Method for collecting automatic replenishment data of fluent goods shelf

Abstract

The invention relates to the technical field of data processing, in particular to a method for collecting automatic replenishment data of a fluent shelf, and the method comprises the steps: obtaining the historical automatic replenishment data of the fluent shelf at a current analysis moment, and constructing a preliminary decision tree for classification; evaluating and calculating the compactness degree of each category and the fusion coefficient of each category and the adjacent category, determining the re-division coefficient of each category and the adjacent category according to the compactness degree of the category and the fusion coefficient, and updating the decision tree in real time according to the size of the re-division coefficient. And classifying the automatic replenishment data generated between the current analysis moment and the next analysis moment by using the decision tree updated in real time so as to realize classified collection of the data. According to the method, the initially constructed decision tree is dynamically updated, so that the feature rule of the automatic replenishment data is accurately reflected; the accuracy of classified collection of the automatic replenishment data of the fluent goods shelf is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method for collecting data of automatic replenishment of a flow-type shelf. Background Art

[0002] At a time when the retail, warehousing and logistics industries are booming, flow racks, as important commodity storage and display equipment, are widely used in supermarkets, warehouses and e-commerce distribution centers. Its high efficiency is reflected in the orderly arrangement of commodities, which is convenient for picking up and replenishing. The automatic replenishment system closely connected with it is the core link to give full play to the advantages of flow racks and improve operational efficiency. In order to achieve scientific analysis and management of the automatic replenishment data of flow racks, accurate data collection methods are essential.

[0003] At present, the main method used to process replenishment data collected in a time series on a flow rack is data classification, so that staff can analyze data more intuitively. Among them, the classification method based on decision trees is the most commonly used. This method usually builds a decision tree classification model based on the automatic replenishment data collected during an initial period of time. After the decision tree is built, the internal structure and division rules are relatively fixed. The automatic replenishment data is classified according to the constructed decision tree classification model to achieve the classification collection of automatic replenishment data.

[0004] However, automatic replenishment data has the characteristics of dynamic changes, and its characteristic distribution will change over time and with business development. For example, when initially building a decision tree, a certain replenishment threshold is used as an important basis for division based on the sales season and product promotion activities at the time. However, as the seasons change and promotion activities end, the sales speed and replenishment rules of the products change, and the original division basis is no longer applicable to the new replenishment data, and the decision tree cannot automatically adjust to adapt to these changes.

[0005] Therefore, the traditional decision tree built based on automatic replenishment data collected over an initial period of time is difficult to accurately classify when faced with continuously updated automatic replenishment data, resulting in errors in the classification of automatic replenishment data, affecting the accurate judgment of inventory status, and may also cause problems such as untimely replenishment or excessive replenishment, thereby increasing operating costs. Summary of the invention

[0006] In order to solve the problem that it is difficult to accurately classify the continuously updated automatic replenishment data when classifying the automatic replenishment data of the flow-type shelves based on the traditional decision tree, which affects the accurate judgment of the inventory status, the present invention proposes a method for collecting the automatic replenishment data of the flow-type shelves, including: Obtain the historical automatic replenishment data of the flow rack at the current analysis moment, and construct a preliminary decision tree at the current analysis moment to divide the historical automatic replenishment data into multiple categories, each category contains multiple data points, and each data point includes multiple data; Calculate the reclassification coefficient of each category and its adjacent categories to reflect the possibility that each category can be integrated with its adjacent categories: ; In the formula, For the categories and The reclassification coefficient of each category, and Separate indivual, The compactness of the categories, For the categories and The fusion coefficient of each category; the compactness is determined based on the difference between the data of all data points of each category and the data of the central data point of the category, and is used to reflect the degree of aggregation of all data points in each category; the fusion coefficient is determined by the degree of penetration between each category and its adjacent categories, and is used to reflect the possibility that each category can be fused with its adjacent categories; In response to the size of the re-division coefficient of each category and its adjacent categories, the preliminary decision tree at the current analysis moment is updated in real time. Based on the decision tree updated in real time at the current analysis moment, the automatic replenishment data generated between the current analysis moment and the next analysis moment is classified to realize data collection.

[0007] The above technical solution can quickly sort out the characteristics and laws of historical automatic replenishment data through preliminary decision tree classification, preliminarily organize and classify the historical automatic replenishment data, and provide an orderly data structure for subsequent calculation and real-time update of decision trees. And further, according to the degree of aggregation of internal data of each category and the closeness of the relationship between categories, the possibility of fusion and update classification between categories is comprehensively determined, that is, the re-classification coefficient, and the re-classification coefficient provides a quantitative judgment basis for the real-time update of the decision tree. And further, according to the size of the re-classification coefficient, the decision tree is dynamically updated in real time, which enhances the adaptability of the decision tree to dynamically changing data, and is no longer limited to the fixed structure initially constructed, so that the decision tree can adapt to the dynamic changes of automatic replenishment data, and timely adjust the classification rules and structure of the decision tree. The real-time updated decision tree can more accurately reflect the actual characteristics and laws of the current replenishment data, and based on this, the newly collected automatic replenishment data is classified and collected, which improves the pertinence and accuracy of data collection, and provides more reliable data support for subsequent data analysis and replenishment decisions.

[0008] Furthermore, the method for updating the preliminary decision tree at the current analysis moment in real time is: A threshold of the reclassification coefficient is preset; if the reclassification coefficient of a category and its adjacent category is greater than the threshold, the category and its adjacent category are merged; if the reclassification coefficient of a category and its adjacent category is not greater than the threshold, the category and its adjacent category are not merged; Recalculate the Gini impurity of each category, and re-divide the nodes of the preliminary decision tree at the current analysis time according to the Gini impurity of each category to obtain the updated decision tree at the current analysis time; The specific method of re-dividing nodes is consistent with the specific method of dividing nodes when constructing the preliminary decision tree at the current analysis moment.

[0009] The above technical solution can make the decision tree adapt to the changes of automatic replenishment data in a timely manner by updating the decision tree in real time. When the data characteristics change, the structure and classification rules of the decision tree can be adjusted through category fusion and node re-division, so that the newly collected data can be continuously and accurately classified, which enhances the adaptability and stability of the decision tree in a dynamic environment.

[0010] Furthermore, the compactness is determined based on the following formula:

[0011] In the formula, For the The compactness of the categories, For the The importance weight of the item data is determined based on the hierarchical analysis method. For the The number of data points contained in each category, For the Category The data point Item data, For the The center data point of the category Item data, For the All data points in the category The extreme difference of the data, is the total number of data items.

[0012] The above technical solution comprehensively considers the compactness of data points in each category from multiple dimensions. It not only considers the difference between each data point and the central data point of the category, but also combines the importance weight of each data, as well as the number of data points in the category and the range of the data. Through such comprehensive calculation, it can more comprehensively and accurately reflect the degree of clustering of data points in the category, avoiding the limitations of single factor measurement.

[0013] Furthermore, the fusion coefficient is determined based on the following formula:

[0014] In the formula, For the categories and The fusion coefficient of each category, For the The total number of data points in each category, To obtain the maximum value function, For the Category Data points and The penetration level of each category, For the Category Data points and The penetration level of each category, For the The total number of data points in each category.

[0015] The above technical solution comprehensively considers the degree of penetration of each data point in two adjacent categories into the other category, and obtains the fusion coefficient by processing the penetration degree and combining it with the total number of category data points. It provides a quantitative basis for judging whether adjacent categories can be fused, and helps to identify which categories have blurred boundaries and data penetration, and need to be merged or adjusted.

[0016] Furthermore, the historical automatic replenishment data at the current analysis time is the automatic replenishment data generated by the flow rack at all times before the current analysis time as the historical automatic replenishment data at the current analysis time.

[0017] Furthermore, each category and its adjacent categories are determined based on the following method: Perform a depth-first traversal of the preliminary decision tree at the current analysis moment, and sort all categories in the order in which they were traversed; Each sorted category and its next adjacent category are taken as the adjacent categories of each category.

[0018] Further, Category Data points and The penetration level of each category is determined based on the following formula:

[0019] In the formula, For the Category Data points and The penetration level of each category, For the The importance weight of the item data, and Respectively , Category The data point Item data, and Respectively , The center data point of the category Item data, is the total number of data items.

[0020] The above technical solution can comprehensively and fully reflect the degree to which a data point deviates from its own category and approaches the adjacent category by considering the differences in each data point between the data point of each category and the central data point of its own category and the central data point of the adjacent category, combined with the importance weight of each data, and provides a basis for judging whether the boundaries between categories are blurred.

[0021] Further, Category Data points and The penetration level of each category is determined based on the following formula:

[0022] In the formula, For the Category Data points and The penetration level of each category, For the The importance weight of the item data, and Respectively , Category The data point Item data, and Respectively , The center data point of the category Item data, is the total number of data items.

[0023] Furthermore, the automatic replenishment data includes: replenishment interval, inventory level at replenishment, replenishment quantity, and commodity sales speed.

[0024] Furthermore, the method for obtaining the central data point of the category is: Calculate the mean of all data points in each category on each data to form a multidimensional data point, and use this data point as the central data point of each category.

[0025] The present invention has the following effects: The present invention analyzes the automatic replenishment data of the flow-type shelves to dynamically update the initially constructed decision tree, so that the decision tree can adapt to the distribution law of the new automatic replenishment data and timely adjust its own classification rules and structure, thereby accurately reflecting the characteristics and laws of the current replenishment data. Data collection is performed based on the real-time updated decision tree, which improves the pertinence and accuracy of the automatic replenishment data collection of the flow-type shelves and provides reliable data support for subsequent replenishment decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic flow chart of the method of the present invention; Figure 2 It is a schematic diagram of the method flow of step S3 of the present invention. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0028] Reference Figure 1 The present invention provides a method for collecting data of automatic replenishment of a flow-type shelf, comprising steps S1 to S4: S1: Collect historical automatic replenishment data of the flow rack at the current analysis time.

[0029] The automatic replenishment data of the flowing shelves include: replenishment interval, inventory at replenishment, replenishment quantity, and commodity sales speed, a total of these four data items. The administrator of the flowing shelves collects automatic replenishment data every time the flowing shelves are replenished.

[0030] In one embodiment, the current analysis time refers to a time arbitrarily selected after the flow rack has been used for a period of time (greater than or equal to 3 months), which is used as the current analysis time. The automatic replenishment data generated in all time periods before the current analysis time is used as the historical automatic replenishment data at the current analysis time.

[0031] The administrator of the flow rack can analyze the historical automatic replenishment data at the current analysis moment to better understand the sales situation of the goods and the real-time inventory of the goods in order to make subsequent automatic replenishment decisions.

[0032] S2: Construct a preliminary decision tree for the current analysis moment to classify the historical automatic replenishment data.

[0033] At the current analysis time, the historical automatic replenishment data of the flow rack has been collected, and each historical automatic replenishment data is taken as a data point to obtain multiple data points. Each data point consists of the values ​​of four data items: replenishment interval, inventory at replenishment, replenishment quantity, and commodity sales speed.

[0034] In one embodiment, the method for constructing the preliminary decision tree at the current analysis moment is: First, divide the root node: Select one data item from all data items in turn. For the selected data item, traverse each value in its value range and use it as a partition point. For each partition point, use the partition point to divide all data points into two sub-datasets: one sub-dataset is smaller than the partition point, and the other sub-dataset is greater than or equal to the partition point.

[0035] Calculate the Gini impurity of each of the two sub-datasets, and calculate the sum of the Gini impurities of the two sub-datasets. After traversing all the values ​​of the selected data item, find the partition point that minimizes the sum of the Gini impurities of the two sub-datasets after partitioning, and use it as the suspected root node of the selected data item; perform the above operation on all data items to obtain the suspected root nodes of all data items.

[0036] A suspected root node is selected from the suspected root nodes of all data items. If the suspected root node minimizes the sum of the Gini impurities of the two sub-data sets after division, the suspected root node is used as the root node of the decision tree, and the data points less than the root node are divided into the left subtree of the root node, and the data points greater than or equal to the root node are divided into the right subtree of the root node.

[0037] Then, for all the data points of the left subtree and the right subtree of the root node, divide them in the same way as in the above steps, and divide all the data points of the left subtree of the root node into a left subtree and a right subtree; divide all the data points of the right subtree of the root node into a left subtree and a right subtree.

[0038] Finally, for each left subtree and each right subtree generated, the above processing steps for the left subtree and the right subtree are repeated continuously, that is, the operation is performed recursively, and in this recursive process, whether the stopping condition is met is continuously checked: For the current subtree (left subtree or right subtree), after completing the attempts of all the above data items and partitioning points, it is found that no matter which data item and which partitioning point are selected, the sum of the Gini impurities of the two new sub-data sets after the partition cannot be less than the Gini impurity of the current subtree, which indicates that the data confusion of the current subtree cannot be further reduced through the partitioning operation. At this time, stop the partitioning of the current subtree and use the current subtree as a leaf node of the decision tree.

[0039] When all subtrees meet the stopping conditions and no further division is performed, the construction of the preliminary decision tree at the current analysis moment is completed.

[0040] S3: According to the data distribution characteristics in the classification results, the preliminary decision tree at the current moment is updated in real time to obtain the decision tree updated in real time at the current analysis moment.

[0041] This step obtains the data distribution characteristics of the preliminary decision tree classification results, including: first analyzing the compactness of each category, then analyzing the fusion coefficient between each category and its adjacent categories, and finally analyzing the compactness of each category and the fusion coefficient between each category and its adjacent categories.

[0042] After obtaining these data distribution characteristics, the re-division coefficients of each category and its adjacent categories are comprehensively determined, and the preliminary decision tree is updated in real time based on the re-division coefficients.

[0043] Specifically, if Figure 2 As shown in , the following steps are included: S31: Determine the compactness of each category according to the difference between the data items of all data points in each category and the data items of the central data point in the category.

[0044] For each category, the method for obtaining the central data point is: calculate the mean of all data points of each category on each data, form a multi-dimensional data point, and use this data point as the central data point of each category.

[0045] In one embodiment, the compactness of each category is determined based on the following formula:

[0046] In the formula, For the The compactness of a category is used to reflect the degree of aggregation of the data points contained in the category. For the Importance weight of each data item: By introducing the importance weight of each data item, different importance can be assigned to different data items when calculating the compactness of the category. For example, given that the sales speed of goods has a significant impact on the replenishment decision, it can be given a higher weight so that the calculation result can more accurately reflect the actual replenishment needs. For the The number of data points contained in each category, For the Category The data point Item data, For the The center data point of the category Item data, For the All data points in the category The data is extremely poor. is the total number of data items. The value of is 4.

[0047] In this formula, The molecule of part of all data points in the category Item data and The center data point of the category The difference between the data items, the denominator is All data points in the category The range of the data ( The maximum value of each item of data minus the minimum value), the overall value of this part represents the of all data points in the category The dispersion degree of the data is Partially indicates of all data points in the category The degree of clustering of the data items. The greater the degree of clustering, the Among the categories, The more compact the distribution of item data is.

[0048] This formula is obtained by The aggregation degree of all data items in a category is calculated by summing up, which fully integrates the distribution information of each data, avoids the limitation of focusing on a single data item, and can more accurately characterize the The aggregate features of all item data in a category.

[0049] In one embodiment, the importance weight of each data is determined based on the hierarchical analysis method, which is a prior art that comprehensively considers the interrelationships and relative importance among all data, can determine the importance weight of each data, and can more comprehensively and accurately reflect the importance of each data in the automatic replenishment data for subsequent calculations.

[0050] S32: Determine the fusion coefficient between each category and its adjacent categories according to the mutual penetration degree between each category and its adjacent categories.

[0051] Each category and its adjacent categories are determined based on the following method: the preliminary decision tree at the current analysis moment is traversed in depth-first order, and all categories are sorted in the order in which they are traversed; Combine each sorted category with its next adjacent category as the adjacent category of each category. Category as the The adjacent categories of a category.

[0052] When analyzing data categories, the compactness of each category is important, but the relative distribution between each category and its adjacent categories cannot be ignored.

[0053] If a category has a high similarity in data distribution characteristics with its adjacent categories, and shows consistency in the distribution patterns of various data items (such as replenishment interval, inventory at replenishment, replenishment quantity, commodity sales speed, etc.), making them more likely to be regarded as the same category in a logical and practical sense, then it can be determined that the two categories have a strong tendency to merge. In this case, a higher fusion coefficient is assigned, which will serve as an important quantitative basis for subsequent judgments on whether to merge and how to merge. In this way, data categories can be integrated and divided more accurately, making them more in line with the inherent characteristics of the data and actual application needs.

[0054] In one embodiment, the mutual penetration degree between each category and its adjacent categories is specifically reflected by the mutual penetration degree between the data points of each category and its adjacent categories, and the mutual penetration degree between the data points of the adjacent category and the category.

[0055] For category, and its adjacent category is categories; No. Category Data points and The calculation formula for the penetration degree of each category is:

[0056] In the formula, For the Category Data points and The penetration level of each category, For the The importance weight of the item data, and Respectively , Category The data point Item data, and Respectively , The center data point of the category Item data, is the total number of data items.

[0057] In this formula, Measured the Category The data point in The data is the same as the The difference between the center data points of the categories, Measured the Category The difference between the data point and the center data point of its own category is divided by the two. Category The difference between the data point and the central data point of the adjacent category and the central data point of its own category is compared to highlight the Category The greater the value obtained by dividing the data point, the greater the deviation of the data point from its own category and the closer it is to the adjacent category. Category Data points and The greater the penetration of a category.

[0058] This formula takes into account the weighted sum of all items and the square root. Category The difference between the data point and the adjacent category center in each data item is taken into account at the same time, which more comprehensively reflects the Category Data points and The overall penetration level of each category.

[0059] No. Category Data points and The calculation formula for the penetration degree of each category is:

[0060] In the formula, For the Category Data points and The penetration level of each category, For the The importance weight of the item data, and Respectively , Category The data point Item data, and Respectively , The center data point of the category Item data, is the total number of data items. The design logic of this formula is similar to the previous one.

[0061] Combining the above two formulas, a two-way evaluation is achieved: the two formulas are respectively Category to categories and from Category to The two-way evaluation can more accurately determine the closeness of the relationship between the two categories.

[0062] If the mutual penetration degree of data points between two categories is high, it means that the two categories are relatively similar in data characteristics, and merging them into one category may be more in line with the actual distribution of the data. On the contrary, if the mutual penetration degree is very low, it means that the boundary between the two categories is relatively clear and no fusion is needed.

[0063] In one embodiment, the fusion coefficient between each category and its adjacent categories is determined based on the following formula:

[0064] In the formula, For the categories and The fusion coefficient of each category is used to reflect the possibility of fusion operation between categories. For the The total number of data points in each category, To obtain the maximum value function, For the Category Data points and The penetration level of each category, For the Category Data points and The penetration level of each category, For the The total number of data points in each category.

[0065] In this formula, through and Considering the Category and The degree of two-way penetration of data points between the categories fully reflects the relationship between the data points of the two categories. Category and The average value of the two-way penetration degree of data points between categories is calculated, so that the calculation of the fusion coefficient can take into account the overall situation of the data points of each category, avoid excessive influence of individual data points, and more accurately reflect the overall fusion trend between categories.

[0066] In this formula, The function is to ensure that the fusion coefficient will not be excessively weakened due to too low a penetration level. When the penetration level is less than 1, 1 is taken as the lower limit, which to a certain extent emphasizes the potential fusion possibility between categories. Even if the current penetration level is relatively low, a certain basic weight is given to avoid the unreasonable impact of extremely low penetration values ​​on the fusion coefficient caused by accidental factors.

[0067] In summary, this formula can quantify the possibility of fusion between two adjacent categories and obtain a specific fusion coefficient. This coefficient can be used as an important basis for judging whether two categories should be fused and the degree of fusion. It provides a clear numerical reference for subsequent data processing and analysis, and facilitates comparison and decision-making between different categories.

[0068] S33: Comprehensively determine the re-division coefficient of each category and its adjacent categories.

[0069] In one embodiment, a method for comprehensively determining the reclassification coefficients of each category and its adjacent categories is as follows:

[0070] In the formula, For the categories and The reclassification coefficient of each category is used to reflect the possibility of merging each category with its adjacent categories. and Separate indivual, The compactness of the categories, For the categories and The fusion coefficient of each category.

[0071] This formula cleverly combines the compactness of each category (the data points within it) and the possibility of fusion between each category and its adjacent categories, providing a quantitative indicator for whether two adjacent categories (each category and its adjacent categories) can be fused.

[0072] If the distribution of data points within two adjacent categories is very compact ( and The larger the value, the fusion coefficient between them The smaller it is, the better the data aggregation is within the two adjacent categories, and the lower the data penetration between the two adjacent categories is, that is, The larger the value of , the less the two adjacent categories need to be fused and updated. The smaller the value, the greater the reclassification coefficient of these two adjacent categories. The smaller it is, the smaller it is, and vice versa.

[0073] S34: Update the preliminary decision tree at the current moment in real time.

[0074] The preset threshold of the reclassification coefficient is 0.3; if the reclassification coefficient of a category and its adjacent category is greater than 0.3, the category and its adjacent category are merged, that is, all data points contained in the two categories are merged to form a new category; if the reclassification coefficient of a category and its adjacent category is not greater than 0.3, the category and its adjacent category are not merged, that is, the category and its adjacent category are kept independent and no fusion operation is performed.

[0075] After completing the above fusion judgment and operation, all categories are obtained (including categories that have not been fused and remain independent). For each category, the Gini impurity of the category is calculated based on all data points of the category (according to the method of calculating the Gini impurity when constructing the preliminary decision tree at the current analysis moment in step S2).

[0076] Based on the recalculated Gini impurity of each category, the nodes of the preliminary decision tree at the current analysis moment are re-divided. The specific method of dividing the nodes here is consistent with the specific method of dividing the nodes when constructing the preliminary decision tree, that is: Starting from the root node, by continuously trying different data items and their values ​​as partitioning points, find the partitioning point that minimizes the sum of the Gini impurities of the sub-datasets after partitioning, and recursively process each subtree until the stopping condition is met (that is, for a subtree, no matter which data item and partitioning point are selected, the sum of the Gini impurities of the two new sub-datasets after partitioning cannot be less than the Gini impurity of the current subtree). At this time, the decision tree updated at the current analysis moment is obtained.

[0077] S4: Use the decision tree updated in real time at the current analysis moment to accurately classify the automatic replenishment data between the current analysis moment and the next analysis moment, so as to realize the collection of automatic replenishment data of the flow shelf.

[0078] In one embodiment, the interval between analysis moments is set to 1 month (empirical value). For example, if the current analysis moment is the last day of January, the next analysis moment of the current analysis moment is the last day of February. All time before the last day of January is taken as the historical time period of the current analysis moment, and all automatic replenishment data of the flow rack in this historical time period is taken as the historical automatic replenishment data of the current moment. Then, the decision tree updated in real time at the current analysis moment is obtained according to the method in steps S1 to S3 of this scheme. Finally, for the automatic replenishment data generated between the last day of January and the last day of February, the automatic replenishment data is accurately classified according to the decision tree updated in real time according to the current analysis moment, so as to realize the accurate collection of the automatic replenishment data generated by the flow rack during this time interval.

[0079] When the time reaches the last day of February, the last day of February is used as the current analysis moment, the last day of March is used as the next analysis moment, all the time before the last day of February is used as the historical time period of the current analysis moment, and all the automatic replenishment data of the flow rack in this historical time period is used as the historical automatic replenishment data of the current moment, and then the decision tree updated in real time at the current analysis moment is obtained according to the method in steps S1 to S3 of this scheme. Finally, for the automatic replenishment data generated between the last day of February and the last day of March, the automatic replenishment data is accurately classified according to the decision tree updated in real time according to the current analysis moment, so as to realize the accurate collection of the automatic replenishment data generated by the flow rack during this time interval. By analogy, the accurate collection of the automatic replenishment data of the flow rack is realized.

[0080] In one embodiment, the method for accurately classifying the automatic replenishment data between the current analysis time and the next analysis time using the decision tree updated in real time at the current analysis time is: For each piece of automatic replenishment data between the current analysis moment and the next analysis moment, the values ​​of multiple data items including replenishment interval, inventory at replenishment, replenishment quantity, and commodity sales speed are extracted and combined to form a data point.

[0081] Match the data point with the decision tree updated in real time at the current analysis moment, and traverse from the root node of the decision tree. Each node of the decision tree corresponds to a judgment condition of a data item (for example, judging whether the replenishment interval is greater than a certain value, etc.). According to the value of the corresponding data item in the data point, move down step by step according to the branch conditions of the decision tree.

[0082] This process continues until a leaf node of the decision tree is reached. Since each leaf node of the decision tree corresponds to a category, the category represented by the leaf node is the classification result of the data point corresponding to this automatic replenishment data, that is, this automatic replenishment data is classified into the category represented by the leaf node, which is consistent with the previous method of dividing the historical automatic replenishment data at the current analysis moment into different leaf node categories according to the preliminary decision tree. The final classification is determined based on the matching of data points with decision tree node features and the branching conditions of the decision tree.

[0083] In summary, the method of this step has the following advantages: The data timeliness and dynamic adaptability are guaranteed. Specifically, through fixed analysis time intervals, sufficient data can be guaranteed for effective analysis, and the decision tree can keep up with the dynamic changes of data in a timely manner. Over time, commodity sales patterns and replenishment needs will change. Updating the decision tree monthly can capture these changes and make classification collection more realistic. For example, seasonal commodity sales vary greatly, and updating the decision tree monthly can better reflect the impact of seasonal changes on the classification of replenishment data. Using the same decision tree to classify and collect automatic replenishment data between adjacent analysis times ensures the consistency and coherence of data classification within a stable time period, which facilitates subsequent data analysis and management.

[0084] This method is an effective use of historical data. Specifically, each time a decision tree is constructed, it is based on all historical data before the current analysis moment to fully explore the information contained therein. Historical data contains patterns and trends in commodity sales and replenishment. By continuously accumulating and analyzing these data, the decision tree can learn more accurate classification rules. For example, by analyzing multiple months of historical data, the sales peak patterns of certain commodities in specific months can be discovered, making subsequent classification collection and judgment more reasonable.

[0085] This method is repeatable: managers regularly collect historical data, update decision trees and collect new data according to established processes every month to achieve standardization and regularization of replenishment data collection. This repeatability is conducive to comparing and evaluating data collection results in different periods and discovering potential problems and improvement directions.

[0086] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A method for collecting data on automatic replenishment of flow shelves, characterized in that: include: Obtain the historical automatic replenishment data of the flow rack at the current analysis moment, and construct a preliminary decision tree at the current analysis moment to divide the historical automatic replenishment data into multiple categories, each category contains multiple data points, and each data point includes multiple data; The reclassification coefficient of each category and its adjacent categories is calculated to reflect the possibility that each category can be integrated with its adjacent categories: ; In the formula, For the categories and The reclassification coefficient of each category, and Separate indivual, The compactness of the categories, For the categories and The fusion coefficient of each category; the compactness is determined according to the difference between the data of all data points of each category and the data of the central data point of the category, and is used to reflect the degree of aggregation of all data points in each category; The fusion coefficient is determined by the degree of penetration between each category and its adjacent categories, and is used to reflect the possibility that each category can be fused with its adjacent categories; In response to the size of the re-division coefficient of each category and its adjacent categories, the preliminary decision tree at the current analysis moment is updated in real time. Based on the decision tree updated in real time at the current analysis moment, the automatic replenishment data generated between the current analysis moment and the next analysis moment is classified to realize data collection.

2. The method for collecting data of automatic replenishment of flow shelves according to claim 1 is characterized in that: The method for real-time updating of the preliminary decision tree at the current analysis moment is: A threshold of the reclassification coefficient is preset; if the reclassification coefficient of a category and its adjacent category is greater than the threshold, the category and its adjacent category are merged; if the reclassification coefficient of a category and its adjacent category is not greater than the threshold, the category and its adjacent category are not merged; Recalculate the Gini impurity of each category, and re-divide the nodes of the preliminary decision tree at the current analysis time according to the Gini impurity of each category to obtain the updated decision tree at the current analysis time; The specific method of re-dividing nodes is consistent with the specific method of dividing nodes when constructing the preliminary decision tree at the current analysis moment.

3. The method for collecting data of automatic replenishment of flow shelves according to claim 1 is characterized in that: The degree of compactness is determined based on the following formula: ； In the formula, For the The compactness of the categories, For the The importance weight of the item data is determined based on the hierarchical analysis method. For the The number of data points contained in each category, For the Category The data point Item data, For the The center data point of the category Item data, For the All data points in the category The extreme difference of the data, is the total number of data items.

4. The method for collecting data of automatic replenishment of flow shelves according to claim 1 is characterized in that: The fusion coefficient is determined based on the following formula: ； In the formula, For the categories and The fusion coefficient of each category, For the The total number of data points in each category, To obtain the maximum value function, For the Category Data points and The penetration level of each category, For the Category Data points and The penetration level of each category, For the The total number of data points in each category.

5. The method for collecting data of automatic replenishment of flow shelves according to claim 1 is characterized in that: The historical automatic replenishment data at the current analysis time is the automatic replenishment data generated by the flow rack at all times before the current analysis time as the historical automatic replenishment data at the current analysis time.

6. The method for collecting data of automatic replenishment of flow shelves according to claim 2 is characterized in that: Each category and its adjacent categories are determined based on the following method: Perform a depth-first traversal of the preliminary decision tree at the current analysis moment, and sort all categories in the order in which they were traversed; Each sorted category and its next adjacent category are taken as the adjacent categories of each category.

7. The method for collecting data of automatic replenishment of flow shelves according to claim 4 is characterized in that: No. Category Data points and The penetration level of each category is determined based on the following formula: ； In the formula, For the Category Data points and The penetration level of each category, For the The importance weight of the item data, and Respectively , Category The data point Item data, and Respectively , The center data point of the category Item data, is the total number of data items.

8. The method for collecting data of automatic replenishment of flow shelves according to claim 4 is characterized in that: No. Category Data points and The penetration level of each category is determined based on the following formula: ； In the formula, For the Category Data points and The penetration level of each category, For the The importance weight of the item data, and Respectively , Category The data point Item data, and Respectively , The center data point of the category Item data, is the total number of data items.

9. The method for collecting data of automatic replenishment of flow shelves according to claim 5 is characterized in that: The automatic replenishment data includes: replenishment interval, inventory level at replenishment, replenishment quantity, and commodity sales speed.

10. The method for collecting data of automatic replenishment of flow shelves according to claim 7 or 8, characterized in that: The method for obtaining the central data point is: calculate the mean of all data points of each category on each data, form a multi-dimensional data point, and use this data point as the central data point of each category.

Images