Demand prediction method based on intelligent pre-packaging
By analyzing store sales data and combining machine learning algorithms, we predict the optimal pre-package weight and quantity of fresh products, solving the problem that is difficult to accurately predict in the existing technology, and achieving efficient packaging strategy adjustment and market adaptability.
Patent Information
- Application Number
- CN202411779335.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-05
- Publication Date
- 2025-04-25
AI Technical Summary
The prior art is difficult to accurately predict the optimal pre-package weight and quantity of fresh products, resulting in waste of goods, backlog of inventory or out of stock.
Through mining and analysis of store sales data, combined with machine learning algorithms (such as time series prediction, random forest, deep learning models, etc.), the optimal pre-package weight and quantity of each product are predicted, and the packaging strategy is dynamically adjusted to adapt to market changes.
It significantly improves packaging accuracy, reduces product waste and inventory backlog, reduces operating costs, improves consumer satisfaction, and can flexibly adapt to market changes.
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of commodity pre-packaging, and in particular to a demand forecasting method based on intelligent pre-packaging. Background Art
[0002] As consumer demand for fresh goods diversifies and supply chain management becomes increasingly complex, how to effectively pre-package fresh goods to meet consumer demand and reduce losses has become a key challenge for retailers and suppliers. Fresh goods usually have a short shelf life, and their sales volume is affected by many factors (such as seasonality, holidays, promotions, etc.), and demand fluctuates greatly. Therefore, accurately predicting the optimal pre-packaging weight and packaging quantity of fresh goods can not only help improve product sales efficiency, reduce inventory backlogs and waste, but also optimize supply chain management and improve customer satisfaction.
[0003] In the process of selling fresh goods, stores often face the problem of how to optimize the pre-packaging weight and packaging quantity of goods. Due to the short shelf life and large sales volatility of fresh goods, traditional packaging strategies (such as fixed packaging weight or quantity) are prone to cause product waste, inventory backlog or out-of-stock phenomenon. In response to this problem, the limitations of existing technologies are mainly reflected in the following aspects: (1) Disadvantages of fixed packaging strategy: Traditional pre-packaging usually adopts a fixed weight or quantity method. For example, a certain type of goods is pre-packaged according to a standard weight (such as 250 grams, 500 grams). Although this method is simple to operate, it ignores the differences in sales characteristics and consumer preferences of different stores. Some goods may have poor sales or even waste due to packaging specifications that are too large or too small. In addition, fixed packaging strategies are difficult to adjust flexibly when facing market demand fluctuations, which will cause inventory backlogs or out of stock.
[0004] (2) Insufficient use of sales data: Currently, most stores have accumulated a large amount of sales data on fresh products in their daily operations, such as sales volume, peak demand during certain time periods, and consumer preferences. However, most of this data has not been fully utilized to guide the optimization of pre-packaging strategies. For example, predicting the optimal pre-packaging weight and quantity of different products through historical data can significantly improve sales efficiency and reduce waste, but existing technologies have not yet provided an effective solution. (3) Complexity of demand forecasting: The demand for fresh products is affected by many factors, including seasonal changes, holidays, weather, promotional activities, etc. This complexity makes demand forecasting difficult. Traditional manual experience makes it difficult to accurately predict the optimal packaging specifications for each product. Fixed packaging design cannot adapt to rapidly changing market demand and is prone to oversupply or undersupply. Summary of the invention
[0005] The present invention mainly solves the deficiencies in the prior art and provides a demand forecasting method based on intelligent pre-packaging, which predicts the optimal pre-packaging weight and quantity of each commodity by mining and analyzing store sales data combined with a machine learning algorithm.
[0006] The above technical problems of the present invention are mainly solved by the following technical solutions: A demand forecasting method based on intelligent pre-packaging includes the following steps: Step 1: Collect and integrate data; use the store's sales system to collect data, including sales volume, timestamp, sales price, and promotion activity identification; associate different data sources and update data regularly every week through scheduled tasks. Different data sources include sales data, inventory data, and external environment data. Step 2: Process the data to achieve demand forecasting and evaluation; use machine learning algorithms to analyze the collected data, and determine the optimal pre-packaging weight and quantity for each product based on product demand and sales fluctuations through data processing, feature expansion and derivation, prediction model selection, prediction and evaluation operations. Machine learning algorithms include time series prediction, random forest, deep learning models, etc.
[0007] Step 3: Based on the prediction results, dynamically generate the packaging specifications for each product to achieve packaging optimization.
[0008] Step 4: Update the forecasting model regularly and continuously optimize the forecasting accuracy through the latest sales data; achieve dynamic learning to adapt to sales changes in different stores and at different times.
[0009] Preferably, the sales volume of each commodity in the most recent year is accumulated and counted through the data interface of the store sales system; the sales data of each commodity is aggregated and counted according to different time windows; the standard deviation of each commodity in one year is calculated; and holiday information and promotional activities are arranged through the computer calendar interface.
[0010] As a preference, weather data is obtained through an API interface; the current inventory quantity of each product is obtained through a real-time synchronization inventory system, and future inventory changes are predicted based on sales. Weather data includes temperature, humidity, precipitation, etc.
[0011] As a preference, data processing uses values such as mean and median to fill in missing sales data for products; uses box plots to detect outliers in product sales data, and uses z-scores to process outliers after detection. The presence of outliers can lead to deviations in model predictions; and filters out duplicate data in the data to remove the impact of duplicate data caused by human errors or errors in data collection on the data end.
[0012] The z-score, also known as the standard score, is the difference between a number and the mean and divided by the standard deviation. In statistics, the standard score is the number of standard deviations that an observation or data point has above the mean of the observed or measured values.
[0013] Preferably, feature expansion derivation is performed by extracting time feature columns, extracting time features such as year, month, day, week, quarter, and whether it is a holiday or a working day; seasonal features are calculated based on time, date, etc., and weather features are generated by combining the impact of external weather data on sales volume. Data during promotional activities are marked to generate feature columns for promotional time.
[0014] As a preferred choice, the prediction model is selected to use the time series model autoregressive integrated moving average model and the seasonal difference autoregressive moving average model to train the processed data, and the test result is result1; then two machine learning algorithms, random forest and support vector machine, are used to train by combining multiple decision trees, and the test result is result2; deep learning (LSTM, GRU) is used for training, and the test result is result3; by comparing the results of different prediction models (result1, result2, and result3), the error method uses the comparison of root mean square deviation, mean absolute error and root mean square error to evaluate the best prediction model as the final prediction model.
[0015] As a preference, the number of pre-packaged items for each product and the weight of each pre-packaged item are obtained based on the model's prediction results; the number of packaging units is dynamically adjusted based on current inventory and forecasted demand; during promotions, the system recommends appropriate packaging specifications based on promotional strategies. Promotional strategies include full-price discounts, discounts, and packaged sales.
[0016] As a preferred option, when there is too much inventory, the ratio of large packages should be appropriately increased; when there is insufficient inventory, smaller packages should be used to avoid overproduction and inventory backlogs; the optimized packaging plan should be converted into specific production instructions and automatically transmitted to the manual packaging link to ensure accurate execution of packaging.
[0017] Preferably, the demand forecasting model in the system will be updated based on new data. After the model has been used for a period of time, the new data will be used to fine-tune the existing model instead of retraining it, reducing computing overhead.
[0018] The present invention can achieve the following effects: The present invention provides a demand forecasting method based on intelligent pre-packaging. Compared with the prior art, the method predicts the optimal pre-packaging weight and quantity of each commodity by mining and analyzing store sales data in combination with a machine learning algorithm.
[0019] Improve packaging accuracy: The system dynamically adjusts packaging weight and quantity according to actual needs, effectively avoiding the problem of unsalable goods or waste due to mismatched packaging specifications.
[0020] Reduce operating costs: reduce inventory backlogs and loss of fresh goods, optimize logistics and packaging resource allocation, and reduce supply chain management costs.
[0021] Improve consumer satisfaction: Provide diversified packaging specifications to meet the needs of different consumer groups and improve consumer shopping experience.
[0022] Flexible adaptation to market changes: The system can quickly respond to special needs such as holidays and promotions, and automatically adjust packaging strategies to ensure market competitiveness. DETAILED DESCRIPTION
[0023] The technical solution of the invention is further specifically described below through embodiments.
[0024] Embodiment: A demand forecasting method based on intelligent pre-packaging includes the following steps: Step 1: Collect and integrate data. Use the store’s sales system to collect data, including sales volume, timestamp, sales price, and promotion activity identifier. Update data from different data sources regularly every week through scheduled tasks.
[0025] Through the data interface of the store sales system, the sales volume of each product in the past year is accumulated and calculated; the sales data of each product is aggregated and counted according to different time windows; the standard deviation of each product within one year is calculated; and holiday information and promotion activities arrangements are collected through the computer calendar interface.
[0026] Obtain weather data through the API interface; obtain the current inventory quantity of each product through real-time synchronization with the inventory system, and predict future inventory changes based on sales.
[0027] Step 2: Process the data to realize demand forecasting and evaluation; through the calculation process of data processing, feature expansion and derivation, selection of forecasting model, forecasting and evaluation.
[0028] Data processing uses values such as mean and median to fill in missing sales data of products; uses box plots to detect outliers in product sales data, and uses z-scores to process outliers after detection. The existence of outliers will cause deviations in model predictions; filters out duplicate data in the data to remove the impact of duplicate data caused by human errors or data collection errors on the data end.
[0029] Feature expansion and derivation uses the extraction of time feature columns to extract time features such as year, month, day, week, quarter, and whether it is a holiday or a working day; seasonal features are calculated based on time, date, etc., and the impact of external weather data on sales volume is combined to generate weather features, mark data during promotional activities, and generate feature columns for promotional time.
[0030] The prediction model is selected to use the time series model autoregressive integral moving average model and the seasonal difference autoregressive moving average model to train the processed data, and the test result is result1; then two machine learning algorithms, random forest and support vector machine, are used to train by combining multiple decision trees, and the test result is result2, and deep learning (LSTM, GRU) is used for training, and the test result is result3; by comparing the results of different prediction models (result1, result2, and result3), the error method uses the comparison of root mean square deviation, mean absolute error and root mean square error to evaluate the best prediction model as the final prediction model.
[0031] Step 3: Based on the prediction results, dynamically generate the packaging specifications for each product to achieve packaging optimization.
[0032] The system obtains the pre-packaged quantity and weight of each product based on the model's prediction results. It dynamically adjusts the number of packaging units based on current inventory and forecasted demand. During promotions, the system recommends appropriate packaging specifications based on promotion strategies.
[0033] When there is too much inventory, increase the proportion of large packages appropriately; when there is insufficient inventory, use smaller packages to avoid overproduction and inventory backlogs; convert the optimized packaging plan into specific production instructions and automatically transmit them to the manual packaging link to ensure accurate execution of packaging.
[0034] Step 4: Update the forecasting model regularly and continuously optimize the forecasting accuracy through the latest sales data; achieve dynamic learning to adapt to sales changes in different stores and at different times.
[0035] The demand forecasting model in the system will be updated based on new data. After the model has been used for a period of time, the new data will be used to fine-tune the existing model instead of retraining, reducing computing overhead.
[0036] By mining and analyzing store sales data and combining it with machine learning algorithms, the optimal pre-packaging weight and quantity for each product are predicted. The core functions of the system include: (1) Collect historical sales data of products, such as sales volume, sales time, and seasonal characteristics.
[0037] (2) Build a demand forecasting model and dynamically adjust the forecast results based on environmental factors (weather, holidays, etc.).
[0038] (3) Based on the prediction results, design the optimal pre-packaging weight and quantity plan for each product. (4) Dynamically update the model and continuously optimize the pre-packaging strategy based on real-time sales data.
[0039] In summary, this demand forecasting method based on intelligent pre-packaging, through the mining and analysis of store sales data, combined with machine learning algorithms, predicts the optimal pre-packaging weight and quantity of each commodity. It can significantly reduce the waste of fresh goods, optimize store operation efficiency, and improve consumer satisfaction, providing an intelligent solution for the pre-packaging design of fresh goods.
[0040] The above description is only a specific embodiment of the present invention, but the structural features of the present invention are not limited thereto. Any changes or modifications made by any technician in the field of the present invention are included in the patent scope of the present invention.
Claims
1. A demand forecasting method based on intelligent pre-packaging, characterized in that The steps are as follows: Step 1: Collect and integrate data. Use the store's sales system to collect data, including sales volume, timestamp, sales price, and promotion activity identifier. Associate different data sources and update data regularly every week through scheduled tasks. Step 2: Process the data to realize demand forecasting and evaluation; through data processing, feature expansion and derivation, prediction model selection, prediction and evaluation calculation process; Step 3: Based on the prediction results, dynamically generate the packaging specifications for each product to achieve packaging optimization; Step 4: Update the forecasting model regularly and continuously optimize the forecasting accuracy through the latest sales data; achieve dynamic learning to adapt to sales changes in different stores and at different times.
2. The demand forecasting method based on intelligent pre-packaging according to claim 1 is characterized in that: Through the data interface of the store sales system, the sales volume of each product in the past year is accumulated and calculated; the sales data of each product is aggregated and counted according to different time windows; Calculate the standard deviation of each product within one year; collect holiday information and promotion activity arrangements through the computer calendar interface.
3. The demand forecasting method based on intelligent pre-packaging according to claim 2 is characterized in that: Obtain weather data through the API interface; obtain the current inventory quantity of each product through real-time synchronization with the inventory system, and predict future inventory changes based on sales.
4. The demand forecasting method based on intelligent pre-packaging according to claim 1 is characterized in that: Data processing uses values such as mean and median to fill in missing sales data of products; uses box plots to detect outliers in product sales data, and uses z-scores to process outliers after detection. The existence of outliers will cause deviations in model predictions; filters out duplicate data in the data to remove the impact of duplicate data caused by human errors or data collection errors on the data end.
5. The demand forecasting method based on intelligent pre-packaging according to claim 4 is characterized in that: Feature expansion and derivation uses the extraction of time feature columns to extract time features such as year, month, day, week, quarter, and whether it is a holiday or a working day; seasonal features are calculated based on time, date, etc., and the impact of external weather data on sales volume is combined to generate weather features, mark data during promotional activities, and generate feature columns for promotional time.
6. The demand forecasting method based on intelligent pre-packaging according to claim 5 is characterized in that: The prediction model is selected to use the time series model autoregressive integral moving average model and the seasonal difference autoregressive moving average model to train the processed data, and the test result is result1; then two machine learning algorithms, random forest and support vector machine, are used to train by combining multiple decision trees, and the test result is result2, and deep learning (LSTM, GRU) is used for training, and the test result is result3; by comparing the results of different prediction models (result1, result2, and result3), the error method uses the comparison of root mean square deviation, mean absolute error and root mean square error to evaluate the best prediction model as the final prediction model.
7. The demand forecasting method based on intelligent pre-packaging according to claim 1 is characterized in that: The system obtains the pre-packaged quantity and weight of each product based on the model's prediction results. It dynamically adjusts the number of packaging units based on current inventory and forecasted demand. During promotions, the system recommends appropriate packaging specifications based on promotion strategies.
8. The demand forecasting method based on intelligent pre-packaging according to claim 7 is characterized in that: When there is too much inventory, increase the proportion of large packages appropriately; when there is insufficient inventory, use smaller packages to avoid overproduction and inventory backlogs; convert the optimized packaging plan into specific production instructions and automatically transmit them to the manual packaging link to ensure accurate execution of packaging.
9. The demand forecasting method based on intelligent pre-packaging according to claim 1 is characterized in that: The demand forecasting model in the system will be updated based on new data. After the model has been used for a period of time, the new data will be used to fine-tune the existing model instead of retraining, reducing computing overhead.