CN117541149A

CN117541149A - Intelligent supply chain replenishment method, system, electronic equipment and storage medium

Info

Publication number: CN117541149A
Application number: CN202210891453.2A
Authority: CN
Inventors: 李嘉稷; 赵申宜; 秦川; 蔡恒兴; 邹昊晟; 涂威威
Original assignee: 4Paradigm Beijing Technology Co Ltd
Current assignee: 4Paradigm Beijing Technology Co Ltd
Priority date: 2022-07-27
Filing date: 2022-07-27
Publication date: 2024-02-09

Abstract

The disclosure relates to an intelligent supply chain goods supplementing method, system, electronic equipment and storage medium, wherein the method comprises the following steps: obtaining a plurality of parameters related to the current replenishment quantity, wherein the estimated arrival time of the current replenishment, the estimated arrival time of the next replenishment closest to the estimated arrival time of the current replenishment and the estimated arrival time of the replenishment quantity estimated to arrive between the current arrival time of the current replenishment are taken as a group of estimated arrival times, at least one group of estimated arrival times is obtained through a dynamic simulator, and under each group of estimated arrival times, the current replenishment quantity is respectively predicted through a pre-trained replenishment model based on the plurality of parameters to obtain at least one predicted current replenishment quantity, wherein the replenishment model is trained through the dynamic simulator based on different arrival time sequences; and determining a first estimated replenishment quantity according to at least one predicted current replenishment quantity, and determining a current replenishment decision based on the first estimated replenishment quantity.

Description

Intelligent supply chain replenishment method, system, electronic equipment and storage medium

Technical Field

The present disclosure relates to the field of artificial intelligence, and more particularly, to an intelligent supply chain replenishment method, system, electronic device, and storage medium.

Background

One of the goals of supply chain management is to minimize the number of inventory turnover days while maintaining a certain spot rate, that is, to avoid both backorders and backlogs, so how to restock is a critical issue for stores.

Various replenishment strategies exist in the related art, for example, a conventional rule-based replenishment method mainly includes an RQ method mainly consisting in replenishing the inventory of quality Q (quantity Q) when the inventory is smaller than reorder-point R (order point R), and an sS method mainly consisting in replenishing to a specific inventory S when the inventory is smaller than S. In addition, the goods demand in a future period can be predicted, the date and the quantity of the occurrence of the shortage can be judged in advance by combining the transportation information of each store, and the quantity of the replenishment can be determined by combining the predicted goods demand with an operation study method.

Although the replenishment strategy in the related art is proposed based on a certain theoretical basis, due to the complex and changeable actual production environment, the phenomenon of backlog or backlog of stock still occurs, which affects the operation cost and income of the supply chain.

Disclosure of Invention

The present disclosure provides an intelligent supply chain replenishment method, an intelligent supply chain replenishment device, an electronic device and a storage medium, so as to at least solve the above-mentioned problems in the related art.

According to a first aspect of embodiments of the present disclosure, there is provided an intelligent supply chain restocking method, comprising: obtaining a plurality of parameters related to the current replenishment quantity, wherein the parameters comprise the actual sales quantity of a product in a first historical preset time period, the predicted sales quantity of the product in a first future preset time period, the predicted arrival time of the current replenishment, the predicted arrival time of the next replenishment closest to the predicted arrival time of the current replenishment, the current spot situation, and the predicted arrival quantity of the product between the current arrival time and the predicted arrival time of the current replenishment, and the predicted arrival time of the next replenishment closest to the predicted arrival time of the current replenishment and the predicted arrival time of the current replenishment are taken as a group of predicted arrival times, and at least one group of the predicted arrival times is obtained through a dynamic simulator, and the dynamic simulator is used for simulating the inventory condition of the product in a period after the current time; under each group of expected arrival time, predicting the current replenishment quantity through a pre-trained replenishment model based on the plurality of parameters to obtain at least one predicted current replenishment quantity, wherein the replenishment model is trained through the dynamic simulator based on different arrival time sequences, the arrival time sequences comprise at least one arrival time, and the arrival time represents the time required from delivery to arrival; determining a first estimated replenishment quantity according to the at least one predicted current replenishment quantity; and determining the current replenishment decision based on the first estimated replenishment quantity.

Optionally, the determining the current replenishment decision based on the first estimated replenishment quantity includes: determining replenishment coefficients according to the current inventory health condition of the product and an evaluation result aiming at a historical replenishment decision, wherein the current inventory health condition of the product is obtained by simulating the inventory condition of the product for a period of time after the current time under each group of expected arrival times respectively, and the historical replenishment decision is at least one replenishment decision determined for a period of time before the current time; determining a second estimated replenishment quantity according to the replenishment coefficient and the estimated backorder condition in a period of time after the current time; and determining the current replenishment decision based on the first estimated replenishment quantity and/or the second estimated replenishment quantity.

Optionally, the determining the current replenishment decision based on the first estimated replenishment volume and/or the second estimated replenishment volume includes: and determining the first estimated replenishment quantity as a current replenishment decision, or determining the second estimated replenishment quantity as a current replenishment decision, or determining a weighted average of the first estimated replenishment quantity and the second estimated replenishment quantity as a current replenishment decision, or determining the current replenishment decision according to the first estimated replenishment quantity, the second estimated replenishment quantity and a buffer quantity for resisting an abnormal point, wherein the abnormal point resisting means that in the process of determining the current replenishment decision, the abnormal fluctuation meeting an abnormal judgment standard is detected in the real sales quantity of the product in a second historical preset time period, and measures are taken to cope with the abnormal fluctuation possibly appearing again.

Optionally, the buffer amount is determined by: under the condition that the current replenishment decision is a first replenishment decision for the product, determining the buffer quantity according to the real sales quantity of the product in the second historical preset time period and the historical abnormal sales quantity statistic value of the product; and under the condition that the current replenishment decision is not the first replenishment decision for the product, determining the buffer quantity according to the real sales volume of the product in the second historical preset time period, the historical abnormal sales volume statistic value of the product and the execution effect of resisting the abnormal point in the second historical preset time period.

Optionally, the restocking model is pre-trained by: acquiring a training data set; training the restocking model by using a preset algorithm based on a training data set to obtain the trained restocking model, wherein an optimization objective of the preset algorithm is: the comprehensive value of the spot rate and the inventory turnover number of the product in the first preset time length is minimum; the stock rate and the inventory turnover number are obtained by deducting the stock quantity of the product in the first preset time by the dynamic simulator according to the predicted sales quantity and the real stock quantity of the product in the first preset time and the stock quantity predicted by the stock supplement model based on different arrival time sequences, and the first preset time is obtained by the real arrival time of the product.

Optionally, the predicted sales in the first future preset time period is predicted by at least one sales prediction model trained in advance, wherein each sales prediction model is different in type.

Optionally, the predicted sales in the first future preset time period is predicted by one or more sales prediction models of the at least one sales prediction model according to sales characteristics of the product.

Optionally, retraining the restocking model and the at least one sales prediction model based on real sales data and real inventory data generated within the preset time interval at preset time intervals to obtain a new trained restocking model and the at least one sales prediction model.

According to a second aspect of embodiments of the present disclosure, there is provided an intelligent supply chain restocking system comprising: parameter acquisition means configured to: obtaining a plurality of parameters related to the current replenishment quantity, wherein the parameters comprise the actual sales quantity of a product in a first historical preset time period, the predicted sales quantity of the product in a first future preset time period, the predicted arrival time of the current replenishment, the predicted arrival time of the next replenishment closest to the predicted arrival time of the current replenishment, the current spot situation, and the predicted arrival quantity of the product between the current arrival time and the predicted arrival time of the current replenishment, and the predicted arrival time of the next replenishment closest to the predicted arrival time of the current replenishment and the predicted arrival time of the current replenishment are taken as a group of predicted arrival times, and at least one group of the predicted arrival times is obtained through a dynamic simulator, and the dynamic simulator is used for simulating the inventory condition of the product in a period after the current time; a replenishment quantity predicting device configured to: under each group of expected arrival time, predicting the current replenishment quantity through a pre-trained replenishment model based on the plurality of parameters to obtain at least one predicted current replenishment quantity, wherein the replenishment model is trained through the dynamic simulator based on different arrival time sequences, the arrival time sequences comprise at least one arrival time, and the arrival time represents the time required from delivery to arrival; the first estimated replenishment quantity determining device is configured to: determining a first estimated replenishment quantity according to the at least one predicted current replenishment quantity; a restocking decision-making device configured to: and determining the current replenishment decision based on the first estimated replenishment quantity.

Optionally, the replenishment decision determining device may be configured to determine a replenishment coefficient according to a current stock health status of the product and an evaluation result for a historical replenishment decision, the current stock health status of the product being obtained by simulating, by the dynamic simulator, the stock status of the product for a period of time after the current time, respectively, at each set of the expected arrival times, the historical replenishment decision being at least one replenishment decision determined for a period of time before the current time; determining a second estimated replenishment quantity according to the replenishment coefficient and the estimated backorder condition in a period of time after the current time; and determining the current replenishment decision based on the first estimated replenishment quantity and/or the second estimated replenishment quantity.

Optionally, the replenishment decision determining device may be configured to determine the first estimated replenishment amount as a current replenishment decision, or determine the second estimated replenishment amount as a current replenishment decision, or determine a weighted average of the first estimated replenishment amount and the second estimated replenishment amount as a current replenishment decision, or determine the current replenishment decision according to the first estimated replenishment amount, the second estimated replenishment amount, and a buffer amount for performing an abnormal point resistance, where the abnormal point resistance is that, in a process of determining the current replenishment decision, an abnormal fluctuation that meets an abnormal judgment standard is detected in a real sales amount of the product in a second historical preset time period, and measures are taken to cope with the abnormal fluctuation that may occur again.

Optionally, the intelligent supply chain restocking system further comprises a model retraining device, which may be configured to retrain the restock model and the at least one sales prediction model based on real sales data and real inventory data generated within a preset time interval at intervals to obtain the new trained restock model and the at least one sales prediction model.

According to a third aspect of embodiments of the present disclosure, there is provided an electronic device, comprising: at least one processor; at least one memory storing computer-executable instructions, wherein the computer-executable instructions, when executed by the at least one processor, cause the at least one processor to perform an intelligent supply chain restocking method according to the present disclosure.

According to a fourth aspect of embodiments of the present disclosure, there is provided a computer-readable storage medium storing instructions that, when executed by at least one processor, cause the at least one processor to perform an intelligent supply chain restocking method according to the present disclosure.

The technical scheme provided by the embodiment of the disclosure at least brings the following beneficial effects:

according to the intelligent supply chain replenishment method, the intelligent supply chain replenishment device, the electronic equipment and the storage medium, a plurality of groups of expected arrival times are obtained through the dynamic simulator, the current replenishment quantity is respectively predicted through the pre-trained replenishment model under each group of expected arrival times to obtain the estimated replenishment quantity, and the replenishment model is also obtained through training through the dynamic simulator on the basis of a plurality of leader time (time required for the arrival of a product from delivery) sequences due to the fact that uncertainty of the arrival time of the product is considered, so that the estimated replenishment quantity obtained through the replenishment model has high robustness in an actual production environment, and the replenishment decision made based on the estimated replenishment quantity meets the actual demand in the actual production environment.

In addition, sales of the product in a future period of time is respectively predicted through a plurality of sales prediction models of different types, the future sales of the product is determined by combining the prediction results of the sales prediction models, the accuracy of sales prediction can be improved, and then, the replenishment decision made based on the predicted sales better meets the real requirements in the actual production environment.

In addition, the replenishment coefficient is determined according to the current stock health condition of the product and the evaluation result of the replenishment decision in a historical period of time, and the replenishment decision is guided to be determined by the replenishment coefficient, so that the replenishment decision can be further more in line with the real requirement in the actual production environment.

In addition, the abnormal point detection is carried out on the real sales volume within a period of time before the current time, the detection result is used as auxiliary information to guide the determination of the replenishment decision, and the replenishment decision can be further made to more meet the real requirements in the actual production environment.

In summary, since the replenishment decisions more meet the real demands in the actual production environment, the occurrence frequency of backlog or backlog of stock can be reduced, and the number of inventory turnover days can be made smaller under the condition of meeting a certain spot rate, so that the supply chain operation cost can be minimized and the income can be maximized in a long period of time.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description, serve to explain the principles of the disclosure and do not constitute an undue limitation on the disclosure.

FIG. 1 is a flowchart illustrating an intelligent supply chain restocking method according to an example embodiment of the present disclosure.

FIG. 2 is a frame diagram illustrating a sales prediction model according to an example embodiment of the present disclosure.

Fig. 3 is an evaluation diagram showing the prediction effect of each sales prediction model shown in fig. 2 in different combinations according to an exemplary embodiment of the present disclosure.

Fig. 4 is a transition diagram illustrating an outlier rejection state according to an exemplary embodiment of the present disclosure.

Fig. 5 is an overall frame diagram illustrating an intelligent supply chain restocking method according to an exemplary embodiment of the present disclosure.

FIG. 6 is a block diagram illustrating an intelligent supply chain restocking system according to an example embodiment of the present disclosure.

Fig. 7 is a block diagram illustrating an electronic device 700 according to an exemplary embodiment of the present disclosure.

Detailed Description

In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

It should be noted that the terms "first," "second," and the like in the description and claims of the present disclosure and in the foregoing figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the disclosure described herein may be capable of operation in sequences other than those illustrated or described herein. The embodiments described in the examples below are not representative of all embodiments consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present disclosure as detailed in the accompanying claims.

It should be noted that, in this disclosure, "at least one of the items" refers to a case where three types of juxtaposition including "any one of the items", "a combination of any of the items", "an entirety of the items" are included. For example, "including at least one of a and B" includes three cases side by side as follows: (1) comprises A; (2) comprising B; (3) includes A and B. For example, "at least one of the first and second steps is executed", that is, three cases are juxtaposed as follows: (1) performing step one; (2) executing the second step; (3) executing the first step and the second step.

FIG. 1 is a flowchart illustrating an intelligent supply chain restocking method according to an example embodiment of the present disclosure. Here, the method may be performed by a computer program, or by a dedicated hardware device or an aggregate of hardware and software resources for implementing intelligent supply chain restocking, as an example.

In one embodiment, the method is used to assist a retail store in determining restocking decisions. Specifically, retail stores sell a plurality of products, each of which may be referred to as a sku (stock keeping unit, stock level unit, and is referred to as a product unit number, with a sku corresponding to a unique one of the products), inventory of each of the products is checked at regular intervals (e.g., daily or weekly, etc.) by the sales store, and the amount of restocking of each of the products is determined for a subsequent period of time using the intelligent supply chain restocking method of the disclosed embodiments, wherein the determined amount of restocking is required to meet sales needs for a week, month, etc., each time a restocking decision is made. In other embodiments, the method is also used to assist the chain merchant in determining restocking decisions for each store, which is not limiting of the present disclosure. In the following description, the intelligent supply chain restocking method of the present disclosure is described in terms of retail stores as an implementation scenario.

Referring to fig. 1, in step 101, a plurality of parameters related to the current replenishment amount may be obtained, where the plurality of parameters include a real sales amount of the product in a first historical preset time period, a predicted sales amount of the product in a first future preset time period, a predicted arrival time of the current replenishment, a predicted arrival time of a next replenishment closest to the predicted arrival time of the current replenishment, a current spot situation, and an expected arrival amount of the product between the current arrival times of the current replenishment, and wherein the predicted arrival time of the current replenishment, the predicted arrival time of the next replenishment closest to the predicted arrival time of the current replenishment, and the predicted arrival time of the expected arrival amount of the product between the current arrival times of the current replenishment are taken as a set of predicted arrival times, and at least one set of the predicted arrival times is obtained by a dynamic simulator for simulating an inventory condition of the product within a period after the current time.

Here, the length of the first history preset period is the same as the length of the first future preset period, which may be 28 days, 35 days, etc. before the current time, and the first future preset period may be 28 days, 35 days, etc. after the current time, which is not limited by the present disclosure. The estimated time of arrival of the present restocking is the time when a restocking decision is currently made (e.g., the day) and the corresponding restocking volume arrives at the retail store in the future. The estimated time of the next replenishment closest to the estimated time of the present replenishment refers to a replenishment decision made at a time after the present time, the corresponding replenishment amount of which arrives after the estimated arrival time of the replenishment amount corresponding to the present replenishment decision, and the estimated arrival time of which is closest to the estimated arrival time of the present replenishment amount (in fact, the demand to be satisfied by the replenishment amount corresponding to the present replenishment decision is the demand between the present arrival time and the next replenishment arrival time). The current spot condition refers to whether the current stock quantity can meet the current demand (e.g., may be the difference between the stock quantity on the day of the replenishment decision and the sales quantity on the day). The estimated arrival of the replenishment quantity between the current arrival time and the estimated arrival time of the current replenishment refers to the replenishment quantity corresponding to the replenishment decision made at a time before the current time, and arrives between the current time and the estimated arrival time of the replenishment quantity corresponding to the current replenishment decision, that is, the intelligent supply chain replenishment method of the present disclosure considers the influence of the continuous multiple replenishment decisions on the current replenishment decision.

In a specific implementation, each estimated time of each set of estimated time of arrival is determined by a shippable time of the supplier and an arrival time of the product (time required for arrival from shipment), the dynamic simulator may determine a plurality of arrival time sequences by a plurality of arrival time of the product in the past and comprehensively considering transportation capacity (planned route, transportation speed, etc.) of the transportation means, weather conditions of a future period, traffic conditions, social events that may affect the arrival time, etc., wherein each arrival time sequence is composed of arrival time of a replenishment amount estimated to arrive between the current arrival time of the replenishment, arrival time of the current replenishment, and arrival time of the next replenishment closest to the estimated arrival time of the current replenishment, and the dynamic simulator may determine a plurality of sets of estimated arrival times in combination with the shippable time of the respective replenishment amounts under each arrival time sequence. For example, assuming that the current time is 1 day of 4 months of 2022, the arrival time of the replenishment amount corresponding to the replenishment decision made on the 3 months 31 days before the day is 3 days, the arrival time of the replenishment amount corresponding to the replenishment decision made on the 4 months 1 days is 4 days, the arrival time of the replenishment amount corresponding to the replenishment decision made on the 4 months 2 days after the day is 5 days, and the replenishment amount corresponding to each replenishment decision made may be issued by the supplier on the following day of the decision day, the simulator determines the sequence of replenishment time periods as [3,4,5] with a corresponding set of estimated time to replenishment of 4 months and 4 days (i.e., the estimated time to replenishment of the amount of replenishment expected to arrive between the current and current estimated time to replenishment), 4 months and 6 days (i.e., the estimated time to replenishment of the current replenishment) and 4 months and 8 days (i.e., the estimated time to replenishment of the next replenishment closest to the current estimated time of replenishment).

In one embodiment, the predicted sales in the first future preset time period are predicted by at least one sales prediction model trained in advance, wherein each sales prediction model is of a different type. Here, the sales prediction model may be, for example, but not limited to, a linear model such as a tree model (lightgbm), a deep learning model (DNN), ARMA (Auto-Regressive and Moving Average Model, a hybrid model based on an autoregressive model and a moving average model)/ARIMA (Auto-Regressive Integrated Moving Average Model, an autoregressive moving average model), and some simple prediction models (for example, the predicted sales is determined directly from the average sales of a historic period, or the actual sales of a historic period is taken as the predicted sales directly, etc.), which is not limited. The prediction results (e.g., additive averaging or weighted averaging, etc.) of at least one sales prediction model may be combined to determine a final predicted sales.

Referring to fig. 2, sales prediction models of exemplary embodiments of the present disclosure include a tree model, a deep learning model, and a simple prediction model, it being understood that more types of sales prediction models may be provided according to actual conditions, which is not limited by the present disclosure. Depending on the structural features of each model, the selected input features include raw features of the product, such as, but not limited to, geographic information, geographic hierarchy, geographic aggregation dimension, product information, product hierarchy, product aggregation dimension, temporal features, such as, but not limited to, including year, quarter, month, week, and weekend, etc., sliding window statistics refer to statistics of feature data of the product at different window sizes, where within each window, the data may be counted as mean, variance, median, maximum, minimum, etc. The Lag feature refers to another time series related to the feature obtained by taking a time after a delay on the previous time series related to the feature. When the predicted sales are obtained, the corresponding features are input into corresponding sales prediction models, and finally the prediction results of the models are weighted and fused, so that the predicted sales in a future preset time period can be obtained. Here, as an example, sales for 1-21 days in the future are predicted.

Referring to fig. 3, a combination of a simple prediction model, a tree model, and a deep learning model has the best prediction effect in an offline experiment, wherein lgb represents a lightgbm tree model, DNN represents a deep learning model, conv1d represents one-dimensional hole convolution, wmae represents an evaluation index of the model.

In one embodiment, the predicted sales in the first future preset time period is predicted by one or more of the at least one sales prediction model based on sales characteristics of the product. Here, the sales characteristic may be, for example, but not limited to, "high sales/low sales", "high return rate/low return rate", etc., and the sales characteristic of the product may be obtained by clustering different products, or manually analyzing historical sales data of the product, etc. As an example, one sales prediction model may be employed to obtain a predicted sales for low sales products, while multiple sales prediction models may be employed to obtain a predicted sales for high sales products.

According to the predicted sales volume obtaining scheme, the features considered by different sales volume prediction models are different, so that the accuracy of the predicted sales volume obtained by combining the prediction results of the sales volume prediction models is higher. And because the predicted sales volume has a serious influence on the replenishment decision, the replenishment decision obtained based on the predicted sales volume acquisition scheme better meets the real requirements in the actual production environment.

Referring back to fig. 1, in step 102, the present replenishment quantity may be predicted by a pre-trained replenishment model based on a plurality of parameters under each group of expected arrival times, to obtain at least one predicted present replenishment quantity, where the replenishment model is trained by the aforementioned dynamic simulator based on different arrival time sequences, and the arrival time sequences include at least one arrival time, and the arrival time represents a time required from shipment to arrival. Here, as an example, the different sequences of arrival durations used in training the restocking model are generated by a dynamic simulator randomly sampling arrival durations that are actually generated by the product over a historical period of time in some distribution (e.g., poisson distribution, etc.).

In one embodiment, the plurality of parameters may be processed into the feature vector of the preset dimension under a set of estimated time of arrival, and then the feature vector is input into the replenishment model to obtain a predicted current replenishment volume, and the process is repeated to obtain a predicted current replenishment volume under each set of estimated time of arrival. For example, the multiple parameters may be processed into a 6-dimensional vector as shown below:

wherein,and->Respectively representing the mean and the variance of the real sales in a first historical preset time period; />And->Respectively representing a mean value and a variance of the predicted sales in a first future preset time period; /> Representing the estimated time of next restocking closest to the estimated time of the present restocking, t representing the time of the present restocking decision; e, e _s，n (t)＝min(0，q _s，n (t)-I _s，n (t))，q _s，n (t) represents the predicted sales at the current time (e.g., the current day), I _s，n (t) represents the inventory of retail stores at the current time; f (f) _s，n (t) represents the cumulative restocking amount expected to arrive between the expected arrival times of the current to current restocking.

In other embodiments, the present invention is not limited thereto, and a plurality of parameters may be processed into corresponding feature vectors under a set of expected time and then input into the restocking model to obtain a predicted present restocking amount.

In one embodiment, the restocking model may be pre-trained by: acquiring a training data set; training the restocking model by using a preset algorithm based on the training data set to obtain a trained restocking model, wherein an optimization objective of the preset algorithm is: the comprehensive value of the spot rate and the inventory turnover days of the product in the first preset time length is minimum; the stock rate and the stock turnover number are obtained by deducting the stock quantity of the product in the first preset time by the dynamic simulator according to the predicted sales quantity and the real stock quantity of the product in the first preset time and the stock supplement quantity predicted by the stock supplement model based on different arrival time sequences. Here, the preset algorithm may be, for example, a random optimization algorithm or the like, and is not limited thereto.

In one embodiment, the data training dataset may include historical real data of the product over a period of time, which may include historical real sales, real inventory, lead time (time required from shipment to arrival), shippable day, receivable day, etc., the historical real data over the period of time being divided by the longest lead time or multiples of the longest lead time as a time slice, a plurality of training samples may be obtained that exist in a time series. In addition, the training data set further includes a predicted sales of a plurality of training samples, the predicted sales being obtainable by the sales prediction model described above. For a training sample, establishing an optimization objective of the following formula (1) over a historical time period T (i.e., a first preset duration) defined by the training sample:

Wherein s, n represent sku (product) and node (retail store), respectively; c represents a super parameter; q (Q) _s，n (t) represents the historical actual sales of the sku-node (representing a product in a retail store) on day t; i _s，n (t) represents the initial inventory of the sku-node on day t (i.e., the inventory just at the beginning of day t) as deduced by the dynamic simulator based on the replenishment decisions.

Here, in the training process of the replenishment model, the dynamic simulator may perform a fluctuation simulation for the leader time of the product, and under the condition that each leader time is defined, train the replenishment model separately, so as to improve the robustness of the output result of the replenishment model in the actual production environment. Specifically, the dynamic simulator may automatically generate a plurality of leader time sequences conforming to a certain distribution (for example, poisson distribution, etc.), and select one leader time sequence for each training, because the time length of the training samples is divided according to the real longest leader time of the product, for each training sample, in the process of obtaining the replenishment quantity during training, at least one replenishment quantity corresponding to the replenishment decision arrives within a time period defined by the training sample, and the dynamic simulator may deduce the daily spot rate and the inventory turnover number in a historical time period defined by the training sample according to the real initial inventory of the first day in the training sample, the predicted sales quantity of the subsequent day and the replenishment quantity output by the replenishment model during the current training, train the replenishment model with the sum of the spot rate and the inventory turnover number as an optimization target, and obtain the model parameters of the replenishment model under the leader time sequence. The training process is repeated for other leader sequences, so that multiple groups of model parameters can be obtained finally, multiple groups of model parameters (for example, weighted average and the like) are synthesized, and a final trained replenishment model can be obtained.

Wherein, in the deduction process of the dynamic simulator, I _s，n (t+1) and I _s，n The recurrence relation of (t) can be expressed as the following formula (2):

I _s，n (t+1)＝max(0，I _s，n (t)-q _s，n (t))+∑ _i≤t X(a _s，n (i)＝t)R _w (i，s，n) (2)

wherein I (1) represents the actual initial stock quantity of the sku-node at the start date of the historical time period defined by the training sample; q _s，n (t) represents the predicted sales of the sku-node on day t; a, a _s，n (i) Representing the arrival time of the replenishment quantity corresponding to the replenishment decision of the sku-node on the i-th day, wherein i is less than or equal to t; x () represents an indicator function of 0 or 1; r is R _w (i, s, n) represents the replenishment quantity corresponding to the replenishment decision on the i-th day.

In addition, in the training process of the replenishment model, the model may be input with reference to the aforementioned 6-dimensional vector, but it should be noted that each item of data in the 6-dimensional vector should be data over a historical period defined by the training sample, and under the determined leader time sequence, the arrival time of the replenishment volume corresponding to the replenishment decision day in the training sample, the next arrival time closest to the arrival time of the replenishment volume, and the arrival time of the replenishment volume arriving between the replenishment decision day and the arrival time of the replenishment volume corresponding thereto are all determined. In addition, the time t in the formula (1) and the formula (2) involved in the replenishment model training process is also the historical time defined by the training sample.

Referring back to fig. 1, in step 103, a first estimated replenishment quantity is determined based on at least one predicted current replenishment quantity. Here, the at least one predicted current replenishment quantity may be weighted and averaged, the weighted and averaged value may be determined as the first estimated replenishment quantity, or the at least one predicted current replenishment quantity may be added and averaged, and the average value may be determined as the first estimated replenishment quantity.

In step 104, the current replenishment decision is determined based on the first estimated replenishment quantity.

In one embodiment, the replenishment coefficient may be determined according to a current inventory health status of the product and an evaluation result for a historical replenishment decision, where the current inventory health status of the product is obtained by simulating, by the dynamic simulator, an inventory status of the product for a period of time after the current time at each set of the expected arrival times, and the historical replenishment decision is at least one replenishment decision determined for a period of time before the current time. Specifically, the present real inventory amount of the product and the real sales amount of the past time period (for example, the past two weeks, etc.) can be used by the dynamic simulator to deduce (i.e., simulate) the future inventory condition of the product, and whether the present inventory is healthy or not can be determined according to the deduced result. For example, the current inventory can meet sales requirements between the current to the time of the replenishment amount corresponding to the current replenishment decision, and the current inventory condition can be considered healthy. In addition, the theoretical optimal inventory turnover number can be calculated, whether the current inventory is healthy or not is determined according to the approaching degree between the inventory turnover number simulated by the dynamic simulator and the theoretical optimal inventory turnover number, the two deduction conditions are combined, the coefficient reflecting the health condition of the inventory is calculated as a part of the replenishment coefficient, and the coefficient reflecting the health condition of the inventory can be calculated according to any one of the two deduction conditions. Another portion of the restocking coefficients may be determined based on historical restocking decisions affecting inventory, meeting real demand, etc.

After the replenishment program is determined, a second estimated replenishment quantity may be determined based on the replenishment program and the predicted backorder condition within a period of time after the current time, and the current replenishment decision may be determined based on the first estimated replenishment quantity and/or the second estimated replenishment quantity. Specifically, the future quantity of the stock is estimated approximately based on the current actual stock quantity, the quantity of the stock expected to be reached before the quantity of the stock corresponding to the current stock decision arrives, and the predicted sales quantity, and the second estimated stock quantity is obtained by multiplying the stock quantity by the stock quantity. Here, the future backorder amount, for example, but not limited to, can be estimated according to the following formula (3):

in this formula, the need to make a week per decision is an exemplary premise of implementation, given that the time required for a shipment, the product from shipment to arrival, is 14 days on the day after the make-up decision is determined on the day. Wherein, the stop is out _14-21 Indicating the quantity of backorder on days 14-21; net demand _14-21 Indicating the net demand (sales minus returns) on days 14-21; inventory of ₀ Representing a real inventory of the first day;indicating the amount of arrival (determined by previous restocking decisions) that may exist on days 1-14; / >Indicating demand on days 1-14.

According to the above scheme for determining the second estimated replenishment quantity, under each group of estimated time, the dynamic simulator simulates the inventory condition within a period of time after the current time, and guides the determination of the second estimated replenishment quantity by the current inventory health condition of the product determined based on the simulation result, and because the uncertainty of the arrival time of the product is considered, the replenishment decision made based on the determined second estimated replenishment quantity more accords with the real demand in the actual production environment.

In one embodiment, the first estimated replenishment quantity may be determined as the current replenishment decision.

In another embodiment, the second estimated replenishment quantity may be determined as the current replenishment decision.

In another embodiment, a weighted average of the first and second estimated replenishment amounts may be determined as the current replenishment decision.

In other embodiments, the present replenishment decision may be determined according to the first estimated replenishment quantity, the second estimated replenishment quantity, and the buffer quantity for performing the abnormal point protection, where the abnormal point protection refers to that in determining the present replenishment decision, it is detected that the actual sales quantity of the product in the second historical preset time period (for example, in the first 1 week, the first 2 weeks, etc. of the present time) has abnormal fluctuations that meet the abnormal determination criterion, and measures are taken to cope with the possible reoccurrence abnormal fluctuations. Here, the abnormality determination criterion may be set according to actual conditions, for example, but not limited to, that the sales of the product are all within a set abnormal sales interval for a period of time, or that the number of days that the sales of the product are within a set abnormal sales interval for a period of time reaches a set threshold, or the like.

In one embodiment, the buffer size may be determined by: under the condition that the replenishment decision is a first replenishment decision for the product, determining the buffer quantity according to the real sales quantity of the product in a second historical preset time period and the historical abnormal sales quantity statistic value of the product, and under the condition that the replenishment decision is not the first replenishment decision for the product, determining the buffer quantity according to the real sales quantity of the product in the second historical preset time period, the historical abnormal sales quantity statistic value of the product and the resisting execution effect of the abnormal point in the second historical preset time period. In particular, during actual sales, the sales of the product is not always stable, for example, there may be various regular or irregular sales promotion activities, resulting in situations such as higher sales of the product during certain time periods, which may lead to errors between the predicted sales and the actual sales, and thus to errors in replenishment decisions with actual demands in the actual production environment. In order to make the replenishment decision more in line with the actual demand in the actual production environment, the historical abnormal sales volume of the product occurring in the past time period (for example, the past year, etc.) and the frequency of occurrence of the historical abnormal sales volume can be counted in advance, the historical abnormal sales volume statistics value of the product is determined based on the counted result (for example, different historical abnormal sales volumes can be calculated according to the frequency of occurrence of the historical abnormal sales volume, the historical abnormal sales volume can be multiplied by the corresponding weight and then added up, the historical abnormal sales volume statistics value of the product can be obtained), whether the abnormal point is started or not is determined according to the actual sales volume of the product in the second historical preset time period under the condition that the replenishment decision is made for the first time of the product, if the actual sales volume of the product in the second historical preset time period accords with the abnormal fluctuation of the abnormality judgment standard, the abnormal point is determined to be needed, the abnormal point can be determined according to the actual sales volume of the product in the second historical preset time period (for example, the abnormal point is realized as a matrix), and the current abnormal point resistance coefficient and the current abnormal point resistance statistics value of the product can be obtained. And if the actual sales volume of the product in the second historical preset time period accords with the abnormal fluctuation of the abnormal judgment standard, determining that the abnormal point is required to be resisted, determining an abnormal resisting coefficient according to the actual sales volume of the product in the second historical preset time period and the executing effect of the abnormal point resisting in the second historical preset time period, and multiplying the abnormal resisting coefficient by the historical abnormal sales volume statistical value of the product to obtain the buffering quantity for resisting the abnormal point. Here, the execution effect of the abnormal point defense in the second history preset period refers to the execution effect of the abnormal point defense in the second history preset period included in the replenishment decision made before the second history preset period, and the execution effect may be measured by the spot rate in the second history preset period as an example.

Referring to fig. 4, in determining that there is abnormal fluctuation in the real sales amount in 2 weeks before the 4 th week in the process of determining the present restocking decision in 4 th week, and performing the present restocking decision in 2 weeks, the level of the present restocking can be determined according to the degree of abnormal fluctuation in the real sales amount in the previous 2 weeks and the effect of performing the present restocking in 2 weeks (e.g., measured in spot rate and stock turnover days), and as an example, fig. 4 represents different levels of the present restocking in terms of the shade of the state color patch in which the level of the real sales amount in 2 weeks is large (the color of the state color patch is deep in the anomaly point) while in determining the present restocking decision in 4 th week, the present restocking decision is performed in combination with the effect of the real sales amount in the previous 2 weeks and the point of performing the anomaly point defense in 2 weeks, but the level of the present restocking is weaker (the state color patch is light in the anomaly point) than in 2 weeks, which may be due to the degree of the abnormal fluctuation in the previous 2 weeks or the high in the spot point of the state color patch is performed in the 2 weeks. And by analogy, determining whether to conduct outlier rejection according to the process every time a replenishment decision is determined after the 4 th week, and determining the degree of outlier rejection under the condition that the outlier rejection is determined to be needed. That is, the abnormal point resisting process is a dynamic process following the real situation, and the resisting force of the abnormal point can be timely adjusted according to the latest real sales volume situation and the executing situation of the abnormal point resisting.

In one embodiment, the second estimated replenishment quantity may be added to the buffer quantity for performing outlier rejection to obtain a third estimated replenishment quantity, and a weighted average of the first estimated replenishment quantity and the third estimated replenishment quantity is determined as the current replenishment decision. As an example, the third estimated replenishment quantity may be expressed as formula (4):

Rep＝Adapative _coef *stockout+anomalies_state_matrix*AADS-Value (4)

wherein, rep represents a third estimated replenishment quantity; adaptive_coef represents the aforementioned replenishment coefficient; the stock out represents the estimated future backorder quantity; the analysis_state_matrix represents an anomaly defense coefficient matrix whose value varies depending on the degree of anomaly of the actual sales in the second preset history period and the effect of execution of anomaly point defense in the second preset history period; aads_value represents a historical abnormal sales statistic for a product.

The current replenishment decision determined based on equation (4), for example, but not limited to, may be expressed as:

Final_rep＝w*model_pred+(1-w)*Rep (5)

wherein, final_rep represents the Final decision of the current replenishment quantity; model_pred represents a first estimated replenishment quantity; and Rep represents the third estimated replenishment quantity.

In one embodiment, the restocking model and the at least one sales prediction model may be retrained at preset time intervals based on the real sales data and the real inventory data generated during the preset time intervals to obtain a new trained restocking model and the at least one sales prediction model. Because the model is retrained based on the newly generated data, new situations (such as new sales distribution rules and the like) which may occur in the actual production environment can be followed in time, so that the replenishment decision is more in line with the actual demand in the actual production environment.

Referring to fig. 5, a time sequence prediction model (i.e., sales prediction model) predicts a predicted sales volume for a period of time after a current time, when making a replenishment decision optimization, outputs the predicted sales volume to a replenishment model, the replenishment model interacts with a dynamic simulator according to the foregoing related description to output a first estimated replenishment volume, in addition, automatically generates a replenishment coefficient and a predicted future period of shortage volume according to the foregoing related description to obtain a second estimated replenishment volume, performs weighted average on the first estimated replenishment volume and the second estimated replenishment volume to obtain a replenishment volume meeting the actual demand of an actual production environment, further determines whether to perform an abnormal point resistance according to decision-making auxiliary information (e.g., the actual sales volume in the foregoing second preset historical time period, etc.), determines a buffer volume for performing an abnormal point resistance according to the foregoing related description when determining that an abnormal point resistance is required, and determines a current replenishment decision according to the first estimated replenishment volume, the second estimated replenishment volume and the buffer volume to further obtain a real replenishment volume meeting the actual demand of an actual production environment.

In addition, parts related to the replenishment decision (including decision assistance, time sequence prediction model, decision optimization and dynamic simulator) can be used online, real data generated by the online environment can be timely updated to the dynamic simulator, and the replenishment model and the time sequence prediction model can be retrained according to real data generated by the online environment at preset time intervals (for example, two months and the like) to obtain a new trained replenishment model and time sequence prediction model for determining the replenishment decision.

Referring to fig. 6, an intelligent supply chain restocking system 600 of an exemplary embodiment of the present disclosure includes a parameter acquisition device 601, a restocking amount prediction device 602, a first estimated restocking amount determination device 603, and a restocking decision determination device 604.

The parameter obtaining device 601 may obtain a plurality of parameters related to the current replenishment quantity, where the plurality of parameters include a real sales quantity of the product in a first historical preset time period, a predicted sales quantity of the product in a first future preset time period, a predicted arrival time of the current replenishment, a predicted arrival time of a next replenishment closest to the predicted arrival time of the current replenishment, a current spot situation, and a predicted arrival quantity of the product between the predicted arrival times of the current replenishment, and the predicted arrival time of the next replenishment closest to the predicted arrival time of the current replenishment, and the predicted arrival time of the predicted arrival quantity of the product between the predicted arrival times of the current replenishment are used as a set of predicted arrival times, and the dynamic simulator obtains at least one set of the predicted arrival times as a set of predicted arrival times.

In one embodiment, the predicted sales in the future predetermined time period is predicted from one or more of the at least one sales prediction model based on sales characteristics of the product. Here, the sales characteristic may be, for example, but not limited to, "high sales/low sales", "high return rate/low return rate", etc., and the sales characteristic of the product may be obtained by clustering different products, or manually analyzing historical sales data of the product, etc. As an example, one sales prediction model may be employed to obtain a predicted sales for low sales products, while multiple sales prediction models may be employed to obtain a predicted sales for high sales products.

Referring back to fig. 6, the replenishment quantity predicting device 602 may predict, based on a plurality of parameters, the current replenishment quantity by using a pre-trained replenishment model to obtain at least one predicted current replenishment quantity, where the replenishment model is trained by using the dynamic simulator based on different arrival time sequences, and the arrival time sequences include at least one arrival time, and the arrival time represents a time required from shipment to arrival. Here, as an example, the different sequences of arrival durations used in training the restocking model are generated by a dynamic simulator randomly sampling arrival durations that are actually generated by the product over a historical period of time in some distribution (e.g., poisson distribution, etc.).

In one embodiment, the foregoing parameters may be processed into feature vectors of a predetermined dimension under a set of estimated time of arrival, and then the feature vectors are input into the replenishment model to obtain a predicted current replenishment volume, and the process is repeated to obtain a predicted current replenishment volume for each set of estimated time of arrival. Here, for example, a plurality of parameters may be processed as the aforementioned 6-dimensional vector.

In one embodiment, the replenishment model may be trained in advance through the foregoing training process, and for brevity of description, details are not repeated here.

Referring back to fig. 6, the first estimated replenishment quantity determining device 603 may determine the first estimated replenishment quantity according to at least one predicted current replenishment quantity. Here, the first estimated replenishment quantity determining device 603 may determine the weighted average of the at least one predicted current replenishment quantity as the first estimated replenishment quantity, or may determine the average value as the first estimated replenishment quantity by adding up the at least one predicted current replenishment quantity and then taking the average value.

The replenishment decision determining device 604 may determine the current replenishment decision based on the first estimated replenishment volume.

In one embodiment, the replenishment decision determining device 604 may determine the replenishment coefficient according to the current inventory health status of the product and the evaluation result for the historical replenishment decision, where the current inventory health status of the product is obtained by simulating the inventory status of the product for a period of time after the current time by the dynamic simulator at each set of the expected arrival times, and the historical replenishment decision is at least one replenishment decision determined for a period of time before the current time. Specifically, the replenishment decision determining device 604 may calculate, by the dynamic simulator, whether the current inventory is healthy or not with the current actual inventory of the product and the actual sales of the past time (for example, two weeks in the past, etc.), determine whether the current inventory is healthy or not based on the result of the deduction (for example, the current inventory can satisfy the sales demand between the arrival time of the replenishment amount corresponding to the current replenishment decision and the current inventory condition can be considered healthy), calculate the theoretical optimal inventory turnover number, determine whether the current inventory is healthy or not with the proximity between the inventory turnover number simulated by the dynamic simulator and the optimal theoretical turnover number, calculate the coefficient reflecting the inventory health condition as a part of the replenishment coefficient based on the result of the deduction, and the replenishment decision determining device 603 may calculate the coefficient reflecting the inventory health condition based on either of the two deduction cases, which is not limited by the present disclosure. Further, the restocking decision-determining device 603 may determine another part of the restocking coefficient according to the influence condition of the historical restocking decision on the stock quantity, the satisfaction condition of the real demand, or the like.

After determining the replenishment coefficient, the replenishment decision determining device 604 may determine a second estimated replenishment amount based on the replenishment coefficient and the estimated backorder condition within a period of time after the current time, and determine the current replenishment decision based on the first estimated replenishment amount and/or the second estimated replenishment amount. Specifically, the replenishment decision determining device 604 may multiply the replenishment coefficient with the future quantity of the stock estimated approximately based on the current actual stock quantity, the quantity of the stock estimated to be likely to be reached before the arrival of the quantity of the stock corresponding to the current replenishment decision, and the predicted sales quantity to obtain the second estimated quantity of the stock. Here, the future backorder amount, for example, but not limited to, can be expressed by the aforementioned formula (3).

In one embodiment, the replenishment decision determining device 604 may determine the first estimated replenishment quantity as the current replenishment decision.

In another embodiment, the replenishment decision determining device 604 may determine the second estimated replenishment quantity as the current replenishment decision.

In another embodiment, the replenishment decision determining device 604 may determine a weighted average of the first estimated replenishment volume and the second estimated replenishment volume as the current replenishment decision.

In other embodiments, the replenishment decision determining device 604 may determine the current replenishment decision according to the first estimated replenishment quantity, the second estimated replenishment quantity, and the buffer quantity for performing the abnormal point protection, where the abnormal point protection refers to that in determining the current replenishment decision, it is detected that the actual sales quantity of the product within the second historical preset time period (for example, the first 1 week, the first 2 weeks, etc. of the current time) has abnormal fluctuations meeting the abnormal decision criteria, and measures are taken to cope with the possible reoccurring abnormal fluctuations. Here, the abnormality determination criterion may be set according to actual conditions, for example, but not limited to, that the sales of the product are all within a set abnormal sales interval for a period of time, or that the number of days that the sales of the product are within a set abnormal sales interval for a period of time reaches a set threshold, or the like.

In one embodiment, the buffer size may be determined by: under the condition that the replenishment decision is a first replenishment decision for the product, determining the buffer quantity according to the real sales quantity of the product in a second historical preset time period and the historical abnormal sales quantity statistic value of the product, and under the condition that the replenishment decision is not the first replenishment decision for the product, determining the buffer quantity according to the real sales quantity of the product in the second historical preset time period, the historical abnormal sales quantity statistic value of the product and the resisting execution effect of the abnormal point in the second historical preset time period. Specifically, in the actual sales process, the sales volume of the product is not always stable, for example, various regular or irregular sales promotion activities may exist, which results in the situation that the sales volume of the product is higher in certain time periods, and the like, and thus, errors exist between the predicted sales volume and the actual sales volume, so that, in order to make the replenishment decision more in line with the actual demand in the actual production environment, in advance, the historical abnormal sales volume of the product which occurs in the past time period (for example, the past year, etc.) and the frequency of occurrence of the historical abnormal sales volume of the product may be counted, for example, the weights may be calculated for different historical abnormal sales volumes according to the frequency of occurrence of the historical abnormal sales volume, the historical abnormal sales volume of the product may be added up after being multiplied by the corresponding weights, and the historical abnormal sales volume statistical value of the product may be obtained. And if the actual sales volume of the product in the second historical preset time period accords with the abnormal fluctuation of the abnormal judgment standard, determining that the abnormal point is required to be resisted, determining an abnormal resisting coefficient according to the actual sales volume of the product in the second historical preset time period and the executing effect of the abnormal point resisting in the second historical preset time period, and multiplying the abnormal resisting coefficient by the historical abnormal sales volume statistical value of the product to obtain the buffering quantity for resisting the abnormal point. Here, the execution effect of the abnormal point defense in the second history preset period refers to the execution effect of the abnormal point defense in the second history preset period included in the replenishment decision made before the second history preset period, and the execution effect may be measured by the spot rate in the second history preset period as an example.

A specific abnormal point protection state transition process may refer to the foregoing description of fig. 4, and is not repeated herein for brevity of description.

In one embodiment, the replenishment decision determining device 604 may add the second estimated replenishment quantity to the buffer quantity for performing the abnormal point protection to obtain the third estimated replenishment quantity, and determine a weighted average of the first estimated replenishment quantity and the third estimated replenishment quantity as the current replenishment decision. As an example, the third estimated replenishment quantity may be represented as formula (4) above, and the current replenishment decision determined based on formula (4), for example, but not limited to, may be represented as formula (5) above.

In one embodiment, the intelligent supply chain restocking system further includes a model retraining device 605 (not shown in fig. 6), the model retraining device 605 retraining the restocking model and the at least one sales prediction model based on the real sales data and the real inventory data generated during the preset time interval at intervals to obtain a new trained restocking model and the at least one sales prediction model. Because the model is retrained based on the newly generated data, new situations (such as new sales distribution rules and the like) which may occur in the actual production environment can be followed in time, so that the replenishment decision is more in line with the actual demand in the actual production environment.

Fig. 7 is a block diagram of an electronic device 700 according to an exemplary embodiment of the present disclosure.

Referring to fig. 7, an electronic device 700 includes at least one memory 701 and at least one processor 702, the at least one memory 701 having stored therein a set of computer-executable instructions that, when executed by the at least one processor 702, perform an intelligent supply chain restocking method according to an example embodiment of the disclosure.

By way of example, the electronic device 700 may be a PC computer, tablet device, personal digital assistant, smart phone, or other device capable of executing the above-described set of instructions. Here, the electronic device 700 is not necessarily a single electronic device, but may be any apparatus or a collection of circuits capable of executing the above-described instructions (or instruction set) individually or in combination. The electronic device 700 may also be part of an integrated control system or system manager, or may be configured as a portable electronic device that interfaces with either locally or remotely (e.g., via wireless transmission).

In electronic device 700, processor 702 may include a Central Processing Unit (CPU), a Graphics Processor (GPU), a programmable logic device, a special purpose processor system, a microcontroller, or a microprocessor. By way of example, and not limitation, processors may also include analog processors, digital processors, microprocessors, multi-core processors, processor arrays, network processors, and the like.

The processor 702 may execute instructions or code stored in the memory 701, wherein the memory 701 may also store data. The instructions and data may also be transmitted and received over a network via a network interface device, which may employ any known transmission protocol.

The memory 701 may be integrated with the processor 702, for example, RAM or flash memory disposed within an integrated circuit microprocessor or the like. In addition, the memory 701 may include a separate device, such as an external disk drive, a storage array, or any other storage device usable by a database system. The memory 701 and the processor 702 may be operatively coupled or may communicate with each other, for example, through an I/O port, a network connection, etc., such that the processor 702 is able to read files stored in the memory.

In addition, the electronic device 700 may also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, mouse, touch input device, etc.). All components of the electronic device 700 may be connected to each other via a bus and/or a network.

According to an exemplary embodiment of the present disclosure, a computer-readable storage medium storing instructions may also be provided, wherein the instructions, when executed by at least one processor, cause the at least one processor to perform an intelligent supply chain restocking method according to the present disclosure. Examples of the computer readable storage medium herein include: read-only memory (ROM), random-access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random-access memory (RAM), dynamic random-access memory (DRAM), static random-access memory (SRAM), flash memory, nonvolatile memory, CD-ROM, CD-R, CD + R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD + R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-RLTH, BD-RE, blu-ray or optical disk storage, hard Disk Drives (HDD), solid State Disks (SSD), card-type memories (such as multimedia cards, secure Digital (SD) cards or extreme digital (XD) cards), magnetic tapes, floppy disks, magneto-optical data storage devices, hard disks, solid state disks, and any other devices configured to store computer programs and any associated data, data files and data structures in a non-transitory manner and to provide the computer programs and any associated data, data files and data structures to a processor or computer to enable the processor or computer to execute the programs. The computer programs in the computer readable storage media described above can be run in an environment deployed in a computer device, such as a client, host, proxy device, server, etc., and further, in one example, the computer programs and any associated data, data files, and data structures are distributed across networked computer systems such that the computer programs and any associated data, data files, and data structures are stored, accessed, and executed in a distributed fashion by one or more processors or computers.

In addition, sales of the product in a future period of time is respectively predicted through a plurality of sales prediction models of different types, the prediction results of the sales prediction models are combined to determine the future sales of the product, the accuracy of sales prediction can be improved, and then the replenishment decision made based on the predicted sales prediction accords with the real requirements in the actual production environment.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any adaptations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It is to be understood that the present disclosure is not limited to the precise arrangements and instrumentalities shown in the drawings, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. An intelligent supply chain replenishment method, comprising: obtaining a plurality of parameters related to the amount of the present replenishment, the plurality of parameters including a true sales amount of the product in a first historical preset time period, a predicted sales amount of the product in a first future preset time period, a predicted arrival time of the present replenishment, a predicted arrival time of a next replenishment closest to the predicted arrival time of the present replenishment, a current spot, and an expected arrival amount between the current arrival time of the present replenishment, wherein,

taking the estimated time of the current replenishment, the estimated time of the next replenishment closest to the estimated time of the current replenishment and the estimated time of the replenishment amount estimated to be reached between the current estimated time of the current replenishment and the estimated time of the current replenishment as a set of estimated time of the replenishment, obtaining at least one set of estimated time of the replenishment by a dynamic simulator, wherein the dynamic simulator is used for simulating the inventory condition of the product in a period of time after the current time;

Under each group of expected arrival time, predicting the current replenishment quantity through a pre-trained replenishment model based on the plurality of parameters to obtain at least one predicted current replenishment quantity, wherein the replenishment model is trained through the dynamic simulator based on different arrival time sequences, the arrival time sequences comprise at least one arrival time, and the arrival time represents the time required from delivery to arrival;

determining a first estimated replenishment quantity according to the at least one predicted current replenishment quantity;

and determining the current replenishment decision based on the first estimated replenishment quantity.

2. The method of claim 1, wherein determining the current replenishment decision based on the first estimated replenishment quantity comprises:

determining replenishment coefficients according to the current inventory health condition of the product and an evaluation result aiming at a historical replenishment decision, wherein the current inventory health condition of the product is obtained by simulating the inventory condition of the product for a period of time after the current time under each group of expected arrival times respectively, and the historical replenishment decision is at least one replenishment decision determined for a period of time before the current time;

Determining a second estimated replenishment quantity according to the replenishment coefficient and the estimated backorder condition in a period of time after the current time;

and determining the current replenishment decision based on the first estimated replenishment quantity and/or the second estimated replenishment quantity.

3. The method of claim 2, wherein determining the current replenishment decision based on the first estimated replenishment volume and/or the second estimated replenishment volume comprises:

determining the first estimated replenishment quantity as the current replenishment decision, or,

determining the second estimated replenishment quantity as the current replenishment decision, or,

determining a weighted average of the first and second estimated replenishment amounts as a current replenishment decision, or,

determining a current replenishment decision according to the first estimated replenishment quantity, the second estimated replenishment quantity and a buffer quantity for resisting an abnormal point, wherein the abnormal point resisting means that in the process of determining the current replenishment decision, abnormal fluctuation meeting an abnormal judgment standard is detected to exist in the real sales quantity of the product in a second historical preset time period, and measures are taken to cope with the possible reappearance of the abnormal fluctuation.

4. A method according to claim 3, wherein the buffer amount is determined by:

under the condition that the current replenishment decision is a first replenishment decision for the product, determining the buffer quantity according to the real sales quantity of the product in the second historical preset time period and the historical abnormal sales quantity statistic value of the product;

and under the condition that the current replenishment decision is not the first replenishment decision for the product, determining the buffer quantity according to the real sales volume of the product in the second historical preset time period, the historical abnormal sales volume statistic value of the product and the execution effect of resisting the abnormal point in the second historical preset time period.

5. The method of claim 1, wherein the restocking model is pre-trained by:

acquiring a training data set;

training the restocking model with a preset algorithm based on a training dataset to obtain a trained restocking model,

wherein, the optimization objective of the preset algorithm is: the comprehensive value of the spot rate and the inventory turnover number of the product in the first preset time length is minimum;

the stock rate and the inventory turnover number are obtained by deducting the stock quantity of the product in the first preset time by the dynamic simulator according to the predicted sales quantity and the real stock quantity of the product in the first preset time and the stock quantity predicted by the stock supplement model based on different arrival time sequences, and the first preset time is obtained by the real arrival time of the product.

6. The method of claim 1, wherein,

the predicted sales in the first future preset time period is predicted by at least one sales prediction model trained in advance, wherein each sales prediction model is different in type.

7. The method of claim 6, wherein,

the predicted sales in the first future preset time period is predicted by one or more sales prediction models of the at least one sales prediction model according to sales characteristics of the product.

8. An intelligent supply chain restocking system, comprising:

parameter acquisition means configured to: obtaining a plurality of parameters related to the amount of the present replenishment, the plurality of parameters including a true sales amount of the product in a first historical preset time period, a predicted sales amount of the product in a first future preset time period, a predicted arrival time of the present replenishment, a predicted arrival time of a next replenishment closest to the predicted arrival time of the present replenishment, a current spot, and an expected arrival amount between the current arrival time of the present replenishment, wherein,

A replenishment quantity predicting device configured to: under each group of expected arrival time, predicting the current replenishment quantity through a pre-trained replenishment model based on the plurality of parameters to obtain at least one predicted current replenishment quantity, wherein the replenishment model is trained through the dynamic simulator based on different arrival time sequences, the arrival time sequences comprise at least one arrival time, and the arrival time represents the time required from delivery to arrival;

the first estimated replenishment quantity determining device is configured to: determining a first estimated replenishment quantity according to the at least one predicted current replenishment quantity;

a restocking decision-making device configured to: and determining the current replenishment decision based on the first estimated replenishment quantity.

9. An electronic device, comprising:

at least one processor;

at least one memory storing computer-executable instructions,

wherein the computer-executable instructions, when executed by the at least one processor, cause the at least one processor to perform the intelligent supply chain restocking method of any one of claims 1 to 7.

10. A computer-readable storage medium storing instructions that, when executed by at least one processor, cause the at least one processor to perform the intelligent supply chain restocking method of any of claims 1-7.