Replenishment quantity prediction method and device for supply chain, storage medium and electronic equipment

Abstract

This disclosure provides a method, apparatus, storage medium, and electronic device for supply chain replenishment quantity forecasting. The method includes: constructing a first feature vector based on the inventory status on the replenishment decision date; inputting the first feature vector into a prediction model, and determining that the output value of the prediction model is the first replenishment quantity corresponding to the current replenishment decision date, wherein the prediction model is used to characterize the correlation between the feature vector and the replenishment quantity. This disclosure obtains a prediction model for characterizing the relationship between the feature vector and the replenishment quantity by training a preset model, uses the prediction model to perform end-to-end replenishment quantity forecasting based on the inventory status on the current replenishment decision date, and uses the predicted replenishment quantity to assist supply chain inventory management decisions, achieving the lowest inventory cost while ensuring service levels.

Description

Technical Field

[0001] This disclosure relates to the field of supply chain, and in particular to a method, apparatus, storage medium, and electronic device for predicting replenishment volume in a supply chain. Background Technology

[0002] In lean supply chain management, a key application scenario involves using historical demand data over a given period, current inventory data, replenishment time, and relevant information about replenishment units (such as product type and geographical coordinates). This information, combined with techniques like time-series forecasting and operations optimization, allows for the determination of replenishment quantities for each unit on the replenishment decision date. The goal is to reduce inventory levels and lower inventory costs while ensuring that inventory levels largely meet demand and prevent supply disruptions.

[0003] The current related technologies involve two stages for determining replenishment quantities. The first stage is based on historical demand and uses time series forecasting technology to predict the demand within a certain period of time in the future. The second stage combines the current inventory level of the replenishment unit and the amount of goods in transit to determine whether the inventory level after the goods arrive at the warehouse can meet a series of predicted demands during the transit period. The replenishment quantity is then calculated by back-calculating the inventory level after the goods arrive at the warehouse and the current inventory level.

[0004] However, when determining the replenishment quantity based on the above methods, it is necessary to first predict the demand in the future time period (mainly the demand between the current replenishment decision date and the next replenishment decision date), and then back-calculate the replenishment quantity based on the predicted demand. In the actual prediction process, there may be deviations in the prediction of demand, resulting in inaccurate calculation of the replenishment quantity, inventory backlog or stockouts, and failure to achieve the goal of reducing inventory costs. Summary of the Invention

[0005] The purpose of this disclosure is to provide a method, apparatus, storage medium, and electronic device for predicting replenishment volume in a supply chain, in order to solve the problem of inaccurate back-calculation results when back-calculating replenishment volume based on predicted demand in the prior art.

[0006] The embodiments of this disclosure adopt the following technical solution: a supply chain replenishment quantity forecasting method, comprising: constructing a first feature vector based on the inventory status on the replenishment decision date; inputting the first feature vector into a forecasting model, and determining that the output value of the forecasting model is the first replenishment quantity corresponding to the current replenishment decision date, wherein the forecasting model is used to characterize the correlation between the feature vector and the replenishment quantity.

[0007] In some embodiments, the prediction model is trained based on the following steps: constructing historical feature vectors corresponding to m replenishment decision days based on historical inventory data for each of the m replenishment decision days; determining a second replenishment quantity corresponding to each historical feature vector to minimize the overall cost between two adjacent replenishment decision days; training a preset model based on a training set formed by all the historical feature vectors and the second replenishment quantities corresponding to all the historical feature vectors until the loss function of the preset model is minimized, thereby obtaining the prediction model.

[0008] In some embodiments, the total cost between two adjacent replenishment decision days is determined based on the following formula:

[0009]

[0010] in, The total cost is represented between the m-th replenishment decision day and the (m+1)-th replenishment decision day, where h is the inventory cost per unit, b is the stockout cost per unit, and v is the total cost. m Let m be the arrival time of the m-th replenishment. Let m be the current inventory level at the time of the m-th replenishment. For from v m The total inventory demand over a period of time s, where s represents any date between the m-th replenishment decision date and the (m+1)-th replenishment decision date, and y m The second replenishment quantity is determined on the m-th replenishment decision day; when the total cost is minimized, the quantity is determined. in,

[0011] In some embodiments, the prediction model is based on the following formula:

[0012]

[0013] Wherein, the x i Let represent the feature vector of the i-th replenishment decision day, N represent the number of training data in the training set, and L represent the loss function of the prediction model. This represents the replenishment quantity on the i-th replenishment decision day.

[0014] In some embodiments, training the preset model based on the training set formed by all the historical feature vectors and the second replenishment quantity corresponding to all the historical feature vectors includes: splitting the training data in the training set into two parts, one part of which is used as the input (x) of the encoder of the preset model. i y i The other part serves as the input to the decoder of the preset model. Where i = 1, ..., t, j = t+1, ..., N, t is used to characterize the splitting parameter of the training data, t = N * 60%.

[0015] In some embodiments, the preset model is any one of the following models: MQRNN network model, LightGBM network model, XGBoost network model, and TCN network model.

[0016] In some embodiments, the inventory status includes at least one or more of the following: the date of the replenishment decision, the replenishment duration, the product category, the inventory of different product categories on the replenishment decision date, and the daily demand for each product category between the previous replenishment decision date and the current replenishment decision date.

[0017] This disclosure also provides a supply chain replenishment quantity prediction device, including: an acquisition module, configured to construct a first feature vector based on the inventory status on the replenishment decision date; and a prediction module, configured to input the first feature vector into a prediction model and determine that the value output by the prediction model is the first replenishment quantity corresponding to the current replenishment decision date, wherein the prediction model is used to characterize the correlation between the feature vector and the replenishment quantity.

[0018] In some embodiments, the system further includes a training module, which is specifically configured to: construct historical feature vectors corresponding to m replenishment decision days based on the historical inventory of m replenishment decision days; determine a second replenishment quantity corresponding to each historical feature vector to minimize the overall cost between two adjacent replenishment decision days; and train a preset model based on a training set formed by all the historical feature vectors and the second replenishment quantities corresponding to all the historical feature vectors until the loss function of the preset model is minimized, thereby obtaining the prediction model.

[0019] This disclosure also provides a storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described supply chain replenishment forecasting method.

[0020] This disclosure also provides an electronic device, including at least a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program in the memory to implement the steps of the above-described supply chain replenishment forecasting method.

[0021] The beneficial effects of this disclosure are as follows: by training a preset model, a prediction model is obtained to characterize the relationship between feature vectors and replenishment quantity. The prediction model is used to predict the end-to-end replenishment quantity based on the inventory status on the current replenishment decision day. Based on the predicted replenishment quantity, the supply chain inventory management decision is assisted, and the lowest inventory cost is achieved while ensuring service level. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the supply chain replenishment forecasting method in the first embodiment of this disclosure;

[0024] Figure 2 This is a flowchart of the training method for the prediction model in the first embodiment of this disclosure;

[0025] Figure 3 This is a schematic diagram of the network structure of the MQRNN network model in the first embodiment of this disclosure;

[0026] Figure 4 This is a schematic diagram of the supply chain replenishment quantity prediction device in the second embodiment of this disclosure;

[0027] Figure 5 This is a schematic diagram of the structure of the electronic device in the fourth embodiment of this disclosure. Detailed Implementation

[0028] Various embodiments and features of this disclosure are described herein with reference to the accompanying drawings.

[0029] It should be understood that various modifications can be made to the embodiments described herein. Therefore, the above description should not be considered as limiting, but merely as an example of embodiments. Other modifications within the scope and spirit of this disclosure will be apparent to those skilled in the art.

[0030] The accompanying drawings, which are included in and form part of this specification, illustrate embodiments of the present disclosure and, together with the general description of the disclosure given above and the detailed description of the embodiments given below, serve to explain the principles of the disclosure.

[0031] These and other features of this disclosure will become apparent from the following description of preferred forms of embodiments given as non-limiting examples, with reference to the accompanying drawings.

[0032] It should also be understood that although this disclosure has been described with reference to some specific examples, many other equivalent forms of this disclosure can be definitively implemented by those skilled in the art, which have the features of the claims and are therefore within the scope of protection defined herein.

[0033] The above and other aspects, features and advantages of this disclosure will become more apparent when taken in conjunction with the accompanying drawings and in view of the following detailed description.

[0034] Specific embodiments of this disclosure are described thereafter with reference to the accompanying drawings; however, it should be understood that the claimed embodiments are merely examples of this disclosure, which may be implemented in various ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that could obscure this disclosure. Therefore, the specific structural and functional details claimed herein are not intended to be limiting, but merely to serve as the basis and representative basis for the claims to teach those skilled in the art to use this disclosure in a variety of substantially any suitable detailed structures.

[0035] This specification may use the phrases “in one embodiment,” “in another embodiment,” “in yet another embodiment,” or “in still another embodiment,” all of which may refer to one or more of the same or different embodiments according to this disclosure.

[0036] In lean supply chain management, a key application scenario involves using historical demand data over a given period, current inventory data, replenishment time, and relevant information about replenishment units (such as product type and geographical coordinates). This information, combined with techniques like time-series forecasting and operations optimization, allows for the determination of replenishment quantities for each unit on the replenishment decision date. The goal is to reduce inventory levels and lower inventory costs while ensuring that inventory levels largely meet demand and prevent supply disruptions.

[0037] The current related technologies involve two stages for determining replenishment quantities. The first stage is based on historical demand and uses time series forecasting technology to predict the demand within a certain period of time in the future. The second stage combines the current inventory level of the replenishment unit and the amount of goods in transit to determine whether the inventory level after the goods arrive at the warehouse can meet a series of predicted demands during the transit period. The replenishment quantity is then calculated by back-calculating the inventory level after the goods arrive at the warehouse and the current inventory level.

[0038] However, when determining the replenishment quantity based on the above methods, it is necessary to first forecast the demand in the future time period (mainly the demand between the current replenishment decision date and the next replenishment decision date), and then back-calculate the replenishment quantity based on the forecasted demand. In the actual forecasting process, there may be deviations in the forecast of demand, resulting in inaccurate calculation of the replenishment quantity, inventory backlog or stockouts, and affecting supply chain management decisions.

[0039] To address the aforementioned problems, the first embodiment of this disclosure provides a supply chain replenishment quantity forecasting method, the flowchart of which is shown below. Figure 1 As shown, it mainly includes steps S1 and S2:

[0040] S1, construct the first feature vector based on the inventory situation on the replenishment decision day.

[0041] A replenishment decision day is the date on which the replenishment quantity of goods is determined based on the replenishment strategy. It is generally set periodically, for example, every Monday or the 1st of each month. The inventory status obtained on the replenishment decision day in this embodiment typically refers to the inventory status of the replenishment unit that needs replenishment, including at least one or more of the following: the date of the replenishment decision day, the replenishment duration, the product category, the inventory quantity of different product categories on the replenishment decision day, and the daily demand for each product category between the previous replenishment decision day and the current replenishment decision day. It may also include the geographical location of the replenishment unit that currently needs replenishment. The replenishment duration refers to the time from the issuance of the replenishment order to the goods being available in the warehouse; for example, the replenishment duration can typically be set to 14 days.

[0042] The first feature vector can be constructed based on the inventory status on the replenishment decision date. This embodiment does not limit the specific construction method of the first feature vector, as long as it can represent the inventory status of the current replenishment unit. It should be noted that the first feature vector in this embodiment needs to be used as an input parameter of the prediction model. Its representation and the inventory status of the current replenishment unit it represents should be exactly the same as the representation and the inventory status of the replenishment units represented by the historical feature vectors used when training the prediction model, so as to ensure the accuracy of the first replenishment quantity value output based on the first feature vector.

[0043] S2, input the first feature vector into the prediction model, and determine that the value output by the prediction model is the first replenishment quantity corresponding to the current replenishment decision day.

[0044] The prediction model is mainly used to characterize the relationship between feature vectors and replenishment quantities to achieve end-to-end replenishment quantity prediction. Specifically, this application uses historical data from each replenishment unit to train a pre-set algorithm model. The trained pre-set model is then used as the prediction model for replenishment quantity prediction. After constructing the first feature vector, it is input into the prediction model, and the output value of the prediction model can be used as the first replenishment quantity corresponding to the current replenishment decision day. This assists in supply chain inventory management decisions. By directly outputting the replenishment quantity using an end-to-end model, errors in predicting demand are avoided. This ensures that the determined replenishment quantity meets the minimum inventory requirements of the replenishment unit while maintaining service levels, thereby reducing inventory costs.

[0045] Specifically, the prediction model used in this embodiment is trained based on the historical inventory of the replenishment unit and the decision-maker's minimum requirement for inventory costs. In some embodiments, the preset model can be any one of the following: MQRNN network model, LightGBM network model, XGBoost network model, and TCN network model. In this embodiment, the MQRNN network model is preferably used as the preset model for training to achieve faster training and more accurate results.

[0046] Figure 2 A flowchart illustrating the training method for the prediction model is shown. Figure 2 As shown, the training steps of the prediction model mainly include steps S21 to S23:

[0047] S21, construct historical feature vectors corresponding to m replenishment decision days based on the historical inventory situation of m replenishment decision days;

[0048] S22, determine the second replenishment quantity corresponding to each historical feature vector based on each historical feature vector, so as to minimize the overall cost between two adjacent replenishment decision days;

[0049] S23. Train the preset model using the training set formed by all historical feature vectors and the second replenishment quantity corresponding to all historical feature vectors until the loss function of the preset model is minimized, and obtain the prediction model.

[0050] It is important to note that the historical inventory data obtained during the training of the predictive model for the m replenishment decision days can be for a single replenishment unit or for multiple different replenishment units. The specific selection of the training data source can be determined based on the actual application requirements of the predictive model. If it is necessary to predict the replenishment quantity that meets the supply demand of a single replenishment unit, the historical inventory data of that replenishment unit can be used to train the predictive model to make the output results more accurate. If it is necessary to ensure that the predictive model has better universality, the historical inventory data of multiple different replenishment units can be used for model training.

[0051] The historical inventory information corresponding to the m replenishment decision days obtained in this embodiment also includes the date of the replenishment decision day, the replenishment duration, the product category, the inventory of different product categories on the replenishment decision day, the demand of each product category on each day between the previous replenishment decision day and the current replenishment decision day, and may also include information such as the geographical latitude of the replenishment unit. After obtaining the above information, a corresponding historical feature vector can be constructed based on the historical inventory information of each replenishment decision day, which can be used as one of the input data of the dataset for model training.

[0052] Furthermore, the model training process typically involves training the various parameters within the model by inputting correct input and output data. The model training is considered complete once the corresponding convergence conditions are met. The objective of this embodiment is to ensure that the replenishment quantity output by the model achieves the lowest possible inventory cost. Therefore, during model training, the input data for the training set should be the historical feature vector for each replenishment decision date and the corresponding second replenishment quantity for each historical feature vector. The second replenishment quantity is the replenishment quantity that minimizes the overall inventory cost within the replenishment cycle corresponding to that replenishment decision date (i.e., the number of days between the current replenishment decision date and the next replenishment decision date).

[0053] When preparing training data, the second replenishment quantity for the current replenishment decision date can be determined by the total cost between two adjacent replenishment decision dates. Specifically, the total cost between two adjacent replenishment decision dates can be determined based on the following formula according to the calculation principle of total cost:

[0054]

[0055] in, The total cost is represented between the m-th replenishment decision day and the (m+1)-th replenishment decision day, where h is the inventory cost per unit, b is the stockout cost per unit, and v is the total cost. m Let m be the arrival time of the m-th replenishment. Let m be the current inventory level at the time of the m-th replenishment. For from v m The total inventory demand over a period of time s, where s represents any date between the m-th replenishment decision date and the (m+1)-th replenishment decision date, and y m This represents the second replenishment quantity on the m-th replenishment decision day. It should be understood that the specific values ​​of inventory cost h and stockout cost b can be calculated based on actual circumstances, and are usually constants; this embodiment does not provide a specific description. m , The values ​​of parameters such as s can be represented or derived from historical feature vectors.

[0056] Based on the above formula (1), when it is necessary to guarantee the overall cost At its minimum, the second replenishment quantity y can be determined. m The calculation formula is:

[0057]

[0058] in, This indicates a floor operation. Therefore, a second replenishment quantity can be determined based on the historical feature vectors to minimize the overall inventory cost within the current replenishment cycle. After determining the second replenishment quantity corresponding to each historical feature vector, a training set for the model can be formed, and the preset model can be trained.

[0059] Figure 3 A schematic diagram of the MQRNN network structure is shown below. The following will combine... Figure 3 The training process of the preset model is described, that is, in this embodiment, the MQRNN network model is selected as the preset model, and the MQRNN network model is used to form a prediction model for predicting replenishment quantity. Figure 3 As shown, the training set obtained in step S22 can be split into two parts, one of which serves as the input (x) to the encoder. i y i The other part serves as the input to the decoder section. Where i = 1, ..., t, j = t+1, ..., N, N is the number of training data in the training set, and t has no specific meaning. It is just a parameter used to split the input data according to the actual needs of the MQRNN network model during training. Its actual value can be adjusted according to the network model. In this embodiment, based on the distribution of the training data, the value of t is preferably set to 60% of the total number of training data N, that is, t = 0.6N. At this time, splitting the input data based on t can make the model training results more consistent with the relationship between the actual data, so as to obtain more accurate prediction results.

[0060] (x) i y i After being used as input to the encoder part, the parameter h can be obtained through LSTM encoding. i and h i The input is fed into the decoder section; simultaneously, As input to the Decoder part, combined with h i The parameter c is obtained through Global MLP (Multilayer Perceptron). j and c a Then, the quantile prediction values ​​are obtained through a local MLP. Where q is the quantile, its value is usually between 0 and 1. In this embodiment, q is 0.5, which means that the predicted value is larger than the actual value by a ratio of q, and the predicted value is smaller than the actual value by a ratio of 1-q. The value of q can be adjusted during actual training; finally, based on the output replenishment quantity y j The second replenishment quantity y in the training set mThe values ​​are compared, and the values ​​of each parameter in the MQRNN network model are adjusted according to the comparison results until the MQRNN network model reaches the convergence condition.

[0061] In some embodiments, the convergence condition of the model can be that the convergence condition is met after the number of iterations reaches a certain threshold, or when the model's loss function is minimized, or when the weight change between two iterations is less than a certain threshold, or other convergence conditions set according to actual needs. In this embodiment, the loss function is used as the convergence condition of the model. When the loss function is minimized, the model is considered to meet the convergence condition, and a mapping relationship between the feature vector and the replenishment quantity is formed as shown in formula (3) to characterize the trained prediction model:

[0062]

[0063] Where, x i Let N represent the feature vector of the i-th replenishment decision day, and let N represent the number of training data in the training set. Let L represent the replenishment quantity on the i-th replenishment decision day, and let L represent the loss function of the prediction model. In this embodiment, squared loss can be used, i.e.

[0064] It should be noted that this embodiment uses the MQRNN network model as the preset model and describes the training process of the MQRNN network model. In actual use, if other network models are selected as the preset model, training can be performed according to the characteristics of the selected network model. This embodiment does not impose any specific restrictions.

[0065] Once the prediction model is trained, when the actual replenishment quantity needs to be predicted on the replenishment decision day, the inventory situation on the replenishment decision day can be formed into the first feature vector and input into the prediction model. The value output by the prediction model can be directly used as the replenishment quantity corresponding to the replenishment decision day, realizing end-to-end replenishment quantity prediction and achieving the lowest inventory cost while ensuring service level.

[0066] A second embodiment of this disclosure provides a supply chain replenishment quantity prediction device, which can be installed in an electronic device used for making supply chain replenishment decisions, such as a computer, mobile phone, or tablet computer. Figure 4 A schematic diagram of a replenishment quantity prediction device is shown, which mainly includes an acquisition module 10 and a prediction module 20 coupled to each other. The acquisition module 10 is used to construct a first feature vector based on the inventory situation on the replenishment decision day. The prediction module 20 is used to input the first feature vector into the prediction model and determine that the output value of the prediction model is the first replenishment quantity corresponding to the current replenishment decision day. The prediction model is used to characterize the correlation between the feature vector and the replenishment quantity.

[0067] In some embodiments, the replenishment quantity prediction device may further include a training module 30 (not shown in the figure) coupled to the prediction module 20. The training module 30 is specifically used to construct historical feature vectors corresponding to m replenishment decision days based on the historical inventory situation of m replenishment decision days; determine the second replenishment quantity corresponding to each historical feature vector based on each historical feature vector so as to minimize the overall cost between two adjacent replenishment decision days; and train the preset model based on the training set formed by all historical feature vectors and the second replenishment quantities corresponding to all historical feature vectors until the loss function of the preset model is minimized, thereby obtaining the prediction model.

[0068] In some embodiments, the total cost between two adjacent replenishment decision days is determined based on the following formula: in, The total cost is represented between the m-th replenishment decision day and the (m+1)-th replenishment decision day, where h is the inventory cost per unit, b is the stockout cost per unit, and v is the total cost. m Let m be the arrival time of the m-th replenishment. Let m be the current inventory level at the time of the m-th replenishment. For from v m The total inventory demand over a period of time s, where s represents any date between the m-th replenishment decision date and the (m+1)-th replenishment decision date, and y m The second replenishment quantity is determined on the m-th replenishment decision day; the overall cost is minimized when determining... in,

[0069] In some embodiments, the prediction model is based on the following formula: Where, x i Let represent the feature vector of the i-th replenishment decision day, N represent the number of training data in the training set, and L represent the loss function of the prediction model. This represents the replenishment quantity on the i-th replenishment decision day.

[0070] In some embodiments, training a preset model based on a training set formed by all historical feature vectors and the second replenishment quantity corresponding to all historical feature vectors includes: splitting the training data in the training set into two parts, one part of which is used as the input (x) to the encoder of the preset model. i y i The other part serves as the input to the decoder of the preset model. Where i = 1, ..., t, j = t+1, ..., N, t is used to characterize the splitting parameters of the training data, t = N * 60%.

[0071] In some embodiments, the preset model is any one of the following models: MQRNN network model, LightGBM network model, XGBoost network model, and TCN network model.

[0072] In some embodiments, inventory status includes at least one or more of the following: the date of the replenishment decision, the replenishment duration, the product category, the inventory of different product categories on the replenishment decision date, and the daily demand for each product category between the previous replenishment decision date and the current replenishment decision date.

[0073] This embodiment trains a preset model to obtain a prediction model that represents the relationship between feature vectors and replenishment quantity. The prediction model is used to predict the replenishment quantity end-to-end based on the inventory status on the current replenishment decision day. The predicted replenishment quantity is used to assist supply chain inventory management decisions, achieving the lowest inventory cost while ensuring service levels.

[0074] The third embodiment of this disclosure provides a storage medium that can be installed in any electronic device, system, or platform with supply chain replenishment quantity decision management functions. Specifically, it is a computer-readable medium storing a computer program. When the computer program is executed by a processor, it implements the method provided in any embodiment of this disclosure, including the following steps S31 and S32:

[0075] S31, construct the first feature vector based on the inventory situation on the replenishment decision day;

[0076] S32, input the first feature vector into the prediction model, and determine that the value output by the prediction model is the first replenishment quantity corresponding to the current replenishment decision day, wherein the prediction model is used to characterize the relationship between the feature vector and the replenishment quantity.

[0077] The computer program is further processed by the processor to train the prediction model by: constructing historical feature vectors corresponding to the m replenishment decision days based on the historical inventory data of the m replenishment decision days; determining the second replenishment quantity corresponding to each historical feature vector to minimize the overall cost between two adjacent replenishment decision days; and training the preset model based on the training set formed by all historical feature vectors and the second replenishment quantities corresponding to all historical feature vectors until the loss function of the preset model is minimized, thus obtaining the prediction model.

[0078] Specifically, the total cost between two adjacent replenishment decision days is determined based on the following formula:

[0079]

[0080] in, The total cost is represented between the m-th replenishment decision day and the (m+1)-th replenishment decision day, where h is the inventory cost per unit, b is the stockout cost per unit, and v is the total cost. m Let m be the arrival time of the m-th replenishment. Let m be the current inventory level at the time of the m-th replenishment. For from v m The total inventory demand over a period of time s, where s represents any date between the m-th replenishment decision date and the (m+1)-th replenishment decision date, and y m The second replenishment quantity is determined for the m-th replenishment decision day; the computer program is also processed by the processor to perform the following steps: determine the quantity when the total cost is minimized. in,

[0081] Specifically, the prediction model is based on the following formula:

[0082]

[0083] Where, x i Let represent the feature vector of the i-th replenishment decision day, N represent the number of training data in the training set, and L represent the loss function of the prediction model. This represents the replenishment quantity on the i-th replenishment decision day.

[0084] When a computer program is executed by a processor to train a pre-defined model using a training set consisting of all historical feature vectors and the corresponding second replenishment quantities, the processor specifically performs the following steps: The training data in the training set is split into two parts, one of which is used as the input (x) to the encoder of the pre-defined model. i y i The other part serves as the input to the decoder of the preset model. Where i = 1, ..., t, j = t+1, ..., N, t is used to characterize the splitting parameters of the training data, t = N * 60%.

[0085] Specifically, the preset model is any one of the following models: MQRNN network model, LightGBM network model, XGBoost network model, or TCN network model.

[0086] Specifically, inventory information includes at least one or more of the following: the date of the replenishment decision, the replenishment duration, the product category, the inventory of different product categories on the replenishment decision date, and the daily demand for each product category between the previous replenishment decision date and the current replenishment decision date.

[0087] This embodiment trains a preset model to obtain a prediction model that represents the relationship between feature vectors and replenishment quantity. The prediction model is used to predict the replenishment quantity end-to-end based on the inventory status on the current replenishment decision day. The predicted replenishment quantity is used to assist supply chain inventory management decisions, achieving the lowest inventory cost while ensuring service levels.

[0088] The fourth embodiment of this disclosure provides an electronic device, which can be a computer, server system, or platform with supply chain replenishment quantity decision management functions, as shown in the schematic diagram below. Figure 5 As shown, the system includes at least a memory 100 and a processor 200. The memory 100 stores a computer program, and the processor 200 implements the methods provided in any embodiment of this disclosure when executing the computer program in the memory 100. Exemplarily, the steps of the electronic device computer program are as follows: S41 and S42:

[0089] S41, construct the first feature vector based on the inventory situation on the replenishment decision day;

[0090] S42, input the first feature vector into the prediction model, and determine that the value output by the prediction model is the first replenishment quantity corresponding to the current replenishment decision day, wherein the prediction model is used to characterize the relationship between the feature vector and the replenishment quantity.

[0091] The processor also executes the following computer program stored in memory to train the prediction model: constructing historical feature vectors corresponding to m replenishment decision days based on the historical inventory situation of m replenishment decision days; determining the second replenishment quantity corresponding to each historical feature vector to minimize the overall cost between two adjacent replenishment decision days; training the preset model based on the training set formed by all historical feature vectors and the second replenishment quantities corresponding to all historical feature vectors until the loss function of the preset model is minimized, thus obtaining the prediction model.

[0092] Specifically, the total cost between two adjacent replenishment decision days is determined based on the following formula:

[0093]

[0094] in, The total cost is represented between the m-th replenishment decision day and the (m+1)-th replenishment decision day, where h is the inventory cost per unit, b is the stockout cost per unit, and v is the total cost. m Let m be the arrival time of the m-th replenishment. Let m be the current inventory level at the time of the m-th replenishment. For from v m The total inventory demand over a period of time s, where s represents any date between the m-th replenishment decision date and the (m+1)-th replenishment decision date, and ym The processor determines the second replenishment quantity for the m-th replenishment decision day; the processor also executes the following computer program stored in memory: to determine the optimal replenishment quantity when the total cost is minimized. in,

[0095] Specifically, the prediction model is based on the following formula:

[0096]

[0097] Where, x i Let represent the feature vector of the i-th replenishment decision day, N represent the number of training data in the training set, and L represent the loss function of the prediction model. This represents the replenishment quantity on the i-th replenishment decision day.

[0098] When the processor trains the preset model using the training set stored in execution memory, which is formed based on all historical feature vectors and the second replenishment quantity corresponding to all historical feature vectors, it specifically executes the following computer program: It splits the training data in the training set into two parts, one of which is used as the input (x) to the encoder of the preset model. i y i The other part serves as the input to the decoder of the preset model. Where i = 1, ..., t, j = t+1, ..., N, t is used to characterize the splitting parameters of the training data, t = N * 60%.

[0099] Specifically, the preset model is any one of the following models: MQRNN network model, LightGBM network model, XGBoost network model, or TCN network model.

[0100] Specifically, inventory information includes at least one or more of the following: the date of the replenishment decision, the replenishment duration, the product category, the inventory of different product categories on the replenishment decision date, and the daily demand for each product category between the previous replenishment decision date and the current replenishment decision date.

[0101] This embodiment trains a preset model to obtain a prediction model that represents the relationship between feature vectors and replenishment quantity. The prediction model is used to predict the replenishment quantity end-to-end based on the inventory status on the current replenishment decision day. The predicted replenishment quantity is used to assist supply chain inventory management decisions, achieving the lowest inventory cost while ensuring service levels.

[0102] The foregoing has provided a detailed description of several embodiments of this disclosure. However, this disclosure is not limited to these specific embodiments. Those skilled in the art can make various variations and modifications based on the concept of this disclosure, and all such variations and modifications should fall within the scope of protection claimed by this disclosure.

Claims

1. A method of predicting a replenishment quantity of a supply chain, characterized by, The application relates to a method for determining a replenishment quantity, and a device and a computer program product for determining a replenishment quantity. The first feature vector is constructed according to inventory conditions on a replenishment decision day; The first feature vector is input into a prediction model, and a value output by the prediction model is determined as a first replenishment quantity corresponding to a current replenishment decision day, wherein the prediction model is used to represent an association between a feature vector and a replenishment quantity; the prediction model is trained based on the following steps: m historical feature vectors corresponding to m replenishment decision days are respectively constructed according to historical inventory conditions of the m replenishment decision days; the inventory conditions at least include the following: a date of a replenishment decision day, a replenishment duration, a product category, an inventory quantity of different categories of products on the replenishment decision day, and a demand quantity of each product category in each day between a last replenishment decision day and the current replenishment decision day; a second replenishment quantity corresponding to each historical feature vector is determined according to each historical feature vector, so that a total cost between two adjacent replenishment decision days is minimized; the second replenishment quantity is a replenishment quantity that minimizes a total inventory cost in a replenishment cycle corresponding to the replenishment decision day; a preset model is trained according to a training set formed by all the historical feature vectors and the second replenishment quantities corresponding to all the historical feature vectors, until a loss function of the preset model is minimized, so that the prediction model is obtained; in the process of training the prediction model, input data of a training set used by the prediction model during model training should be the historical feature vector of each replenishment decision day and the second replenishment quantity corresponding to each historical feature vector; the total cost between two adjacent replenishment decision days is determined based on the following formula: wherein, characterizing the overall cost between the mthreplenishment decision day and the m+1threplenishment decision day, h is the inventory cost per unit product, b is the stockout cost per unit product, is the arrival time of the mthreplenishment, is the current inventory level at the mthreplenishment, is the total inventory demand from to s, s is used to characterize any date between the mthreplenishment decision day and the m+1threplenishment decision day, is the second replenishment quantity at the mthreplenishment decision day; determining the total cost at the time of the total cost is the lowest where s = 0 .

2. The replenishment quantity prediction method according to claim 1, characterized by, The prediction model is represented based on the following formula: The method comprises the following steps: Xi represents a feature vector of the i-th replenishment decision day, N represents the number of training data in the training set, and L represents a loss function of the prediction model. Qi represents a replenishment quantity of the i-th replenishment decision day.

3. The method of claim 2, wherein The preset model is trained according to a training set formed by all the historical feature vectors and the second replenishment quantities corresponding to all the historical feature vectors, including: splitting training data in a training set into two parts, one part as input of an encoder of the preset model , and the other part as input of a decoder of the preset model , wherein i = 1, … t, j = t + 1, … N, t is used to represent a split parameter of the training data, t = N * 60%.

4. The method of claim 1, wherein The preset model is any one of the following models: an MQRNN network model, a LightGBM network model, an XGBoost network model, and a TCN network model.

5. A replenishment quantity prediction device of a supply chain, characterized by, The application relates to a method for determining a replenishment quantity, and a device and a computer program product for determining a replenishment quantity. The application relates to a method for determining a replenishment quantity, and a device and a computer program product for determining a replenishment quantity. The prediction module is configured to input the first feature vector into a prediction model, and determine a value output by the prediction model as a first replenishment quantity corresponding to a current replenishment decision day, wherein the prediction model is used to represent an association between a feature vector and a replenishment quantity; and the training module is specifically configured to: m historical feature vectors corresponding to m replenishment decision days are respectively constructed according to historical inventory conditions of the m replenishment decision days; the inventory conditions at least include the following: a date of a replenishment decision day, a replenishment duration, a product category, an inventory quantity of different categories of products on the replenishment decision day, and a demand quantity of each product category in each day between a last replenishment decision day and the current replenishment decision day; determining a second replenishment quantity corresponding to each of the historical feature vectors, so as to minimize the overall cost between two adjacent replenishment decision days; the second replenishment quantity being a replenishment quantity that minimizes the overall inventory cost in a replenishment cycle corresponding to the replenishment decision day; the overall cost between two adjacent replenishment decision days being determined based on the following formula: wherein, characterizing the overall cost between the mthreplenishment decision day and the m+1threplenishment decision day, h is the inventory cost per unit product, b is the stockout cost per unit product, is the arrival time of the mthreplenishment, is the current inventory level at the mthreplenishment, is the inventory demand sum from to s, s is used to characterize any date between the mthreplenishment decision day and the m+1threplenishment decision day, is the second replenishment quantity at the mthreplenishment decision day; determining the total cost at the time of the total cost is the lowest where s = 0 ; training a preset model based on a training set formed by all the historical feature vectors and the second replenishment quantities corresponding to the historical feature vectors, until a loss function of the preset model is minimized, to obtain the prediction model; in the process of training, the input data of the training set used in the model training should be the historical feature vector of each replenishment decision day and the second replenishment quantity corresponding to each historical feature vector.

6. A storage medium storing a computer program, characterized by The computer program, when executed by a processor, implements the steps of the replenishment quantity prediction method of the supply chain according to any one of claims 1 to 5.

7. An electronic device comprising at least a memory, a processor, said memory having stored thereon a computer program, characterized in that, The processor, when executing the computer program on the memory, implements the steps of the replenishment quantity prediction method of the supply chain according to any one of claims 1 to 5.

Images