of trend changes; Configure the trigger threshold. If the update evaluation index value is greater than the trigger threshold, the model update is triggered, the demand update model is refitted, and the model parameters are trained. Otherwise, the order sales volume for the next moment is predicted based on the current demand forecast model.

9. The intelligent stationery supply chain optimization system based on big data analysis according to claim 1, characterized in that: The mining and analysis module includes an anomaly detection unit and a cause classification unit. The anomaly detection unit is configured with an anomaly identification strategy, which is used to automatically monitor the deviation between the predicted value of the demand forecasting model and the actual order demand, and identify anomalies in product orders. The cause classification unit is configured with an association analysis strategy, which is used to determine the type of anomaly through association analysis of the identified anomaly.

10. The intelligent stationery supply chain optimization system based on big data analysis according to claim 9, characterized in that: The anomaly identification strategy includes: Calculate the deviation between the demand forecast model's forecast and actual sales values ​​and smooth the deviation using a weighted moving average method to identify long-term trends and cyclical fluctuations; Configure the deviation threshold. When the weighted moving value of the deviation of the first sampling point is greater than the deviation threshold, the sampling points are marked as abnormal points, otherwise they are marked as normal points.

Images