Replenishment quantity prediction method and device, model training method and device and storage medium

By constructing a replenishment quantity prediction model that combines sales characteristics and planned sample data, the problem of insufficient accuracy in replenishment quantity prediction in the supply chain is solved, and the matching of replenishment quantity with sales demand and overall planning is achieved, thereby improving the automation level and cost-effectiveness of the supply chain.

CN119048150AInactive Publication Date: 2024-11-29BEIJING WODONG TIANJUN INFORMATION TECH CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410101956.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-01-24
Publication Date
2024-11-29
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

In existing technologies, the accuracy of replenishment volume forecasting in the supply chain is insufficient, making it difficult to simultaneously meet market demand and overall planning, resulting in inventory backlog and resource waste.

Method used

By constructing a replenishment quantity prediction model, combining sales characteristics and planned sample data, and using an encoding/decoding network and loss function for training, the sales prediction value is adjusted to improve accuracy, ensuring that the replenishment quantity meets sales demand and overall planning.

Benefits of technology

It improved the accuracy of replenishment forecasting, reduced inventory backlog and resource waste, and enhanced the automation level and cost-effectiveness of the supply chain.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119048150A_ABST
    Figure CN119048150A_ABST
Patent Text Reader

Abstract

The invention provides a replenishment quantity prediction method and device, a model training method and device and a storage medium, and relates to the technical field of intelligent supply chains. The method for training the replenishment volume prediction model comprises the steps that sales volume prediction sample data, sales volume feature sample data of a first dimension, sales volume feature sample data of a second dimension and plan sample data of the second dimension are acquired, and the second dimension comprises a plurality of sales volume statistical objects of the first dimension; according to the sales volume characteristic sample data of the first dimension, the sales volume characteristic sample data of the second dimension and the plan sample data of the second dimension, the sales volume prediction sample data is adjusted based on a to-be-trained replenishment volume prediction model, a loss function value is obtained until training is completed, and the replenishment volume prediction model is determined, parameters of the to-be-trained replenishment quantity prediction model are adjusted according to the loss function value.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of smart supply chain technology, and in particular to a replenishment quantity prediction method, device, model training method, device, and storage medium. Background Art

[0002] In the supply chain industry, maintaining a certain spot rate, reducing the probability of out-of-stock, and avoiding inventory backlogs can meet consumer demand while avoiding excessive storage and capital pressure. Sales forecasting technology can provide reference data for stocking. Summary of the invention

[0003] One object of the present disclosure is to improve the accuracy of replenishment.

[0004] According to one aspect of some embodiments of the present disclosure, a training method for a replenishment quantity prediction model is proposed, comprising: obtaining sales forecast sample data, sales feature sample data of a first dimension, sales feature sample data of a second dimension, and planned sample data of the second dimension, wherein the second dimension comprises a plurality of sales statistical objects of the first dimension; adjusting the sales forecast sample data based on a replenishment quantity prediction model to be trained according to the sales feature sample data of the first dimension, the sales feature sample data of the second dimension, and the planned sample data of the second dimension, and obtaining a loss function value until the training is completed, and determining the replenishment quantity prediction model, wherein the parameters of the replenishment quantity prediction model to be trained are adjusted according to the loss function value.

[0005] In some embodiments, adjusting the sales forecast sample data based on the replenishment quantity prediction model to be trained and obtaining the loss function value includes: adjusting the sales forecast sample data according to the sales feature sample data of the first dimension and the sales feature sample data of the second dimension through the prediction value adjustment network in the replenishment quantity prediction model to be trained to obtain the sales forecast adjustment value; constraining the sales forecast adjustment value according to the plan sample data of the second dimension to obtain the replenishment quantity prediction value and the loss function value of the first dimension.

[0006] In some embodiments, constraining the sales forecast adjustment value according to the planned sample data of the second dimension, obtaining the replenishment quantity forecast value and the loss function value includes: generating a first loss function value according to the sales forecast adjustment value and the planned sample data of the second dimension, wherein the loss function value is the first loss function value.

[0007] In some embodiments, generating a first loss function value based on the sales forecast adjustment value and the planned sample data of the second dimension includes: using the sales forecast adjustment value as the replenishment quantity forecast value of the first dimension, and determining the first loss function value based on the difference between the replenishment quantity forecast value and the replenishment quantity value of the corresponding time period, and the difference between the planned sample data of the second dimension and the replenishment quantity forecast value of the second dimension, wherein the replenishment quantity forecast value of the second dimension is the sum of the sales forecast adjustment values ​​of the sales statistical objects of the first dimension included in the second dimension.

[0008] In some embodiments, obtaining the replenishment quantity forecast value and loss function value of the first dimension according to the planned sample data constraining the sales forecast adjustment value of the second dimension includes: adjusting the sales forecast adjustment value based on the planned constraint network according to the planned sample data and the sales forecast adjustment value of the second dimension to obtain the replenishment quantity forecast value of the first dimension; determining the second loss function value according to the difference between the replenishment quantity forecast value of the first dimension and the replenishment quantity value of the corresponding time period, and the loss function value is the second loss function value.

[0009] In some embodiments, according to the planned sample data and the sales forecast adjustment value of the second dimension, the sales forecast adjustment value is adjusted based on the planned constraint network to obtain the replenishment quantity forecast value of the first dimension, which includes: adjusting the sales forecast adjustment value according to the difference between the planned sample data of the second dimension and the replenishment quantity forecast value of the second dimension to obtain the replenishment quantity forecast value of the first dimension, wherein the replenishment quantity forecast value of the second dimension is the sum of the sales forecast adjustment values ​​of the sales statistical objects of the first dimension included in the second dimension.

[0010] In some embodiments, the sales forecast adjustment value is adjusted according to the difference between the planned sample data of the second dimension and the replenishment quantity forecast value of the second dimension, and obtaining the replenishment quantity forecast value of the first dimension includes: determining the adjustment coefficient according to the ratio of the replenishment quantity forecast value of the first dimension to the replenishment quantity forecast value of the second dimension; adjusting the sales forecast adjustment value according to the product of the adjustment coefficient and the difference between the planned sample data of the second dimension and the replenishment quantity forecast value of the second dimension, and determining the replenishment quantity forecast value of the first dimension.

[0011] In some embodiments, the sales prediction sample data is adjusted through a prediction value adjustment network in a replenishment quantity prediction model to be trained according to the sales feature sample data of the first dimension and the sales feature sample data of the second dimension, and obtaining the sales prediction adjustment value includes: obtaining the prediction residual value of the first dimension through a codec network according to the sales prediction sample data, the sales feature sample data of the first dimension and the sales feature sample data of the second dimension; and adjusting the sales prediction sample data through a residual network according to the prediction residual value of the first dimension to obtain the sales prediction value.

[0012] In some embodiments, the first dimension is a SKU (Stock Keeping Unit) dimension, and the second dimension includes multiple SKUs.

[0013] According to one aspect of some embodiments of the present disclosure, a replenishment quantity prediction method is proposed, including: obtaining sales forecast data, sales feature data of a first dimension, sales feature data of a second dimension, and plan data of the second dimension; obtaining a replenishment quantity prediction value of the first dimension through a replenishment quantity prediction model according to the sales forecast data, the sales feature data of the first dimension, the sales feature data of the second dimension, and the plan data of the second dimension, wherein the replenishment quantity prediction model is generated by training according to any one of the replenishment quantity model training methods described above.

[0014] In some embodiments, the replenishment quantity prediction method further includes: performing a replenishment operation according to the replenishment quantity prediction value of the first dimension.

[0015] According to one aspect of some embodiments of the present disclosure, a training device for a replenishment quantity prediction model is proposed, comprising: a sample acquisition unit, configured to acquire sales forecast sample data, sales feature sample data of a first dimension, sales feature sample data of a second dimension, and planned sample data of the second dimension, wherein the second dimension includes multiple sales statistical objects of the first dimension; a training unit, configured to adjust the sales forecast sample data based on the replenishment quantity prediction model to be trained according to the sales feature sample data of the first dimension, the sales feature sample data of the second dimension, and the planned sample data of the second dimension, and acquire a loss function value until the training is completed to determine the replenishment quantity prediction model, wherein the parameters of the replenishment quantity prediction model to be trained are adjusted according to the loss function value.

[0016] According to one aspect of some embodiments of the present disclosure, a replenishment quantity prediction device is proposed, comprising: a data acquisition unit, configured to acquire sales forecast data, sales feature data of a first dimension, sales feature data of a second dimension, and plan data of the second dimension; a prediction unit, configured to acquire replenishment quantity prediction data through a replenishment quantity prediction model based on the sales forecast data, the sales feature data of the first dimension, the sales feature data of the second dimension, and the plan data of the second dimension, wherein the replenishment quantity prediction model is generated by training according to any one of the replenishment quantity model training methods described above.

[0017] In some embodiments, the prediction device further includes: a replenishment unit configured to perform a replenishment operation according to the replenishment quantity prediction value of the first dimension.

[0018] According to one aspect of some embodiments of the present disclosure, a data processing device is provided, comprising: a memory; and a processor coupled to the memory, wherein the processor is configured to execute any one of the above methods based on instructions stored in the memory.

[0019] According to one aspect of some embodiments of the present disclosure, a non-transitory computer-readable storage medium is provided, on which computer program instructions are stored, and when the instructions are executed by a processor, the steps of any one of the above methods are implemented.

[0020] According to one aspect of some embodiments of the present disclosure, a computer program product is provided, including a computer program or instructions, and when the computer program or instructions are executed by a processor, the steps of any one of the above methods are implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide further understanding of the present disclosure and constitute a part of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute improper limitations on the present disclosure.

[0022] Figure 1 The present invention is a flowchart of some embodiments of the training method of the replenishment quantity prediction model disclosed in the present invention.

[0023] Figure 2A Flowcharts of other embodiments of the training method of the replenishment quantity prediction model disclosed in the present invention.

[0024] Figure 2B Schematic diagrams of other embodiments of the replenishment quantity prediction model in the replenishment quantity prediction model training method disclosed in the present invention.

[0025] Figure 3A Flowcharts of some further embodiments of the method for training a replenishment quantity prediction model disclosed herein.

[0026] Figure 3B Schematic diagrams of some further embodiments of the replenishment quantity prediction model in the training method of the replenishment quantity prediction model disclosed in the present invention.

[0027] Figure 4 The present invention is a flowchart of some embodiments of the replenishment quantity prediction method disclosed in the present invention.

[0028] Figure 5 Schematic diagrams of some embodiments of a training device for a replenishment quantity prediction model disclosed herein.

[0029] Figure 6 Schematic diagrams of some embodiments of the replenishment quantity prediction device disclosed in the present invention.

[0030] Figure 7 Schematic diagrams of some embodiments of the data processing method disclosed herein.

[0031] Figure 8 Schematic diagrams of other embodiments of the data processing method disclosed herein. DETAILED DESCRIPTION

[0032] The technical solution of the present disclosure is further described in detail below through the accompanying drawings and embodiments.

[0033] The planning system is the foundation of the supply chain. Accurate and efficient planning can provide more adequate guidance for supply chain operations and ensure that decisions on commodity granularity, such as replenishment and transfer, are effectively executed. The effective implementation of the plan is related to the effective scheduling of overall resources.

[0034] In related technologies, the planning system is mainly implemented in each link through a set of plans, and then split into product dimensions through rules. For product prediction, there is only a single prediction value as the input of the decision, which is finally executed. A single static plan cannot respond to complex changes in time, and it is difficult to provide probabilistic flexibility for various scenarios, resulting in a deterioration in the automation level of the supply chain and cost loss in complex scenarios.

[0035] In response to the above problems, the present disclosure proposes a replenishment quantity prediction method, device, model training method, device and storage medium, which use plans to constrain based on sales forecasts so that the replenishment quantity can both meet market demand and comply with the overall plan, thereby improving the accuracy of replenishment quantity prediction.

[0036] The flowcharts of some embodiments of the training method of the replenishment quantity prediction model disclosed in the present invention are as follows: Figure 1 shown.

[0037] In step S11, sales forecast sample data, sales feature sample data of the first dimension, sales feature sample data of the second dimension, and planned sample data of the second dimension are obtained. The second dimension includes multiple sales statistical objects of the first dimension. The sales forecast sample data is the sample data of the sales forecast data, the sales feature sample data of the first dimension is the sample data of the sales feature data of the first dimension, the sales feature sample data of the second dimension is the sample data of the sales feature data of the second dimension, and the planned sample data of the second dimension is the sample data of the planned data of the second dimension, so that the types of sample data in training correspond to the input information in the prediction process one by one, thereby improving the matching degree between the model and the application scenario.

[0038] In some embodiments, the first dimension is a SKU dimension, and the second dimension includes multiple SKUs, for example, the second dimension is a three-level department dimension, so as to improve the accuracy of replenishment quantity forecasting and facilitate replenishment for each SKU. In some embodiments, the first dimension is a dimension including multiple SKUs, and the second dimension includes multiple SKUs of each dimension in the first dimension, for example, the first dimension is a three-level department dimension, and the second dimension is a second-level department dimension, etc. In some embodiments, the second dimension is one or more dimensions higher than the first dimension. Through such a method, the first dimension can be used as the forecast granularity of replenishment, and the plan can be used to constrain at a granularity higher than the first dimension, thereby improving the flexibility of replenishment forecasting.

[0039] In some embodiments, the sales forecast sample data is a sales forecast result for a time period in which the actual sales volume is known. In some embodiments, the sales forecast sample data can be obtained by predicting historical sales data. In some embodiments, the sales forecast result generated at a past time point can be extracted as the sales forecast sample data, thereby eliminating the need to perform a forecast operation to obtain the sales forecast sample data, reducing the data processing burden, and improving the efficiency of obtaining sample data.

[0040] In some embodiments, the length of the time period corresponding to each sales forecast sample data can be set or adjusted as needed, such as one or more months, so as to improve the flexibility of replenishment control. In some embodiments, the sales forecast sample data includes sales forecast samples of each first-dimensional sales statistical object in the second dimension, so as to ensure the accuracy of the plan's constraints on replenishment.

[0041] In some embodiments, the sales feature sample data is a feature value determined based on historical sales data. In some embodiments, the sales feature sample data is obtained by calculating the mean or variance of the historical sales data. In some embodiments, the sales feature sample data of the first dimension is obtained by calculating the mean or variance of the historical sales data of the first dimension; in some embodiments, the sales feature sample data of the second dimension is obtained by calculating the mean or variance of the historical sales data of the second dimension. In some embodiments, the historical sales data of the second dimension is the sum of the historical sales data of the sales statistics objects of the first dimension contained in the second dimension. Through the method in the embodiments shown above, the sales fluctuations of different dimensions are taken into account in the subsequent processing, thereby improving the prediction accuracy.

[0042] In step S12, according to the sales feature sample data of the first dimension, the sales feature sample data of the second dimension, and the planned sample data of the second dimension, the sales forecast sample data is adjusted based on the replenishment quantity forecast model to be trained, and the loss function value is obtained until the training is completed to determine the replenishment quantity forecast model, wherein the parameters of the replenishment quantity forecast model to be trained are adjusted according to the loss function value.

[0043] In some embodiments, the planned sample data of the second dimension is used as a constraint parameter to constrain the process of predicting the replenishment prediction value based on the sales feature sample data of the first dimension, the sales feature sample data of the second dimension and the sales forecast sample data, thereby realizing the combination of planned sample data on the basis of sales replenishment.

[0044] In some embodiments, the sales forecast sample data is first adjusted according to the sales feature sample data of the first dimension and the sales feature sample data of the second dimension to obtain the sales forecast adjustment value, and the sales forecast adjustment value is further constrained by the planned sample data of the second dimension to obtain the replenishment quantity forecast value and the loss function value of the first dimension. The constraint method may be through the loss function constraint, or the sales forecast adjustment value is adjusted according to the planned sample data of the second dimension. This improves the flexibility of using the planned sample data of the second dimension to constrain the replenishment quantity forecast value while realizing the combination of the planned sample data on the basis of sales volume replenishment.

[0045] In some embodiments, the prediction residual value of the first dimension can be obtained by first encoding and decoding the sales forecast sample data, the sales feature sample data of the first dimension, and the sales feature sample data of the second dimension. In some embodiments, the prediction residual value corresponds to the sales statistical object in the first dimension. Further, the prediction residual value of the first dimension is used to adjust the sales forecast sample data to obtain the sales forecast value. Through such a method, the sales forecast sample data can be adjusted based on the sales fluctuations of different dimensions, and the replenishment quantity forecast value can be further determined based on the adjusted data to improve the accuracy of the replenishment quantity forecast.

[0046] Based on the method in the embodiment shown above, a plan is used to constrain the sales forecast so that the replenishment quantity can meet the sales demand and comply with the overall plan, thereby improving the accuracy of the replenishment quantity forecast.

[0047] In some embodiments, the method of using the planned sample data of the second dimension to constrain the sales forecast adjustment value is through a loss function constraint, then the above step S12 is as follows Figure 2A shown.

[0048] In step 221, the sales forecast sample data is adjusted through the prediction value adjustment network in the replenishment quantity prediction model to be trained according to the sales feature sample data of the first dimension and the sales feature sample data of the second dimension to obtain the sales forecast adjustment value.

[0049] In some embodiments, Figure 2BAs shown, the sales forecast adjustment value is obtained through the codec network 201 and the forecast value adjustment network 202. In some embodiments, the codec network 201 is a two-layer codec-decoder structure, wherein the coding part can be implemented by TCN (temporal convolutional network), MQCNN (Multi-Quantile Convolutional Neural Network), etc., and the decoding part is solved by MLP (Multilayer Perceptron), ATTENTION (attention) mechanism, etc. The codec network 201 outputs the predicted residual value ∈ i Further, the prediction value adjustment network 202 obtains the sales forecast adjustment value according to the prediction residual value and the sales forecast sample data. In some embodiments, the prediction value adjustment network 202 is a residual network, which learns the adjustment value of the sales forecast sample data of the first dimension of the input, and the design formula of the residual network is shown in formula (1).

[0050]

[0051] In this formula, is the sales forecast adjustment value of object i, the sales forecast sample data of object i, ∈ i is the predicted residual value of object i, parameter w i Through training adjustment, i is the identifier of the sales statistics object in the first dimension, such as the SKU identifier. The sales forecast adjustment value output by the forecast value adjustment network 202 corresponds to the sales statistics object in the first dimension one by one, that is, the forecast value adjustment network 202 outputs The number of sales statistics objects in the first dimension is n.

[0052] In step 222, a first loss function value is generated according to the sales forecast adjustment value and the planned sample data of the second dimension, wherein the loss function value of the replenishment quantity forecast model is the first loss function value. In some embodiments, the sales forecast adjustment value is used as the replenishment quantity forecast value of the first dimension, and the first loss function value is determined according to the difference between the replenishment quantity forecast value and the replenishment quantity value of the corresponding time period, and the difference between the planned sample data of the second dimension and the replenishment quantity forecast value of the second dimension, wherein the replenishment quantity forecast value of the second dimension is the sum of the sales forecast adjustment values ​​of the sales statistics objects of the first dimension included in the second dimension.

[0053] In some embodiments, the deviation between the plan and the actual in the first loss function is used to ensure that the prediction is close to the plan. In some embodiments, the formula of the first loss function is shown in formula (2):

[0054]

[0055] In the above formula, loss1 is the value of the first loss function, α is the hyperparameter, and y i is the replenishment quantity value (real value) of object i in the corresponding time period, is the sales forecast adjustment value of object i, Y plan is the planned sample data for the second dimension, is the replenishment quantity forecast value of the second dimension, where The number of sales statistics objects in the first dimension is n. In some embodiments, L(·) uses a mean square error function. By increasing the learning of the hyperparameter α and This ensures that the prediction accuracy of the first dimension object is guaranteed, and the prediction value of the second dimension is close to the plan, thus achieving the prediction close to the plan. In this way, the sales forecast adjustment value is used as the replenishment quantity forecast value of the first dimension, which reduces the computational burden of performing replenishment quantity forecast after training is completed and improves the prediction efficiency.

[0056] In some embodiments, the method of using the planned sample data of the second dimension to constrain the sales forecast adjustment value is to adjust the sales forecast adjustment value according to the planned sample data of the second dimension. Then, the above step S12 is as follows: Figure 3A shown.

[0057] In step 321, the sales forecast sample data is adjusted through the prediction value adjustment network in the replenishment quantity prediction model to be trained according to the sales feature sample data of the first dimension and the sales feature sample data of the second dimension to obtain the sales forecast adjustment value.

[0058] In some embodiments, Figure 3B As shown, the sales forecast adjustment value is obtained through the codec network 301 and the prediction value adjustment network 302. In some embodiments, the codec network 201 is a two-layer codec-decoder structure, wherein the coding part can be implemented by TCN, MQCNN, etc., and the decoding part is solved by MLP, ATTENTION, etc. The codec network 301 outputs the prediction residual value ∈ i Further, the prediction value adjustment network 302 obtains the sales forecast adjustment value according to the prediction residual value and the sales forecast sample data. In some embodiments, the prediction value adjustment network 302 is a residual network, which learns the adjustment value of the sales forecast sample data of the first dimension of the input, and the design formula of the residual network is shown in formula (1).

[0059]

[0060] In this formula, is the sales forecast adjustment value of object i, the sales forecast sample data of object i, ∈ i is the predicted residual value of object i, parameter wi i is the identifier of the sales statistics object in the first dimension, such as the SKU identifier. The sales forecast adjustment value output by the forecast value adjustment network 302 corresponds to the sales statistics object in the first dimension, that is, the forecast value adjustment network 302 outputs The number of sales statistics objects in the first dimension is n.

[0061] In step 322, according to the planned sample data and the sales forecast adjustment value of the second dimension, the sales forecast adjustment value is adjusted based on the planning constraint network to obtain the replenishment quantity forecast value of the first dimension.

[0062] In some embodiments, the sales forecast adjustment value is adjusted according to the difference between the planned sample data of the second dimension and the replenishment quantity forecast value of the second dimension, and the replenishment quantity forecast value of the first dimension is obtained, so that the replenishment quantity forecast value can be subject to the plan constraint. In some embodiments, the adjustment coefficient is determined according to the ratio of the replenishment quantity forecast value of the first dimension to the replenishment quantity forecast value of the second dimension, and then the sales forecast adjustment value is adjusted according to the product of the adjustment coefficient and the difference between the planned sample data of the second dimension and the replenishment quantity forecast value of the second dimension, and the replenishment quantity forecast value of the first dimension is determined. Through the method in this embodiment, the replenishment quantity forecast value of the first dimension can be made to fully conform to the planned sample data of the second dimension, thereby improving the execution accuracy of the plan.

[0063] In some embodiments, Figure 3B As shown, the planning constraint network 303 obtains the sales forecast adjustment value output by the forecast value adjustment network 302 The number of sales statistics objects in the first dimension is n, and formula (3) is used to obtain the replenishment quantity forecast value of the first dimension.

[0064]

[0065] is the replenishment quantity forecast value of the first dimension of object i, is the sales forecast adjustment value of object i, Y plan is the planned sample data for the second dimension, is the replenishment quantity forecast value of the second dimension, where The number of sales statistics objects in the first dimension is n. β i The sum of is 1.

[0066] In step 323, a second loss function value is determined based on the difference between the replenishment quantity prediction value of the first dimension and the replenishment quantity value of the corresponding time period, and the loss function value of the replenishment quantity prediction model is the second loss function value. In some embodiments, the second loss function 304 is as shown in formula (4).

[0067]

[0068] loss2 is the second loss function value, y i The replenishment quantity value (real value) of object i in the corresponding time period, is the replenishment quantity prediction value of the first dimension of object i. In some embodiments, L(·) adopts the mean square error function. Using the method in this embodiment, The examples are shown in Table 1. The examples in Table 1 are only examples of prediction results for some samples when the first dimension is the SKU dimension, and do not constitute an improper limitation to the present disclosure.

[0069] Table 1 Examples of prediction results during training

[0070] SKU identification True value Sales forecast Plan adjustment increment Replenishment quantity forecast 10001 10 7 4 11 10002 20 15 4 19 10003 30 20 11 31 10005 40 30 9 39 10007 50 60 -9 51 10008 60 45 14 59 Grand total 210 177 33 210

[0071] In Table 1, the cumulative value of the actual value is the planned value. It can be seen that the total amount of the replenishment quantity forecast value is exactly the same as the total amount of the actual value, thus achieving a complete match with the plan.

[0072] Based on the method of the embodiment shown above, the output replenishment quantity forecast value of the first dimension can be made completely consistent with the plan, thereby improving the accuracy of plan execution, and at the same time can be consistent with sales forecasts, thereby improving the accuracy of replenishment quantity forecasts.

[0073] The flowcharts of some embodiments of the replenishment quantity prediction method disclosed in the present invention are as follows: Figure 4 As shown, steps 401 and 402 are included.

[0074] In step 401, sales forecast data, sales feature data of the first dimension, sales feature data of the second dimension, and plan data of the second dimension are obtained. In some embodiments, the sales forecast data can be determined based on a sales forecast algorithm in the related art.

[0075] In some embodiments, the sales characteristic data is a characteristic value determined based on historical sales data. In some embodiments, the sales characteristic data is obtained by calculating the mean or variance of the historical sales data. In some embodiments, the sales characteristic data of the first dimension is obtained by calculating the mean or variance of the historical sales data of the first dimension; in some embodiments, the sales characteristic data of the second dimension is obtained by calculating the mean or variance of the historical sales data of the second dimension. In some embodiments, the historical sales data of the second dimension is the sum of the historical sales data of the sales statistics objects of the first dimension contained in the second dimension. Through the method in the embodiments shown above, the sales fluctuations of different dimensions are taken into account in the subsequent processing, thereby improving the prediction accuracy.

[0076] In step 402, a replenishment quantity forecast value of the first dimension is obtained through a replenishment quantity forecast model according to the sales forecast data, the sales feature data of the first dimension, the sales feature data of the second dimension, and the planned data of the second dimension. The replenishment quantity forecast model is generated by training according to any of the training methods of the replenishment quantity model described above.

[0077] In some embodiments, when the demand forecast result is close to the plan, the following method can be used: Figure 2A , 2B The method in the illustrated embodiment constructs and trains a replenishment quantity prediction model, and then uses the trained model to predict future replenishment quantities. In some embodiments, when the demand forecast result is a perfect match with the plan, the following method can be used: Figure 3A , 3B The method in the illustrated embodiment constructs and trains a replenishment quantity prediction model, and then uses the trained model to predict future replenishment quantities. Through such a method, a replenishment quantity prediction model can be selected according to demand, thereby improving the matching degree between the output result and demand and improving the flexibility of prediction.

[0078] Based on the method in the embodiment shown above, a plan is used to constrain the sales forecast so that the replenishment quantity can meet the sales demand and comply with the overall plan, thereby improving the automation level of the supply chain in complex scenarios, reducing cost losses, and thus improving the accuracy of the replenishment quantity forecast.

[0079] In some embodiments, by adjusting the first dimension and the corresponding second dimension, it is possible to achieve predictions for different levels of warehousing operations, reduce human interference and reduced accuracy caused by layer-by-layer disassembly, and improve the accuracy of global replenishment quantity predictions.

[0080] In some embodiments, Figure 4 As shown, the replenishment quantity prediction method disclosed in the present invention also includes step 403.

[0081] In step 403, a replenishment operation is performed according to the replenishment quantity forecast value of the first dimension. In some embodiments, the first dimension is the SKU dimension, and the replenishment operation is performed according to the replenishment quantity forecast result for each SKU. In some embodiments, the first dimension is the three-level department dimension, and the replenishment operation of the corresponding three-level department is constrained according to the replenishment quantity forecast result for each three-level department.

[0082] Based on the method in this embodiment, the replenishment quantity prediction result can be applied to the replenishment operation of the corresponding dimension, so that the replenishment quantity of the first dimension is consistent with the sales volume and the plan, thereby improving the accuracy of the replenishment operation.

[0083] Schematic diagrams of some embodiments of the training device 51 of the replenishment quantity prediction model disclosed in the present invention are as follows Figure 5 shown.

[0084] The sample acquisition unit 511 can acquire sales forecast sample data, sales feature sample data of the first dimension, sales feature sample data of the second dimension, and planned sample data of the second dimension. The second dimension includes multiple sales statistical objects of the first dimension. The sales forecast sample data is the sample data of the sales forecast data, the sales feature sample data of the first dimension is the sample data of the sales feature data of the first dimension, the sales feature sample data of the second dimension is the sample data of the sales feature data of the second dimension, and the planned sample data of the second dimension is the sample data of the planned data of the second dimension, so that the types of sample data in training correspond to the input information in the prediction process one by one, thereby improving the matching degree between the model and the application scenario.

[0085] In some embodiments, the first dimension is a SKU dimension, and the second dimension includes multiple SKUs, for example, the second dimension is a three-level department dimension, so as to improve the accuracy of replenishment quantity forecasting and facilitate replenishment for each SKU. In some embodiments, the first dimension is a dimension including multiple SKUs, and the second dimension includes multiple SKUs of each dimension in the first dimension, for example, the first dimension is a three-level department dimension, and the second dimension is a second-level department dimension, etc. In some embodiments, the second dimension is one or more dimensions higher than the first dimension. Through such a method, the first dimension can be used as the forecast granularity of replenishment, and the plan can be used to constrain at a granularity higher than the first dimension, thereby improving the flexibility of replenishment forecasting.

[0086] In some embodiments, the sample acquisition unit 511 can execute the method in any of the embodiments of the above step S11.

[0087] The training unit 512 can adjust the sales forecast sample data based on the replenishment quantity forecast model to be trained according to the sales feature sample data of the first dimension, the sales feature sample data of the second dimension, and the planned sample data of the second dimension, and obtain the loss function value until the training is completed to determine the replenishment quantity forecast model, wherein the parameters of the replenishment quantity forecast model to be trained are adjusted according to the loss function value. In some embodiments, the planned sample data of the second dimension is used as a constraint parameter to constrain the process of predicting the replenishment forecast value according to the sales feature sample data of the first dimension, the sales feature sample data of the second dimension, and the sales forecast sample data, thereby realizing the combination of the planned sample data on the basis of sales replenishment.

[0088] In some embodiments, the training unit 512 can perform the above step S12, and Figure 2A , 2B The method of any of the embodiments shown.

[0089] The training device of the replenishment quantity prediction model in the embodiment of the present disclosure uses a plan to constrain based on the sales forecast, so that the replenishment quantity can both meet the sales demand and comply with the overall plan, thereby improving the accuracy of the replenishment quantity prediction.

[0090] Schematic diagrams of some embodiments of the replenishment quantity prediction device 62 of the present disclosure are as follows Figure 6 shown.

[0091] The data acquisition unit 621 can acquire sales forecast data, sales feature data of the first dimension, sales feature data of the second dimension, and plan data of the second dimension. In some embodiments, the sales forecast data can be determined based on a sales forecast algorithm in the related art.

[0092] In some embodiments, the sales characteristic data is a characteristic value determined based on historical sales data. In some embodiments, the sales characteristic data is obtained by calculating the mean or variance of the historical sales data. In some embodiments, the sales characteristic data of the first dimension is obtained by calculating the mean or variance of the historical sales data of the first dimension; in some embodiments, the sales characteristic data of the second dimension is obtained by calculating the mean or variance of the historical sales data of the second dimension. In some embodiments, the historical sales data of the second dimension is the sum of the historical sales data of the sales statistics objects of the first dimension contained in the second dimension, so that the sales fluctuations of different dimensions are taken into account in the subsequent processing, thereby improving the prediction accuracy.

[0093] The prediction unit 622 can obtain the replenishment quantity prediction value of the first dimension through the replenishment quantity prediction model according to the sales forecast data, the sales feature data of the first dimension, the sales feature data of the second dimension, and the plan data of the second dimension. The replenishment quantity prediction model is trained and generated according to any of the training methods of the replenishment quantity model described above.

[0094] In some embodiments, when the demand forecast result is close to the plan, the following method can be used: Figure 2A , 2B In the embodiment shown, the trained model predicts the future replenishment quantity; in some embodiments, when the demand forecast result is completely matched with the plan, the following method can be used: Figure 3A , 3B The trained model in the illustrated embodiment predicts future replenishment quantities, so that a replenishment quantity prediction model can be selected according to demand, thereby improving the matching degree between the output result and demand and improving the flexibility of the prediction.

[0095] The replenishment quantity prediction device in the embodiment of the present disclosure can use a plan to constrain based on the sales forecast, so that the replenishment quantity can both meet the sales demand and comply with the overall plan, thereby improving the accuracy of the replenishment quantity prediction.

[0096] In some embodiments, Figure 6 As shown, the replenishment quantity prediction device also includes a replenishment unit 623, which can perform a replenishment operation according to the replenishment quantity prediction value of the first dimension. In some embodiments, the first dimension is the SKU dimension, and the replenishment operation is performed according to the replenishment quantity prediction result for each SKU. In some embodiments, the first dimension is the three-level department dimension, and the replenishment operation of the corresponding three-level department is constrained according to the replenishment quantity prediction result for each three-level department.

[0097] The replenishment quantity prediction device in the embodiment of the present disclosure can apply the replenishment quantity prediction result to the replenishment operation of the corresponding dimension, so that the replenishment quantity of the first dimension is consistent with the sales volume and the plan, thereby improving the accuracy of the replenishment operation.

[0098] A structural diagram of an embodiment of the data processing device disclosed in the present invention is shown in FIG. Figure 7 As shown. The data processing device includes a memory 701 and a processor 702. Among them: the memory 701 can be a disk, a flash memory or any other non-volatile storage medium. The memory is used to store the instructions in the corresponding embodiments of the training method of the replenishment quantity prediction model or the replenishment quantity prediction method mentioned above. The processor 702 is coupled to the memory 701 and can be implemented as one or more integrated circuits, such as a microprocessor or a microcontroller. The processor 702 is used to execute the instructions stored in the memory, which can improve the accuracy of the replenishment quantity prediction.

[0099] In one embodiment, it is also possible to Figure 8 As shown, the data processing device 800 includes a memory 801 and a processor 802. The processor 802 is coupled to the memory 801 via a BUS bus 803. The data processing device 800 can also be connected to an external storage device 805 via a storage interface 804 to call external data, and can also be connected to a network or another computer system (not shown) via a network interface 806. No further detailed description will be given here.

[0100] In this embodiment, the accuracy of replenishment quantity prediction can be improved by storing data instructions in a memory and processing the instructions through a processor.

[0101] In another embodiment, a computer-readable storage medium stores computer program instructions thereon, which, when executed by a processor, implement the training method of a replenishment quantity prediction model or the steps of the method in the corresponding embodiment of the replenishment quantity prediction method. It should be understood by those skilled in the art that the embodiments of the present disclosure may be provided as methods, apparatus, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable non-transitory storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0102] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0103] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0104] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0105] So far, the present disclosure has been described in detail. In order to avoid obscuring the concept of the present disclosure, some details known in the art are not described. Based on the above description, those skilled in the art can fully understand how to implement the technical solution disclosed here.

[0106] The method and apparatus of the present disclosure may be implemented in many ways. For example, the method and apparatus of the present disclosure may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present disclosure are not limited to the order specifically described above, unless otherwise specifically stated. In addition, in some embodiments, the present disclosure may also be implemented as a program recorded in a recording medium, which includes machine-readable instructions for implementing the method according to the present disclosure. Therefore, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.

[0107] It should be noted that the terms "first", "second", etc. in the specification, claims, and drawings of the present disclosure are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present disclosure rather than to limit it. Although the present disclosure has been described in detail with reference to the preferred embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present disclosure can still be modified or some technical features can be replaced by equivalents without departing from the spirit of the technical solution of the present disclosure, which should be included in the scope of the technical solution for protection of the present disclosure.

Claims

1. A training method for a replenishment quantity prediction model, include: Acquire sales forecast sample data, sales feature sample data of a first dimension, sales feature sample data of a second dimension, and planned sample data of the second dimension, wherein the second dimension includes a plurality of sales statistics objects of the first dimension; According to the sales feature sample data of the first dimension, the sales feature sample data of the second dimension, and the planned sample data of the second dimension, the sales forecast sample data is adjusted based on the replenishment quantity forecast model to be trained, and the loss function value is obtained until the training is completed to determine the replenishment quantity forecast model, wherein the parameters of the replenishment quantity forecast model to be trained are adjusted according to the loss function value.

2. The training method according to claim 1, in, The adjusting the sales volume forecast sample data based on the replenishment quantity forecast model to be trained and obtaining the loss function value includes: According to the sales feature sample data of the first dimension and the sales feature sample data of the second dimension, the sales forecast sample data is adjusted by a prediction value adjustment network in the replenishment quantity prediction model to be trained to obtain a sales forecast adjustment value; The sales forecast adjustment value is constrained according to the planned sample data of the second dimension, and the replenishment quantity forecast value of the first dimension and the loss function value are obtained.

3. The training method according to claim 2, in, The step of constraining the sales forecast adjustment value according to the planned sample data of the second dimension and obtaining the replenishment quantity forecast value and the loss function value comprises: A first loss function value is generated according to the sales forecast adjustment value and the planned sample data of the second dimension, wherein the loss function value is the first loss function value.

4. The training method according to claim 3, in, The generating a first loss function value according to the sales forecast adjustment value and the planned sample data of the second dimension comprises: The sales forecast adjustment value is used as the replenishment quantity forecast value of the first dimension, and the first loss function value is determined according to the difference between the replenishment quantity forecast value and the replenishment quantity value of the corresponding time period, and the difference between the planned sample data of the second dimension and the replenishment quantity forecast value of the second dimension, wherein the replenishment quantity forecast value of the second dimension is the sum of the sales forecast adjustment values ​​of the sales statistical objects of the first dimension included in the second dimension.

5. The training method according to claim 2, in, The step of constraining the sales forecast adjustment value according to the planned sample data of the second dimension and obtaining the replenishment quantity forecast value and the loss function value of the first dimension includes: According to the planned sample data of the second dimension and the sales forecast adjustment value, adjusting the sales forecast adjustment value based on the planning constraint network to obtain a replenishment quantity forecast value of the first dimension; A second loss function value is determined based on the difference between the replenishment quantity prediction value of the first dimension and the replenishment quantity value of the corresponding time period, and the loss function value is the second loss function value.

6. The training method according to claim 5, in, The step of adjusting the sales forecast adjustment value based on the plan sample data of the second dimension and the sales forecast adjustment value based on the plan constraint network to obtain the replenishment quantity forecast value of the first dimension includes: The sales forecast adjustment value is adjusted according to the difference between the planned sample data of the second dimension and the replenishment quantity forecast value of the second dimension to obtain the replenishment quantity forecast value of the first dimension, wherein the replenishment quantity forecast value of the second dimension is the sum of the sales forecast adjustment values ​​of the sales statistical objects of the first dimension included in the second dimension.

7. The training method according to claim 6, in, The step of adjusting the sales forecast adjustment value according to the difference between the planned sample data of the second dimension and the replenishment quantity forecast value of the second dimension to obtain the replenishment quantity forecast value of the first dimension includes: Determine an adjustment coefficient according to the ratio of the replenishment quantity forecast value of the first dimension to the replenishment quantity forecast value of the second dimension; The sales forecast adjustment value is adjusted according to the product of the adjustment coefficient and the difference between the planned sample data of the second dimension and the replenishment quantity forecast value of the second dimension to determine the replenishment quantity forecast value of the first dimension.

8. The training method according to claim 2, in, The step of adjusting the sales forecast sample data by a prediction value adjustment network in the replenishment quantity prediction model to be trained based on the sales feature sample data of the first dimension and the sales feature sample data of the second dimension to obtain the sales forecast adjustment value includes: According to the sales prediction sample data, the sales feature sample data of the first dimension and the sales feature sample data of the second dimension, obtaining the prediction residual value of the first dimension through the encoding and decoding network; According to the prediction residual value of the first dimension, the sales forecast sample data is adjusted through a residual network to obtain the sales forecast value.

9. The training method according to claim 1, in, The first dimension is the smallest stock keeping unit (SKU) dimension, and the second dimension includes a plurality of the SKUs.

10. A replenishment quantity prediction method, include: Obtain sales forecast data, sales feature data of a first dimension, sales feature data of a second dimension, and plan data of the second dimension; According to the sales volume forecast data, the sales volume characteristic data of the first dimension, the sales volume characteristic data of the second dimension, and the planned data of the second dimension, a replenishment quantity forecast value of the first dimension is obtained through a replenishment quantity forecast model, wherein the replenishment quantity forecast model is generated by training according to the training method of the replenishment quantity model according to any one of claims 1 to 9.

11. The method according to claim 10, further comprising: include: The replenishment operation is performed according to the replenishment quantity prediction value of the first dimension.

12. A training device for a replenishment quantity prediction model, include: A sample acquisition unit is configured to acquire sales forecast sample data, sales feature sample data of a first dimension, sales feature sample data of a second dimension, and planned sample data of the second dimension, wherein the second dimension includes a plurality of sales statistical objects of the first dimension; The training unit is configured to adjust the sales prediction sample data based on the replenishment quantity prediction model to be trained according to the sales feature sample data of the first dimension, the sales feature sample data of the second dimension, and the planned sample data of the second dimension, and obtain the loss function value until the training is completed to determine the replenishment quantity prediction model, wherein the parameters of the replenishment quantity prediction model to be trained are adjusted according to the loss function value.

13. A replenishment quantity prediction device, include: A data acquisition unit is configured to acquire sales forecast data, sales feature data of a first dimension, sales feature data of a second dimension, and plan data of the second dimension; The prediction unit is configured to obtain replenishment quantity prediction data through a replenishment quantity prediction model based on the sales volume prediction data, the sales volume feature data of the first dimension, the sales volume feature data of the second dimension, and the plan data of the second dimension, wherein the replenishment quantity prediction model is generated by training according to the training method of the replenishment quantity model according to any one of claims 1 to 9.

14. The prediction device according to claim 13, further comprising: include: The replenishment unit is configured to perform a replenishment operation according to the replenishment quantity prediction value of the first dimension.

15. A data processing device, include: Memory; as well as A processor coupled to the memory, the processor being configured to execute the method according to any one of claims 1 to 10 based on instructions stored in the memory.

16. A non-transitory computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 10.

17. A computer program product, comprising a computer program or instructions, wherein when the computer program or instructions are executed by a processor, the steps of the method according to any one of claims 1 to 10 are implemented.

Citation Information

Cited By

  • E-commerce automatic replenishment system based on cloud computing

    CN121961415A