Dynamic inventory replenishment method and system based on big data

Abstract

The invention relates to the technical field of inventory replenishment, and discloses a dynamic inventory replenishment method and system based on big data, and the method comprises the steps: collecting inventory, historical delivery, supplier performance and social media data. A periodic rule is determined through time sequence analysis, and demand fluctuation is analyzed through a sliding window method. And the delivery time and success rate of the supplier are extracted, and a performance quantized value is determined by combining demand fluctuation and a weighted average method. Analyzing the social media data, generating a hot commodity table, and performing cross validation to determine the replenishment priority. According to the demand fluctuation and the priority, the replenishment amount is optimized through linear programming, and a preliminary plan is generated; and reducing the risk by combining the performance quantized value and Monte Carlo simulation to obtain a low-risk plan. And finally, dynamically and finely adjusting the low-risk replenishment schedule to form a final replenishment execution scheme. According to the method, real-time network hotspot commodities and the supply capacity of suppliers are taken into consideration, so that the accuracy and adaptability of dynamic inventory replenishment are improved.

Description

Technical Field

[0001] The present invention relates to the field of inventory replenishment, and particularly to a dynamic inventory replenishment method and system based on big data. Background Art

[0002] In the retail industry, especially in fields with large demand fluctuations such as fresh food and clothing, the accuracy of inventory management is crucial for the operational efficiency and profitability of enterprises. For example, due to the short shelf life of fresh products, frequent replenishment is required to avoid out-of-stock or expired losses. In such scenarios, a dynamic inventory replenishment method can monitor the inventory level in real time and automatically adjust the replenishment plan according to sales data and market trends to ensure that there is always an appropriate amount of goods in stock.

[0003] In an existing method, inventory replenishment still adopts an inventory replenishment method based on historical sales data and simple prediction models. These methods analyze historical sales data and market trends to predict future demand and formulate replenishment plans accordingly. For example, some systems use statistical models and machine learning algorithms, such as time series analysis and regression analysis, to improve the accuracy of demand prediction. Although these methods can improve the accuracy of replenishment to a certain extent, they lack a dynamic adjustment function and cannot be flexibly adjusted when real-time commodity demand changes, resulting in inaccurate predicted replenishment quantities. Summary of the Invention

[0004] The present invention provides a dynamic inventory replenishment method and system based on big data to solve the problem that the existing methods lack a dynamic adjustment function and cannot be flexibly adjusted when real-time commodity demand changes, resulting in inaccurate predicted replenishment quantities.

[0005] In a first aspect, to solve the above technical problems, the present invention provides a dynamic inventory replenishment method based on big data, including: Obtaining inventory types and corresponding inventory quantities, historical outbound data, supplier fulfillment ability data, and social media data; Analyzing the historical outbound data through time series analysis to determine the periodic outbound pattern, and using a sliding window detection method to obtain the demand fluctuation result; Extracting the delivery time and success rate of items from the supplier fulfillment ability data, and combining with the demand fluctuation result to determine the fulfillment quantification value of the supplier fulfillment ability through the weighted average method; According to the text and images in the social media data, using natural language processing technology and convolutional neural network to analyze the potential hot commodity list; Cross-verifying the hot commodity list with the demand fluctuation result to determine the replenishment priority of different types of inventory; Optimize the replenishment quantities of various types of inventory according to the demand fluctuation results and the replenishment priorities through a linear programming algorithm to obtain a preliminary replenishment plan table; Combine the preliminary replenishment plan table and the fulfillment quantification value, and use the Monte Carlo simulation algorithm to perform multiple iterative calculations on the uncertainty of supplier fulfillment to obtain a low-risk replenishment plan table; Analyze the low-risk replenishment plan table through time series, and perform dynamic fine-tuning on the low-risk replenishment plan table to obtain a final multi-type inventory replenishment execution plan.

[0006] In an alternative implementation manner, the determining the periodic outbound rules by analyzing the historical outbound data through time series and obtaining the demand fluctuation results using the sliding window detection method includes: Process the historical outbound data through time series decomposition to extract the periodic outbound rules; Extract local data features using the sliding window technique according to the periodic outbound rules; Calculate the demand fluctuation amplitude within each window through the local data features, where the fluctuation amplitude is the difference between the maximum value and the minimum value within the window; Determine whether the demand fluctuation amplitude exceeds a preset amplitude threshold. If so, smooth the fluctuation amplitude by the moving average method. If not, do not perform any processing to obtain a stable demand sequence; Adjust the sliding window parameters according to the stable demand sequence until the fluctuation amplitude of the stable demand sequence is minimized to obtain an optimized demand fluctuation result.

[0007] In an alternative implementation manner, the extracting the delivery time and success rate of items from the supplier fulfillment ability data, and determining the fulfillment quantification value of the supplier fulfillment ability through the weighted average method in combination with the demand fluctuation results includes: Generate an initial fulfillment ability data set through data combination according to the delivery time and the success rate; Determine the weights of the delivery time and the success rate in the fulfillment quantification value for the initial fulfillment ability data set and the demand fluctuation results; For each supplier, multiply the delivery time and the success rate by the corresponding weights respectively and sum them to obtain the fulfillment quantification value.

[0008] In an alternative implementation manner, the cross-verifying the hot product table with the demand fluctuation results to determine the replenishment priorities of different types of inventory includes: Extract the matching features of the hot product table and the demand fluctuation results using the cross-comparison method to obtain a product demand association set; Use weighted calculation to process the types of the commodity demand association set and the inventory, and obtain a preliminary quantification value of the replenishment demand; According to the preliminary quantification value, arrange the inventory types through a sorting algorithm to generate a replenishment order list; According to the commodity demand association set, determine whether there are multiple types of commodities with similar association degrees in the replenishment order list. If so, adjust the order through the historical change trend of the demand fluctuation result to obtain an optimized order list; According to the optimized order list, calculate the difference between the expected inventory quantity of the inventory types with higher priorities and the real-time inventory quantity to generate the final replenishment priority.

[0009] In an alternative implementation, combining the preliminary replenishment plan table and the fulfillment quantification value, using the Monte Carlo simulation algorithm to perform multiple iterative calculations on the supplier fulfillment uncertainty to reduce potential risks and obtain a low-risk replenishment plan table, including: Use the Monte Carlo simulation algorithm to perform multiple iterative calculations on the standardized fulfillment quantification value to obtain a fulfillment uncertainty distribution; According to the preliminary replenishment plan table and the fulfillment uncertainty distribution, perform a secondary iterative calculation through the Monte Carlo simulation algorithm to obtain a risk assessment result; For the risk assessment result, use the mean-variance analysis method to screen the preliminary replenishment plan table to obtain a low-risk replenishment plan table.

[0010] In an alternative implementation, through time series analysis of the low-risk replenishment plan table, dynamically fine-tune the low-risk replenishment plan table to obtain a final multi-type inventory replenishment execution plan, including: Obtain the periodic change characteristics from the low-risk replenishment plan table through time series analysis to obtain a rhythm fluctuation distribution; Use a dynamic fine-tuning method to adjust the rhythm fluctuation distribution to obtain a corrected rhythm parameter; For the corrected rhythm parameter, divide the low-risk replenishment plan table through multi-category classification rules to obtain a set of type demands; Generate a multi-type inventory replenishment execution plan according to the set of type demands.

[0011] In an alternative implementation, after obtaining the final multi-type inventory replenishment execution plan, the method further includes: Input the inventory replenishment execution plan into an inventory management system, obtain replenishment process data through the inventory management system, and use a time window partitioning method to extract real-time tracking data to obtain a dynamic replenishment status; Analyze the fulfillment change trend according to the dynamic replenishment status, divide the supplier fulfillment status by using a clustering algorithm, and determine the fulfillment adjustment coefficient; Update the adjustment parameter set according to the fulfillment adjustment coefficient, update the product status through the hot product table, and obtain the classified demand change; Adjust the inventory demand classification according to the classified demand change, and smooth the replenishment execution plan by using the moving average method to obtain a preliminary correction plan; Integrate the inventory demand classification for the preliminary correction plan, and verify the replenishment process data through the inventory management system to obtain a real-time replenishment plan.

[0012] In a second aspect, the present invention provides a dynamic inventory replenishment system based on big data, mainly including: A data acquisition module for acquiring inventory types and corresponding inventory quantities, historical outbound data, supplier fulfillment ability data, and social media data; A time series analysis module for analyzing the historical outbound data through time series to determine the periodic outbound pattern, and obtaining the demand fluctuation result by using the sliding window detection method; A fulfillment ability quantification module for extracting the delivery time and success rate of items from the supplier fulfillment ability data, and combining with the demand fluctuation result to determine the fulfillment quantification value of the supplier fulfillment ability by using the weighted average method; A social media analysis module for analyzing the potential hot product table according to the text and images in the social media data by using natural language processing technology and convolutional neural network; A cross-validation module for cross-validating the hot product table with the demand fluctuation result to determine the replenishment priority of different types of inventory; A replenishment optimization module for optimizing the replenishment quantity of various types of inventory according to the demand fluctuation result and the replenishment priority by using a linear programming algorithm to obtain a preliminary replenishment plan table; A risk simulation module for combining the preliminary replenishment plan table with the fulfillment quantification value, and performing multiple iterative calculations on the supplier fulfillment uncertainty by using the Monte Carlo simulation algorithm to reduce potential risks and obtain a low-risk replenishment plan table; A dynamic adjustment module for dynamically fine-tuning the low-risk replenishment plan table by analyzing the low-risk replenishment plan table through time series to obtain a final multi-type inventory replenishment execution plan.

[0013] In a third aspect, the present invention further provides a computer-readable storage medium, the computer-readable storage medium includes a stored computer program, wherein, when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the above-mentioned big data-based dynamic inventory replenishment method.

[0014] In a fourth aspect, the present invention also provides a computer-readable storage medium, which includes a stored computer program. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute the dynamic inventory replenishment method based on big data described in any one of the above.

[0015] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: The present invention discloses a dynamic inventory replenishment method and system based on big data, and the beneficial effects include: (1) The present invention can monitor the inventory level in real time and dynamically adjust the replenishment plan according to real-time demand fluctuations, avoiding inventory backlogs or out-of-stock problems caused by demand changes. Through time series analysis and the sliding window detection method, real-time demand fluctuations are captured, and combined with the trend of hot products in social media data, the replenishment priority is dynamically adjusted. This real-time dynamic adjustment ability ensures that the replenishment plan can quickly respond to market changes, improving the flexibility and accuracy of inventory management.

[0016] (2) The present invention integrates inventory data, historical sales data, supplier fulfillment ability data, and social media data, comprehensively improving the accuracy of demand forecasting and the scientific nature of the replenishment plan. The present invention analyzes social media data through natural language processing and convolutional neural networks to identify potential hot products and cross-verify them with the results of demand fluctuations. This multi-dimensional data fusion can more comprehensively reflect market demand, reduce forecasting errors, and improve the accuracy of the replenishment plan.

[0017] (3) The present invention quantifies the supplier fulfillment ability through the weighted average method to ensure the reliability of the replenishment plan and reduce the inventory risk caused by supplier uncertainty. The present invention extracts the delivery time and success rate from the supplier fulfillment ability data and calculates the fulfillment quantification value through the weighted average method. This quantification method can objectively evaluate the supplier's fulfillment ability and avoid inventory shortages or delays caused by supplier fulfillment problems.

[0018] (4) The present invention performs multiple iterative calculations on the supplier fulfillment uncertainty through the Monte Carlo simulation algorithm to generate a low-risk replenishment plan table, reducing the risk of inventory management. The present invention combines the preliminary replenishment plan table and the supplier fulfillment quantification value, and performs multiple iterative calculations on the uncertainty of supplier fulfillment through the Monte Carlo simulation algorithm. This method can simulate the replenishment results under different scenarios, optimize the replenishment plan, and ensure the robustness of inventory management.

[0019] (5) The present invention dynamically fine-tunes the low-risk replenishment plan through time series analysis to further optimize the replenishment plan and ensure the accuracy and adaptability of the implementation plan. The present invention dynamically fine-tunes the low-risk replenishment plan through time series analysis and adjusts the replenishment quantity in combination with real-time data. This dynamic fine-tuning can ensure that the replenishment plan always remains consistent with the actual demand, improving the efficiency and accuracy of inventory management.

[0020] (6) The present invention analyzes social media data through natural language processing and convolutional neural networks, and can identify potential hot-selling products in advance, providing forward-looking guidance for the replenishment plan. The present invention analyzes the text and images in social media data to identify potential hot-selling products and cross-verifies them with the results of demand fluctuations. This ability to predict hot-selling products can help enterprises plan inventory in advance and avoid sales losses caused by out-of-stock of hot-selling products.

[0021] In summary, through improvements in aspects such as real-time dynamic adjustment, multi-dimensional data fusion, quantification of supplier fulfillment capabilities, risk control, dynamic fine-tuning optimization, and prediction of hot-selling products, the present invention solves the deficiencies of existing methods in terms of dynamics, accuracy, and risk control, and significantly improves the efficiency of inventory management. Brief Description of the Drawings

[0022] Figure 1 It is a flowchart of a dynamic inventory replenishment method based on big data according to the present invention.

[0023] Figure 2 It is a schematic structural diagram of a dynamic inventory replenishment system based on big data according to the present invention. Detailed Embodiments

[0024] Next, the technical solutions of the present invention will be described clearly and completely in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0025] Refer to Figure 1 , the first embodiment of the present invention provides a dynamic inventory replenishment method based on big data, specifically including the following steps: S101. Obtain the inventory types and corresponding inventory quantities, historical outbound data, supplier fulfillment capability data, and social media data; S102. Determine the periodic outbound pattern through time series analysis of the historical outbound data, and use the sliding window detection method to obtain the demand fluctuation result; S103. Extract the delivery time and success rate of the items from the supplier performance ability data, and combine with the demand fluctuation result to determine the performance quantification value of the supplier performance ability by the weighted average method; S104. According to the text and images in the social media data, use natural language processing technology and convolutional neural network to analyze the potential hot commodity list; S105. Cross-verify the hot commodity list with the demand fluctuation result to determine the replenishment priority of different types of inventory; S106. According to the demand fluctuation result and the replenishment priority, optimize the replenishment quantity of various types of inventory through the linear programming algorithm to obtain a preliminary replenishment plan table; S107. Combine the preliminary replenishment plan table and the performance quantification value, and use the Monte Carlo simulation algorithm to perform multiple iterative calculations on the uncertainty of supplier performance to obtain a low-risk replenishment plan table; S108. Analyze the low-risk replenishment plan table through time series, and perform dynamic fine-tuning on the low-risk replenishment plan table to obtain the final replenishment execution plan for multiple types of inventory.

[0026] In step S101, obtain the inventory types and the corresponding inventory quantities, historical outbound data, supplier performance ability data, and social media data.

[0027] It should be noted that the inventory type refers to the classification of various commodities or materials held in the warehouse. For example, in a retail enterprise, the inventory types include food, daily necessities, electronic products, etc.; the inventory quantity refers to the actual quantity of each inventory type at the current time point; the historical outbound data is a multi-dimensional array, including four data items: outbound time, outbound commodity type, outbound quantity, and outbound purpose; the supplier performance ability data is a multi-dimensional array, including five data items: supplier basic information, delivery time, success rate, order response time, and quality qualification rate. Among them, the supplier information includes basic information such as the supplier's name, contact information, and cooperation history. The delivery time refers to the time required for the supplier to receive the order and actually deliver the goods. The success rate refers to the ratio of the number of times the supplier delivers the goods on time to the total number of orders. The order response time refers to the time required for the supplier to receive the order, confirm the order, and start processing. The quality qualification rate refers to the proportion of the goods delivered by the supplier that meet the quality standards; the social media data is a multi-dimensional array, including five data items: text information, image content, release time, publisher information, and interaction data. The text information is the description content about the commodity. The image content refers to the image of the commodity. The publisher information includes the publisher's geographical location and user portrait, which is conducive to the enterprise to understand the demand differences of different regions and different user groups. The interaction data includes data on interaction behaviors such as likes, comments, and shares. The interaction data can reflect the popularity and market heat of the commodity.

[0028] In step S102, the historical outbound data is analyzed through time series analysis to determine the periodic outbound pattern, and the sliding window detection method is used to obtain the demand fluctuation result, including: The historical outbound data is processed by the time series decomposition method to extract the periodic outbound pattern; According to the periodic outbound pattern, the sliding window technique is adopted to extract the local data features; Based on the local data features, the demand fluctuation amplitude within each window is calculated, where the fluctuation amplitude is the difference between the maximum value and the minimum value within the window; It is judged whether the demand fluctuation amplitude exceeds the preset amplitude threshold. If so, the fluctuation amplitude is smoothed by the moving average method. If not, no processing is performed to obtain the stable demand sequence; According to the stable demand sequence, the sliding window parameters are adjusted until the fluctuation amplitude of the stable demand sequence is minimized to obtain the optimized demand fluctuation result.

[0029] It should be noted that the periodic outbound pattern refers to the regular repeated pattern presented by the inventory outbound data within a certain time range. This pattern is usually related to periodic events such as seasonal factors, holidays, promotional activities, weekdays and weekends. By analyzing the historical outbound data, these periodic patterns can be identified, so as to better predict future demand. By processing the historical outbound data with the time series decomposition method, the complex outbound data can be decomposed into three parts: trend, seasonality and residual, so as to more clearly understand the internal law of the data.

[0030] The formula of the time series decomposition method is as follows: Among them, is the time series data, is the trend component, is the seasonal component, is the residual component. The trend component is calculated using the moving average method, and the trend component is subtracted from the original data to obtain the combination of the seasonal component and the residual component , and the seasonal component is calculated by the seasonal average method, and the seasonal component is removed from the combined data to obtain the residual component .

[0031] Exemplarily, assume that a warehouse has recorded the monthly outbound volume for the past 36 months, and the data shows a year-on-year increase with a relatively high outbound volume in summer each year. After decomposition, the trend part shows that the outbound volume slowly increases over time, the seasonal part reveals that the outbound volume from June to August each year is about 20% higher than other months, and the residual is the random fluctuation after removing the trend and seasonality. This decomposition helps to observe the periodic repetition characteristics of the data.

[0032] It should be noted that local data characteristics refer to the data characteristics within a specific time period extracted from the periodic outbound rules through the sliding window technique. These characteristics can reflect the changes in the data within the local time period and are helpful for analyzing short-term fluctuations and trend changes. Local data characteristics include the maximum value, minimum value, average value of all data points within the sliding window, and standard deviation of data points within the sliding window.

[0033] Exemplarily, assume that the window size is set to 3 months and the sliding step is 1 month. For a certain segment of outbound data such as 100, 120, 150, the local characteristics within the window are assumed to be the mean of 123.3 and the range of 50.

[0034] Preferably, the window size can be adjusted according to business needs. For example, it can be shortened to 2 months to capture more detailed fluctuations or extended to 6 months to smooth the noise. The advantage of this method is that it not only retains the local details of the data but also facilitates subsequent fluctuation analysis.

[0035] In one embodiment, if the data of a certain window is 100, 130, 110, the fluctuation amplitude is 130 - 100 = 30. Assume that the preset amplitude threshold is 40. When the fluctuation amplitude does not exceed the threshold, the original data is retained; if the data of another window is 90, 150, 120 and the fluctuation amplitude is 60, which exceeds the threshold, smoothing processing is required. This judgment mechanism helps to identify and control abnormal fluctuations. By using the moving average method to smooth the fluctuation amplitude, the impact of data mutations can be effectively reduced. For example, for a window with a fluctuation amplitude of 60, calculate the average fluctuation of the previous 3 windows (such as 30, 40, 60) to get 43.3 as the smoothed value. This way can reduce the influence of random noise on the sequence and generate a more stable demand sequence, and its beneficial effect is to improve the robustness of prediction. Adjust the sliding window parameters according to the stable demand sequence, aiming to optimize the fluctuation amplitude.

[0036] For example, the initial window size is 3 months and the fluctuation amplitude of the stable sequence is 25. If the fluctuation drops to 15 after adjusting to 4 months, it indicates that the new parameter is better. This dynamic adjustment can adapt to the data characteristics and ensure that the final demand fluctuation result is closer to the actual business needs.

[0037] It can be understood that this method not only improves the prediction accuracy but also optimizes the resource allocation efficiency.

[0038] In step S103, extract the delivery time and success rate of the items from the supplier performance ability data, and combine the demand fluctuation results to determine the performance quantification value of the supplier performance ability by the weighted average method, including: Generate an initial performance ability data set through data combination according to the delivery time and the success rate; For the initial performance ability data set and the demand fluctuation results, determine the weights of the delivery time and the success rate in the performance quantification value; For each supplier, multiply the delivery time and the success rate by the corresponding weights respectively, and sum to obtain the performance quantification value.

[0039] It should be noted that the initial performance ability data set refers to a basic data set formed by collecting and sorting data such as the delivery time and success rate of suppliers for evaluating the performance ability of suppliers.

[0040] Exemplarily, assume that there are three suppliers A, B, and C in a certain warehouse, and the delivery time and success rate in the past 6 months are respectively recorded. The average delivery time of supplier A is 5 days, and the success rate is 95%; for supplier B it is 7 days, 90%; for supplier C it is 4 days, 98%. Pair and combine these data to form an initial performance ability data set, such as A(5, 95%), B(7, 90%), C(4, 98%), laying a foundation for subsequent analysis.

[0041] It should be noted that the weights of the delivery time and the success rate in the performance quantification value can be adjusted in combination with the demand fluctuation results. Assume that the demand fluctuation results show that the sensitivity to the delivery time is higher during the summer peak season, while the success rate is more concerned about in the off-season. In response to this situation, through historical data analysis, the weight of the delivery time in summer can be set to 0.7 and the success rate to 0.3; in the off-season, it is adjusted to 0.4 and 0.6. Such dynamic weights reflect the change in the priority of business needs and help to be closer to the actual scenario.

[0042] It should be noted that the process of determining the weights can also start from the business objectives. For example, if the warehouse attaches more importance to on-time delivery to reduce inventory backlog, the weight of the delivery time can be appropriately increased. Assume that in a certain analysis, the weight of the delivery time is set to 0.6 and the success rate to 0.4. For the data of supplier A, 5 days and 95%, assuming that the maximum delivery days of suppliers in the existing data is 10 days, when calculating, use (1 - 5 / 10)×0.6 + 95 / 100×0.4 to obtain the performance quantification value of 0.68; other suppliers are processed similarly. The advantage of this method is that the quantification result is more in line with the business focus.

[0043] It should be noted that the establishment of the initial performance ability data set and the weight adjustment are an iterative process.

[0044] In a feasible implementation, if it is found that the delivery time of a certain supplier continues to shorten, data can be recollected and weights adjusted to make the quantization value more time-sensitive. This method can improve the accuracy of decision-making and provide support for inventory management and cost control.

[0045] In step S104, according to the text and images in the social media data, natural language processing technology and convolutional neural network are used to analyze the potential hot product list, including: Extract the text data, image data, repost volume, and comment volume in the social media data, and use natural language processing technology to perform word segmentation and semantic parsing on the text data to obtain a preliminary semantic set; For the preliminary semantic set and the image data, extract visual features through a convolutional neural network to generate a product-related feature set; Use a weighted calculation method to process the product-related feature set to determine the initial value of the popularity of each product; According to the initial value of the product popularity, generate a dynamic adjustment value of popularity based on the associated repost volume and comment volume; Fuse the dynamic adjustment value of popularity with the initial value of the product popularity to obtain a quantified value of product popularity; For the quantified value of product popularity, use a sorting algorithm to rank the products by priority to generate a potential hot product list.

[0046] It should be noted that extracting text data, image data, repost volume, and comment volume from social media data is a basic step in analyzing product popularity. Exemplarily, assume that an e-commerce platform monitors Weibo data, and a post about a certain mobile phone contains the text "The new mobile phone takes super clear photos", a product picture, 500 reposts, and 200 comments. Using natural language processing technology to perform word segmentation and semantic parsing on the text, "takes super clear photos" can be split into "takes photos" and "clear", and a positive evaluation can be parsed to form a preliminary semantic set. The key to this step is to understand user sentiment and concerns and provide semantic support for subsequent analysis.

[0047] It should be noted that for image data, visual features are extracted through a convolutional neural network. For example, the above mobile phone picture contains elements such as a camera and a screen. After the network recognizes these features, a visual feature set related to the product is generated, such as "prominent camera" and "large screen". Combining with the preliminary semantic set "takes clear photos", it can be inferred that users are concerned about the camera function, forming a product-related feature set. The advantage of this method is to fuse text and image information to comprehensively reflect product characteristics.

[0048] In the embodiments of the present invention, a Transformer model is used to improve the accuracy of word segmentation and semantic parsing through the attention mechanism; the extraction of visual features is realized through a VGG convolutional neural network. It should be noted that after generating the commodity-related feature set, a weighted calculation method is adopted to determine the initial heat value. Assume that the positive sentiment weight of the semantic set is 0.6 and the attraction weight of the visual feature is 0.4. A certain mobile phone has a high score for "clear photo taking" and "prominent camera", so the initial heat value is relatively high. This weighting method can highlight user preferences and facilitate the preliminary screening of popular commodities.

[0049] It should be noted that the dynamic adjustment value of the heat is generated according to the number of forwards and comments, and the calculation formula is as follows: where T is the pre-normalized number of forwards, ranging from 0 to 100, is the pre-normalized number of comments, ranging from 0 to 100, 、 are the corresponding weight coefficients, is the dynamic adjustment value of the heat.

[0050] In the embodiments of the present invention, = 0.7, = 0.3.

[0051] It should be noted that the dynamic adjustment value of the heat generated according to the number of forwards and comments can reflect the dissemination power and discussion degree of the commodity. For example, 500 forwards indicate a wide dissemination range, and 200 comments show a high user participation degree. It can be set that the proportion of the number of forwards is 0.7 and the proportion of the number of comments is 0.3. After calculating the adjustment value, it can dynamically reflect the real-time heat change of the commodity. This adjustment can capture the instant feedback of social media and improve the timeliness of evaluation.

[0052] It should be noted that the dynamic adjustment value of the heat is fused with the initial heat value to obtain the quantified value of the commodity heat. For example, the initial heat value of a certain mobile phone is 80, and the dynamic adjustment value is 15. After adding them, the quantified value is 95. This value combines static features and dynamic dissemination and can more truly reflect the popularity of the commodity. Preferably, if the number of forwards of a certain commodity surges, the weight of the adjustment value can be temporarily increased to highlight its explosive potential.

[0053] It is understandable that for the quantified value of product popularity, a sorting algorithm is used to generate a table of potential hot products. In a feasible embodiment, if the quantified value of a certain mobile phone rises to the top of the list due to the continuous increase in the number of comments, it indicates that its degree of discussion has increased and it is worthy of key promotion. For example, the text of a post about a certain pair of headphones is "excellent sound quality", the picture shows a delicate design, the number of forwards is 300 times, and the number of comments is 150. Semantic analysis reveals "good sound quality", and the image feature is "strong sense of design", with a relatively high initial value. The dynamic adjustment value is slightly lower due to the stable number of forwards, but the final quantified value still ranks among the top. This multi-faceted analysis ensures the comprehensiveness and consistency of the evaluation, and helps to accurately lock in potential best-selling products.

[0054] In another feasible embodiment, if the number of negative reviews in the comments of a certain product increases, the positive weight can be reduced during semantic analysis, and the quantified value will be adjusted accordingly. This flexibility makes the evaluation closer to the real feedback of users and provides a basis for product optimization. Similarly, if the image feature shows damaged packaging, the visual score will decrease, which will also affect the final ranking. This comprehensive consideration improves the reliability of the decision-making.

[0055] In step S105, cross-validate the hot product table with the demand fluctuation result to determine the replenishment priorities for different types of inventory, including: Use the cross-comparison method to extract the matching features of the hot product table and the demand fluctuation result to obtain a set of product demand associations; Use weighted calculation to process the set of product demand associations and the types of inventory to obtain a preliminary quantified value of the replenishment demand; According to the preliminary quantified value, arrange the types of inventory through a sorting algorithm to generate a replenishment order table; According to the set of product demand associations, determine whether there are multiple types of products with similar degrees of association in the replenishment order table. If so, adjust the order through the historical change trend of the demand fluctuation result to obtain an optimized order table; According to the optimized order table, calculate the difference between the estimated inventory quantity of the inventory types with higher priorities and the real-time inventory quantity to generate the final replenishment priorities.

[0056] It should be noted that the set of product demand associations refers to a set of products with related demands by analyzing the demand relationships between products. The key to using the cross-comparison method to extract the matching features of the hot product table and the demand fluctuation result is to find the intersection of the two to form a set of product demand associations.

[0057] In the embodiment of the present invention, the Apriori algorithm is used to mine frequent item sets and association rules from the hot product table and the demand fluctuation result. The core idea of the Apriori algorithm is to find frequent item sets that meet the minimum support and confidence through the method of candidate set generation and layer-by-layer search For example, the hot product list shows that a certain mobile phone has a high popularity due to "clear photography", and the demand fluctuation result indicates that the recent demand of users for the photography function has increased. After cross-comparison, the "photography function" becomes the matching feature and is included in the commodity demand association set. This method can accurately lock in the commodity characteristics that users actually pay attention to.

[0058] In a feasible implementation manner, when processing the commodity demand association set and the inventory types, a preliminary quantitative value of the replenishment demand can be generated through weighted calculation. Specifically, assume that there are three mobile phones in the inventory, and the commodity demand association set shows that the "photography function" has a higher weight. A certain mobile phone has a high correlation due to its prominent camera, and its weight is set to 0.8; another mobile phone has a large screen but an average camera, and its weight is 0.5. Combining the inventory quantities, if there are 100 units left of the former and 300 units left of the latter, the preliminary quantitative value after weighting reflects the urgency of replenishment.

[0059] Exemplarily, the photography mobile phone with a high quantitative value ranks first, followed by the screen mobile phone. This ranking intuitively reflects the replenishment priority and facilitates quick decision-making.

[0060] It should be noted that if the correlation degrees of multiple types of commodities in the replenishment order list are similar, such as two mobile phones are both related to "photography" and their quantitative values are close, then the historical trend of the demand fluctuation result needs to be introduced to adjust the order.

[0061] In one embodiment, historical data shows that the demand for a certain mobile phone has been continuously increasing, while the demand for another mobile phone has tended to be stable. The priority of the former is increased, forming an optimized order list. This adjustment can capture the long-term trend and improve the rationality of replenishment.

[0062] Preferably, according to the optimized order list, calculate the difference between the estimated inventory quantity and the real-time inventory quantity of the inventory types with higher priorities to generate the final replenishment priority.

[0063] For example, the photography mobile phone is estimated to require 200 units, and the real-time inventory is 100 units, with a difference of 100 units, having the highest priority. If another mobile phone is estimated to require 150 units and the inventory is 140 units, with a difference of only 10 units, its priority is lower. This difference calculation can intuitively reflect the size of the gap and ensure the pertinence of replenishment.

[0064] It can be understood that if the demand fluctuation shows that a certain mobile phone is about to be promoted, the estimated inventory quantity can be appropriately enlarged, the difference is adjusted accordingly, and the priority is dynamically optimized. This flexibility makes replenishment closer to the actual market situation.

[0065] In step S106, according to the demand fluctuation result and the replenishment priority, optimize the replenishment quantities of various inventories through a linear programming algorithm to obtain a preliminary replenishment plan table.

[0066] It should be noted that the preliminary replenishment plan is a list of replenishment suggestions generated based on demand forecasting and the current inventory situation through optimization algorithms (such as linear programming). It details information such as the current inventory level, forecasted demand, suggested replenishment quantity, and replenishment priority for each item.

[0067] In the embodiment of the present invention, the linear programming objective function is defined as minimizing the sum of replenishment costs and storage costs. Demand satisfaction constraints, warehouse capacity constraints, and safety stock constraints are added, and the model is solved using a linear programming algorithm (such as the PuLP library) to obtain the optimal replenishment quantity for each item. Based on the solution results, a preliminary replenishment plan is generated, listing the suggested replenishment quantity for each item. The table includes information such as item name, current inventory level, forecasted demand, suggested replenishment quantity, and replenishment priority.

[0068] In step S107, the preliminary replenishment plan and the fulfillment quantification value are combined, and the Monte Carlo simulation algorithm is used to perform multiple iterative calculations on the supplier fulfillment uncertainty to reduce potential risks, resulting in a low-risk replenishment plan, including: The Monte Carlo simulation algorithm is used to perform multiple iterative calculations on the standardized fulfillment quantification value to obtain the fulfillment uncertainty distribution; Based on the preliminary replenishment plan and the fulfillment uncertainty distribution, a secondary iterative calculation is performed through the Monte Carlo simulation algorithm to obtain the risk assessment result; For the risk assessment result, the mean-variance analysis method is used to screen the preliminary replenishment plan to obtain a low-risk replenishment plan.

[0069] It should be noted that the preliminary replenishment plan is an initial replenishment plan formulated based on demand forecasting and inventory situation, and the fulfillment quantification value is a quantitative assessment of the supplier's fulfillment ability, reflecting the possibility of the supplier supplying goods on time and in full quantity.

[0070] It should be noted that the fulfillment uncertainty distribution refers to the probability distribution of the fulfillment quantification value obtained through multiple iterative calculations of the supplier's fulfillment ability using the Monte Carlo simulation algorithm. This distribution can reflect the uncertainty of the supplier's fulfillment ability.

[0071] Exemplarily, calculate the mean and variance of the fulfillment quantification value, define the number of simulations (e.g., 10,000 times), set the weight parameters, create an array or list to store the results of each simulation, and perform N iterations. In each iteration: randomly sample a value from the distribution of the fulfillment quantification value (e.g., using normal distribution sampling); calculate the impact of the sampled value on the replenishment plan; store the result in the result array. Based on the mean and variance of the sampled values, obtain the fulfillment uncertainty distribution and evaluate the risk of the preliminary replenishment plan.

[0072] For each replenishment plan in the preliminary replenishment plan table, combined with the fulfillment uncertainty distribution, perform a secondary iterative calculation. According to the mean and variance, screen out the replenishment plans with risk assessment values lower than the threshold to obtain a low-risk replenishment plan table.

[0073] Preferably, the combination of Monte Carlo simulation and mean-variance analysis can support replenishment decisions from multiple aspects.

[0074] In step S108, analyze the low-risk replenishment plan table through time series to dynamically fine-tune the low-risk replenishment plan table to obtain a final multi-category inventory replenishment execution plan, including: Obtain the periodic change characteristics from the low-risk replenishment plan table through time series analysis to obtain the rhythm fluctuation distribution; Adopt a dynamic fine-tuning method to adjust the rhythm fluctuation distribution to obtain corrected rhythm parameters; For the corrected rhythm parameters, divide the low-risk replenishment plan table through multi-category classification rules to obtain a set of category demands; Generate a multi-category inventory replenishment execution plan according to the set of category demands.

[0075] It should be noted that the rhythm fluctuation distribution refers to the periodic change characteristics extracted from the low-risk replenishment plan table through time series analysis. It reflects the fluctuation of replenishment demand in different time periods, including the amplitude, frequency, and regularity of the fluctuation. The category correction parameters refer to a series of parameters for adjusting the replenishment plan according to the replenishment demand and demand fluctuation of different categories of goods, such as the specific amount of increased replenishment demand, and also include the increase or decrease amount of the replenishment cycle and safety inventory level, aiming to improve the accuracy and adaptability of the replenishment plan.

[0076] Exemplarily, assume that the low-risk replenishment plan table covers 4 weeks, and the demands on Monday are 110, 115, 108, and 112 pieces respectively, with an average of about 111 pieces, while it is stable at about 85 pieces on Tuesday. This regular difference is the periodic characteristic. Adjusting the low-risk replenishment plan table according to the rhythm fluctuation distribution using the dynamic fine-tuning method is a flexible way to cope with cycle changes.

[0077] Specifically, for the dynamic fine-tuning method, according to the high-demand characteristic on Monday, the low-risk replenishment plan table can be adjusted from fixed daily replenishment to early replenishment on Monday. For example, the original plan was to replenish 90 pieces per day, but the analysis shows that the average demand on Monday is 111 pieces. The corrected rhythm parameter for Monday can be adjusted to replenish 110 pieces one day in advance, and keep 85 pieces on Tuesday. This not only matches the periodic demand but also avoids the cost of frequent adjustments. In a feasible implementation, the dynamic fine-tuning can also adjust the rhythm according to the peak and off-peak seasons. The replenishment quantity on Monday increases to 120 pieces in the peak season and decreases to 105 pieces in the off-peak season, reflecting the adaptability of the method.

[0078] It should be noted that multi-category classification is based on the sales speed or importance of goods. For example, fast-moving Class A goods such as daily necessities are replenished with 110 pieces on Monday, Class B goods such as seasonal clothing are replenished with 50 pieces, and Class C low-frequency goods are replenished with 20 pieces, forming a category demand set.

[0079] In one embodiment, Class A products are replenished first due to stable demand, while Class B products require additional buffer inventory due to large fluctuations. This classification ensures the accuracy of replenishment demand. Generating multiple types of inventory replenishment execution plans based on the type demand set is a step to convert the classification results into actual operations.

[0080] Preferably, the A-class products can be replenished daily, the B-class products can be replenished weekly, and the C-class products can be replenished on demand. For example, 110 items of the A-class products can be replenished on Monday and 90 items on Tuesday; 200 items of the B-class products can be replenished on Monday to cover the whole week; and 20 items of the C-class products can be replenished only when the inventory is less than 10 items.

[0081] After step S108, the method further includes: The inventory replenishment execution plan is input into the inventory management system, replenishment process data is obtained through the inventory management system, and real-time tracking data is extracted using a time window division method to obtain a dynamic replenishment status; Analyze the performance change trend according to the dynamic replenishment status, use a clustering algorithm to divide the supplier's performance status, and determine the performance adjustment coefficient; The adjustment parameter set is updated for the fulfillment adjustment coefficient, and the commodity status is updated through the hot commodity table to obtain the classification demand change; Adjust the inventory demand classification according to the changes in the classification demand, use the sliding average method to smooth the replenishment execution plan, and obtain a preliminary correction plan; The inventory demand classification is integrated with the preliminary revised plan, and the replenishment process data is verified through the inventory management system to obtain a real-time replenishment plan.

[0082] Specifically, obtaining replenishment process data through the inventory management system can be understood as extracting real-time inventory flow information from the system, such as the records of goods entering, leaving and current inventory in a certain e-commerce warehouse.

[0083] For example, the system shows that 100 items of a certain product were shipped out yesterday, 90 items were shipped in today, and the current inventory is 50 items. These data provide a basis for subsequent analysis. The time window division method is used to extract real-time tracking data. Specifically, the data is divided into fixed time periods, such as a window of 24 hours, to observe the daily replenishment situation.

[0084] In a feasible implementation, the warehouse divides a week into 7 windows and finds that the outbound quantity on Wednesday often reaches 120 pieces, while on Friday it is only 60 pieces. This division can clearly reflect the dynamic replenishment status.

[0085] According to the analysis of the dynamic replenishment status to analyze the trend of fulfillment changes, the timeliness of suppliers' deliveries can be extracted from the data. For example, a certain supplier promised to replenish 100 pieces on Wednesday, but actually only delivered 80 pieces, with a fulfillment rate of 80%. The clustering algorithm is used to divide the suppliers' fulfillment states. Preferably, the suppliers can be divided into three categories: high-fulfillment group (fulfillment rate exceeding 90%), medium-fulfillment group (70%-90%), and low-fulfillment group (below 70%).

[0086] In one embodiment, Supplier A has a fulfillment rate of 95% for 5 consecutive weeks and is classified into the high-fulfillment group; Supplier B's fulfillment rate fluctuates around 75% and is classified into the medium-fulfillment group. This classification provides a basis for subsequent adjustments. Update the parameter set according to the fulfillment adjustment coefficient.

[0087] It should be noted that the adjustment coefficient changes dynamically according to the fulfillment state. For example, the coefficient for high-fulfillment suppliers is set to 1.0, for medium-fulfillment to 0.9, and for low-fulfillment to 0.8.

[0088] In one embodiment, due to a fulfillment rate of 75% for Supplier B, its replenishment quantity is adjusted from 100 pieces to 90 pieces. Update the commodity status through the hot commodity table. Exemplarily, the daily sales volume of a certain best-selling commodity increases from 50 pieces to 70 pieces, and the status is updated from "regular" to "hot", reflecting the change in classification demand. Adjust the inventory demand classification according to the change in classification demand.

[0089] Specifically, commodities can be divided into three categories: high demand, medium demand, and low demand.

[0090] For example, hot commodities are classified as high demand, and the replenishment frequency is increased to daily; ordinary commodities are classified as medium demand and replenished weekly. The moving average method is used to smooth the replenishment execution plan.

[0091] In one implementation, the daily sales volumes of a certain commodity are 60, 70, and 80 pieces respectively, and the moving average value is 70 pieces. The preliminary correction plan adjusts the daily replenishment to 70 pieces to avoid sudden changes affecting stability. Integrate the inventory demand classification for the preliminary correction plan and verify the replenishment process data through the inventory management system.

[0092] It can be understood that when the system compares the actual inventory with the plan and finds that the inventory of high-demand commodities only remains 20 pieces, it immediately adjusts to replenish 80 pieces.

[0093] In one embodiment, after verification by the warehouse, it is found that the replenishment of 70 pieces on Monday is insufficient, and the implementation plan is adjusted to 90 pieces to ensure sufficient supply. This method significantly improves the accuracy and timeliness of replenishment, especially having advantages during peak periods.

[0094] Reference Figure 2 , the present invention provides a dynamic inventory replenishment system based on big data, mainly including: A data acquisition module, configured to acquire inventory types and corresponding inventory quantities, historical outbound data, supplier fulfillment ability data, and social media data; A time series analysis module, configured to determine the periodic outbound pattern by performing time series analysis on the historical outbound data, and obtain the demand fluctuation result using the sliding window detection method; A fulfillment ability quantification module, configured to extract the delivery time and success rate of items from the supplier fulfillment ability data, and determine the fulfillment quantification value of the supplier fulfillment ability by the weighted average method in combination with the demand fluctuation result; A social media analysis module, configured to analyze the potential hot product list according to the text and images in the social media data by using natural language processing technology and convolutional neural network; A cross-validation module, configured to perform cross-validation on the hot product list and the demand fluctuation result to determine the replenishment priority of different types of inventory; A replenishment optimization module, configured to optimize the replenishment quantity of various inventories by a linear programming algorithm according to the demand fluctuation result and the replenishment priority, and obtain a preliminary replenishment plan; A risk simulation module, configured to combine the preliminary replenishment plan and the fulfillment quantification value, and perform multiple iterative calculations on the supplier fulfillment uncertainty by using the Monte Carlo simulation algorithm to reduce potential risks and obtain a low-risk replenishment plan; A dynamic adjustment module, configured to perform dynamic fine-tuning on the low-risk replenishment plan by performing time series analysis on the low-risk replenishment plan to obtain the final multi-type inventory replenishment execution plan.

[0095] The above only lists some preferred embodiments of the present invention, but the present invention is not limited thereto, and many improvements and transformations can be made. As long as the improvements and transformations are made based on the basic principles of the present invention, they should be regarded as falling within the protection scope of the present invention.

[0096] The electronic device may be a computing device such as a desktop computer, a notebook, a palm computer, and a smart tablet. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above components are only examples of the electronic device, and do not constitute a limitation on the electronic device. It may include more or fewer components than the above, or combine some components, or different components. For example, the electronic device may further include input / output devices, network access devices, a bus, etc.

[0097] The so-called processor may be a Central Processing Unit (CPU) , or it may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the electronic device and connects all parts of the entire electronic device through various interfaces and circuits.

[0098] The memory can be used to store the computer modules. The processor realizes various functions of the electronic device by running or executing the computer modules stored in the memory and by calling the data stored in the memory. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store the operating system, application programs required for at least one function (such as the sound playback function, the image playback function, etc.); the data storage area can store the data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory may include high-speed random access memory and may also include non-volatile memory, such as hard disks, memory, plug-in hard disks, SmartMedia Cards (SMCs), Secure Digital (SD) cards, Flash Cards, at least one magnetic disk storage device, flash device, or other volatile solid-state storage devices.

[0099] Among them, if the modules integrated in the electronic device are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-described embodiment methods of the present invention, it can also be completed by a computer program instructing relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-described various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0100] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the attached drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without creative effort.

[0101] The above-described specific embodiments have further elaborated on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above is only the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. It is particularly pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A dynamic inventory replenishment method based on big data, characterized in that, The method includes: Obtaining inventory types, corresponding inventory quantities, historical outbound data, supplier fulfillment ability data, and social media data; Analyzing the historical outbound data through time series analysis to determine the periodic outbound pattern, and using the sliding window detection method to obtain the demand fluctuation result; Extracting the delivery time and success rate of items from the supplier fulfillment ability data, and combining with the demand fluctuation result, determining the fulfillment quantification value of the supplier fulfillment ability through the weighted average method; According to the text and images in the social media data, analyzing the potential hot product list using natural language processing technology and convolutional neural network; Cross-verifying the hot product list with the demand fluctuation result to determine the replenishment priority of different types of inventory; According to the demand fluctuation result and the replenishment priority, optimizing the replenishment quantity of various inventories through the linear programming algorithm to obtain a preliminary replenishment plan; Combining the preliminary replenishment plan and the fulfillment quantification value, and using the Monte Carlo simulation algorithm to perform multiple iterative calculations on the supplier fulfillment uncertainty to obtain a low-risk replenishment plan; Analyzing the low-risk replenishment plan through time series analysis, and dynamically fine-tuning the low-risk replenishment plan to obtain the final multi-type inventory replenishment execution plan.

2. The method according to claim 1, wherein The step of analyzing the historical outbound data through time series analysis to determine the periodic outbound pattern, and using the sliding window detection method to obtain the demand fluctuation result includes: Processing the historical outbound data through time series decomposition to extract the periodic outbound pattern; According to the periodic outbound pattern, using the sliding window technique to extract local data features; Calculating the demand fluctuation amplitude within each window through the local data features, where the fluctuation amplitude is the difference between the maximum value and the minimum value within the window; Judging whether the demand fluctuation amplitude exceeds the preset amplitude threshold. If so, smoothing the fluctuation amplitude through the moving average method. If not, no processing is performed to obtain a stable demand sequence; According to the stable demand sequence, adjusting the sliding window parameters until the fluctuation amplitude of the stable demand sequence is minimized to obtain the optimized demand fluctuation result.

3. The method according to claim 1, characterized in that The step of extracting the delivery time and success rate of items from the supplier fulfillment ability data, and combining with the demand fluctuation result, determining the fulfillment quantification value of the supplier fulfillment ability through the weighted average method includes: Generating an initial fulfillment ability data set through data combination according to the delivery time and the success rate; Determining the weights of the delivery time and the success rate in the fulfillment quantification value for the initial fulfillment ability data set and the demand fluctuation result; For each supplier, multiplying the delivery time and the success rate by the corresponding weights respectively and summing to obtain the fulfillment quantification value.

4. The method according to claim 1, wherein The step of according to the text and images in the social media data, analyzing the potential hot product list using natural language processing technology and convolutional neural network includes: Extracting the text data, image data, forwarding volume, and comment volume in the social media data, and using natural language processing technology to perform word segmentation and semantic parsing on the text data to obtain a preliminary semantic set; For the preliminary semantic set and the image data, extract visual features through a convolutional neural network to generate a product-related feature set; Use a weighted calculation method to process the product-related feature set to determine the initial value of the popularity of each product; According to the initial value of the product popularity, generate a popularity dynamic adjustment value based on the associated number of forwards and the number of comments; Fuse the popularity dynamic adjustment value with the initial value of the product popularity to obtain a quantified value of the product popularity; For the quantified value of the product popularity, use a sorting algorithm to rank the products by priority to generate a table of potential hot products; 5. The method according to claim 1, wherein Cross-validate the hot product table with the demand fluctuation result to determine the replenishment priority of different types of inventory, including: Use a cross-comparison method to extract the matching features of the hot product table and the demand fluctuation result to obtain a product demand association set; Use a weighted calculation to process the product demand association set and the types of inventory to obtain a preliminary quantified value of the replenishment demand; According to the preliminary quantified value, arrange the inventory types through a sorting algorithm to generate a replenishment order table; According to the product demand association set, judge whether there are multiple product association degrees in the replenishment order table that are similar. If so, adjust the order through the historical change trend of the demand fluctuation result to obtain an optimized order table; According to the optimized order table, calculate the difference between the expected inventory of the inventory types with higher priority and the real-time inventory to generate the final replenishment priority; 6. The method according to claim 1, wherein Combine the preliminary replenishment plan table and the fulfillment quantification value, and use the Monte Carlo simulation algorithm to perform multiple iterative calculations on the supplier fulfillment uncertainty to reduce potential risks and obtain a low-risk replenishment plan table, including: Use the Monte Carlo simulation algorithm to perform multiple iterative calculations on the standardized fulfillment quantification value to obtain a fulfillment uncertainty distribution; According to the preliminary replenishment plan table and the fulfillment uncertainty distribution, perform a secondary iterative calculation through the Monte Carlo simulation algorithm to obtain a risk assessment result; For the risk assessment result, use the mean-variance analysis method to screen the preliminary replenishment plan table to obtain a low-risk replenishment plan table; 7. The method according to claim 1, wherein Dynamically fine-tune the low-risk replenishment plan table through time series analysis to obtain a final multi-type inventory replenishment execution plan, including: Obtain the periodic change characteristics from the low-risk replenishment plan table through time series analysis to obtain a rhythm fluctuation distribution; Use a dynamic fine-tuning method to adjust the rhythm fluctuation distribution to obtain a corrected rhythm parameter; For the corrected rhythm parameter, divide the low-risk replenishment plan table through multi-category classification rules to obtain a set of category demands; Generate a multi-type inventory replenishment execution plan according to the set of category demands; 8. The method according to claim 1, characterized in that, After obtaining the final multi-type inventory replenishment execution plan, the method further includes: Input the inventory replenishment execution plan into the inventory management system, obtain replenishment process data through the inventory management system, and use a time window division method to extract real-time tracking data to obtain a dynamic replenishment status; Analyze the fulfillment change trend according to the dynamic replenishment status, divide the supplier fulfillment status using a clustering algorithm, and determine the fulfillment adjustment coefficient; Update the adjustment parameter set according to the fulfillment adjustment coefficient, update the product status through the hot product table, and obtain the classified demand change; Adjust the inventory demand classification according to the classified demand change, and smooth the replenishment execution plan using the moving average method to obtain a preliminary correction plan; Fuse the inventory demand classification for the preliminary correction plan, and verify the replenishment process data through the inventory management system to obtain a real-time replenishment plan.

9. A dynamic inventory replenishment system based on big data, characterized in that, The system includes: A data acquisition module for acquiring inventory types and corresponding inventory quantities, historical outbound data, supplier fulfillment ability data, and social media data; A time series analysis module for analyzing the historical outbound data through time series to determine the periodic outbound pattern, and obtaining the demand fluctuation result using the sliding window detection method; A fulfillment ability quantification module for extracting the delivery time and success rate of items from the supplier fulfillment ability data, and determining the fulfillment quantification value of the supplier fulfillment ability through the weighted average method in combination with the demand fluctuation result; A social media analysis module for analyzing the potential hot product table according to the text and images in the social media data using natural language processing technology and convolutional neural networks; A cross-validation module for cross-validating the hot product table and the demand fluctuation result to determine the replenishment priority of different types of inventory; A replenishment optimization module for optimizing the replenishment quantity of various types of inventory through a linear programming algorithm according to the demand fluctuation result and the replenishment priority to obtain a preliminary replenishment plan table; A risk simulation module for combining the preliminary replenishment plan table and the fulfillment quantification value, and performing multiple iterative calculations on the supplier fulfillment uncertainty using the Monte Carlo simulation algorithm to reduce potential risks and obtain a low-risk replenishment plan table; A dynamic adjustment module for dynamically fine-tuning the low-risk replenishment plan table through time series analysis of the low-risk replenishment plan table to obtain a final multi-type inventory replenishment execution plan.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the big data-based dynamic inventory replenishment method according to any one of claims 1 to 8.

Images