Replenishment method and device, electronic device, and storage medium

Abstract

The present disclosure provides a replenishment method and device, an electronic device, and a storage medium; and relates to the field of big data technology. The method includes: obtaining and determining the historical forecast deviation of the target item at multiple different lead times based on the historical actual transaction data and historical forecast data of the target item at multiple different lead times; establishing a non-parametric model and a parametric model based on the historical forecast deviations of multiple different lead times; integrating the non-parametric model and the parametric model to obtain an integrated model; determining the transaction fluctuation data of the target item at a target future date based on the integrated model, and sending the transaction fluctuation data to a replenishment system, so that the replenishment system determines the replenishment quantity based on the transaction fluctuation data. The present disclosure can improve the accuracy of replenishment.

Description

Technical Field

[0001] The present disclosure relates to the field of big data technology, and in particular to a replenishment method, a replenishment device, an electronic device, and a computer-readable storage medium. Background Art

[0002] With the rapid development of information technology, the supply chain construction of retail enterprises is gradually becoming intelligent. In the replenishment stage, relying on massive transaction data, combined with multiple factors such as marketing information, inventory costs, and transportation costs, an intelligent replenishment model that can automatically give the optimal expected order quantity can be trained, which can greatly save labor costs and improve order accuracy. It is an important part of the current intelligent supply chain construction of retail enterprises.

[0003] Among them, the replenishment model contains two key inputs: the mean forecast of future transactions and the volatility forecast, which can also be called the standard deviation forecast. Most existing studies focus on the mean forecast, and the prediction accuracy is relatively high. However, there are fewer methods for volatility prediction, and the prediction accuracy is relatively low, which ultimately leads to low replenishment accuracy.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The purpose of the present disclosure is to provide a replenishment method and device, an electronic device, and a computer-readable storage medium, thereby overcoming the problem of low replenishment accuracy caused by the limitations and defects of related technologies to a certain extent.

[0006] According to a first aspect of the present disclosure, there is provided a replenishment method, comprising:

[0007] Obtaining and determining historical forecast deviations of the target item at the multiple different lead times based on historical actual transaction data and historical forecast data of the target item at the multiple different lead times;

[0008] Based on the historical forecast deviations of the plurality of different lead times, a non-parametric model and a parametric model are established respectively;

[0009] Integrating the non-parametric model and the parametric model to obtain an integrated model;

[0010] According to the integrated model, transaction fluctuation data of the target item on a target future date is determined, and the transaction fluctuation data is sent to a replenishment system, so that the replenishment system determines a replenishment quantity according to the transaction fluctuation data.

[0011] Optionally, a non-parametric model is established based on the historical forecast deviations of the multiple different lead times, including:

[0012] Based on the historical forecast deviations of the multiple different lead times, a non-parametric model is established by kernel density estimation.

[0013] Optionally, the establishing of a non-parametric model by kernel density estimation based on the historical forecast deviations of the multiple different lead times includes:

[0014] According to factors affecting the transaction data, the historical forecast deviations of the plurality of different lead times are divided, and a local probability distribution of the historical forecast deviations of the groups corresponding to a single factor is determined;

[0015] Based on the local probability distribution of the historical prediction deviations corresponding to a single group, the overall deviation distribution corresponding to the group is determined, and the overall deviation distribution corresponding to each group is used as a non-parametric model.

[0016] Optionally, a parameter model is established based on the historical forecast deviations of the multiple different lead times, including:

[0017] According to the generalized autoregressive conditional heteroskedasticity model and the factors affecting the transaction data, a threshold generalized autoregressive conditional heteroskedasticity model is established, wherein the threshold variable in the threshold generalized autoregressive conditional heteroskedasticity model is determined according to the factors affecting the transaction data;

[0018] Based on the historical forecast deviations of the multiple different lead times, the parameters in the threshold generalized autoregressive conditional heteroskedasticity model are solved by the maximum likelihood estimation method, and the solved threshold generalized autoregressive conditional heteroskedasticity model is used as a parameter model.

[0019] Optionally, the threshold generalized autoregressive conditional heteroskedasticity model is specifically:

[0020]

[0021] Among them, ε i,n =σ i,n z n , ε i,n represents the residual corresponding to the target item i in time series n, represents the variance of target item i in time series n, z n is a normal distribution with a mean of 0 and a standard deviation of 1, ω i ,α i ,β i ,γ represents the model parameter, tv represents the threshold variable, and d represents the threshold value.

[0022] Optionally, the integrating the non-parametric model and the parametric model to obtain an integrated model that meets a target replenishment condition includes:

[0023] The non-parametric model and the parametric model are integrated by a stacking method, and the weight of the non-parametric model and the weight of the parametric model are determined to obtain an integrated model.

[0024] Optionally, determining the transaction fluctuation data of the target item on a target future date according to the integrated model includes:

[0025] Determine factors affecting transaction data corresponding to the target future date;

[0026] The transaction fluctuation data of the target item on the target future date is determined according to the integrated model and the factors affecting the transaction data corresponding to the target future date.

[0027] Optionally, the replenishment method of the embodiment of the present disclosure further includes:

[0028] Determining, based on the parameter model, forecast residuals for the target item at multiple future dates;

[0029] When the mean of the prediction residuals is not 0, the mean of the prediction residuals is sent to the replenishment system, so that the replenishment system corrects the predicted transaction data of the multiple future dates according to the mean of the prediction residuals.

[0030] Optionally, the factors affecting the transaction data include one or more of the following: forecast lead time, calendar effect, weather changes, and promotional activities.

[0031] According to a second aspect of the present disclosure, there is provided a replenishment device, comprising:

[0032] A historical forecast deviation determination module, used to obtain and determine the historical forecast deviation of the target item at the multiple different lead times based on the historical actual transaction data and historical forecast data of the target item at the multiple different lead times;

[0033] A model building module, used for building a non-parametric model and a parametric model respectively based on the historical forecast deviations of the plurality of different lead times;

[0034] A model integration module, used for integrating the non-parametric model and the parametric model to obtain an integrated model;

[0035] A data sending module is used to determine the transaction fluctuation data of the target item on a target future date according to the integrated model, and send the transaction fluctuation data to the replenishment system so that the replenishment system determines the replenishment quantity according to the transaction fluctuation data.

[0036] Optionally, the model building module includes:

[0037] The non-parametric model building unit is used to build a non-parametric model through kernel density estimation based on the historical forecast deviations of the multiple different lead times.

[0038] Optionally, the non-parametric model building unit is specifically used to divide the historical forecast deviations of the multiple different lead times according to factors affecting the transaction data, and determine the local probability distribution of the historical forecast deviations of the group corresponding to a single factor;

[0039] Based on the local probability distribution of the historical prediction deviations corresponding to a single group, the overall deviation distribution corresponding to the group is determined, and the overall deviation distribution corresponding to each group is used as a non-parametric model.

[0040] Optionally, the model building module further includes:

[0041] A parameter model building unit, used to build a threshold generalized autoregressive conditional heteroskedasticity model according to the generalized autoregressive conditional heteroskedasticity model and factors affecting transaction data, wherein the threshold variable in the threshold generalized autoregressive conditional heteroskedasticity model is determined according to the factors affecting transaction data;

[0042] Based on the historical forecast deviations of the multiple different lead times, the parameters in the threshold generalized autoregressive conditional heteroskedasticity model are solved by the maximum likelihood estimation method, and the solved threshold generalized autoregressive conditional heteroskedasticity model is used as a parameter model.

[0043] Optionally, the threshold generalized autoregressive conditional heteroskedasticity model is specifically:

[0044]

[0045] Among them, ε i,n =σ i,n z n , ε i,n represents the residual corresponding to the target item i in time series n, represents the variance of target item i in time series n, z n is a normal distribution with a mean of 0 and a standard deviation of 1, ω i ,α i ,β i ,γ represents the model parameter, tv represents the threshold variable, and d represents the threshold value.

[0046] Optionally, a model integration module is specifically used to integrate the non-parametric model and the parametric model through a stacking method, and determine the weight of the non-parametric model and the weight of the parametric model to obtain an integrated model.

[0047] Optionally, the data sending module includes:

[0048] The transaction fluctuation data determination unit is specifically used to determine the factors affecting the transaction data corresponding to the target future date; based on the integrated model and the factors affecting the transaction data corresponding to the target future date, determine the transaction fluctuation data of the target item on the target future date.

[0049] Optionally, the replenishment device of the embodiment of the present disclosure further includes:

[0050] A prediction residual determination module, configured to determine the prediction residuals of the target item at multiple future dates based on the parameter model;

[0051] When the mean of the prediction residuals is not 0, the mean of the prediction residuals is sent to the replenishment system, so that the replenishment system corrects the predicted transaction data of the multiple future dates according to the mean of the prediction residuals.

[0052] Optionally, the factors affecting the transaction data include one or more of the following: forecast lead time, calendar effect, weather changes, and promotional activities.

[0053] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute any one of the above-mentioned methods by executing the executable instructions.

[0054] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, any one of the methods described above is implemented.

[0055] The exemplary embodiments of the present disclosure may have at least some or all of the following beneficial effects:

[0056] In a replenishment method provided in an example embodiment of the present disclosure, a parametric model and a non-parametric model are generated based on the historical forecast deviations of the target item at multiple different lead times. The parametric model and the non-parametric model are integrated to obtain an integrated model, which can make up for the shortcomings of the parametric model and the non-parametric model, thereby improving the accuracy of the prediction of transaction data volatility and providing an important basis for the accurate calculation of the intelligent replenishment quantity. Therefore, the accuracy of replenishment can be improved. In addition, with the change of time, the historical actual transaction data and historical forecast data of the target item at multiple different lead times can be dynamically changing, and accordingly, the integrated model can also be dynamically changing, so the accuracy of replenishment can be further improved.

[0057] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0059] Figure 1 A flow chart of a replenishment method in an embodiment of the present disclosure is shown;

[0060] Figure 2 A flow chart of a method for establishing a non-parametric model by kernel density estimation in an embodiment of the present disclosure is shown;

[0061] Figure 3 A schematic diagram of a data partitioning method in an embodiment of the present disclosure is shown;

[0062] Figure 4 A distribution diagram of a kernel density estimation is shown;

[0063] Figure 5 A flow chart of a method for establishing a parameter model in an embodiment of the present disclosure is shown;

[0064] Figure 6 A structural schematic diagram of a replenishment device in an embodiment of the present disclosure is shown;

[0065] Figure 7 A schematic diagram of the structure of a computer system suitable for implementing an electronic device of an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0066] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the example embodiments are fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0067] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0068] In the replenishment model, the future sales volume of the product can be predicted to determine whether to replenish and the replenishment quantity. The forecast results of product sales can be determined by the mean forecast and volatility forecast. At present, the mean forecast method can include machine learning regression algorithm, econometric time series forecasting model, etc.; compared with the forecast method based on the (weighted / simple) average value of historical sales, the forecast accuracy can be improved. However, there are few algorithmic studies on volatility forecasting. The common method is to directly calculate the sample standard deviation based on historical sales, that is, according to the following formula:

[0069]

[0070] Directly fit the sample standard deviation S x , x i (i=1,…,n) represents the sales volume of historical n days, The average of historical sales.

[0071] However, the above method has the following disadvantages:

[0072] (1) There are many factors that affect sales fluctuations. Historical sales series that are not based on the same factor, such as sales with different promotional efforts, cannot be used to fit and calculate the same standard deviation.

[0073] (2) Even if the historical promotion intensity is the same, the same standard deviation cannot be used if the promotion intensity in the future forecast period is different.

[0074] (3) Simple statistical calculations cannot capture the serial dependence that is unique to the volatility risk of sales time series, such as the fact that a larger standard deviation is often followed by a larger standard deviation.

[0075] (3) Replenishment considers multi-day forecasts, which are dynamic forecast values, and the corresponding volatility risk is also dynamic. Giving a single standard deviation of historical fitting obviously cannot describe the enlargement of the confidence interval of multi-period forecast results over time (i.e., the increase in standard deviation).

[0076] It can be seen that the accuracy of volatility prediction in the prior art is low. Even if the mean prediction is very accurate, it will lead to abnormal replenishment recommendation quantity, that is, the replenishment accuracy is low, affecting business revenue. In order to solve this problem, the embodiment of the present disclosure provides a replenishment method that can improve the accuracy of replenishment.

[0077] See also Figure 1 , Figure 1 A flow chart of a replenishment method in an embodiment of the present disclosure is shown, comprising the following steps:

[0078] Step S110, obtaining and determining the historical forecast deviation of the target item at multiple different lead times based on the historical actual transaction data and historical forecast data of the target item at multiple different lead times.

[0079] Step S120, based on the historical forecast deviations of multiple different lead times, respectively establish a non-parametric model and a parametric model.

[0080] Step S130, integrating the non-parametric model and the parametric model to obtain an integrated model.

[0081] Step S140, determining the transaction fluctuation data of the target item on the target future date according to the integrated model, and sending the transaction fluctuation data to the replenishment system, so that the replenishment system determines the replenishment quantity according to the transaction fluctuation data.

[0082] The replenishment method of the disclosed embodiment generates a parametric model and a non-parametric model based on the historical forecast deviations of the target item at multiple different lead times. The parametric model and the non-parametric model are integrated to obtain an integrated model, which can make up for the shortcomings of the parametric model and the non-parametric model, thereby improving the accuracy of the prediction of transaction data volatility and providing an important basis for the accurate calculation of the intelligent replenishment quantity. Therefore, the accuracy of replenishment can be improved. In addition, with the change of time, the historical actual transaction data and historical forecast data of the target item at multiple different lead times can be dynamically changed, and accordingly, the integrated model can also be dynamically changed, so the accuracy of replenishment can be further improved.

[0083] The replenishment method of the embodiment of the present disclosure is described in more detail below.

[0084] In step S110, historical forecast deviations of the target item at multiple different lead times are determined based on historical actual transaction data and historical forecast data of the target item at multiple different lead times.

[0085] In the disclosed embodiment, the items sold by the enterprise can be divided based on SKU (Stock Keeping Unit), that is, different SKUs can be considered as different products, and the target item can be any SKU. Of course, it can also be divided in other ways, which are not limited here.

[0086] For target items, the present disclosure can predict the volatility of future transactions based on the historical transaction data of the target items. Specifically, EDW (Enterprise Data Warehouse) can call the data layer to the Spark computing platform, and extract the historical transaction data of the target items through MySQL (relational database management system) and HBASE (a distributed, column-oriented open source database). Among them, the enterprise data warehouse is a subject-oriented, integrated, non-volatile and time-varying data set used to support managers' decision-making. Spark is a fast and general computing engine designed for large-scale data processing.

[0087] The historical transaction data includes: historical actual transaction data and historical forecast data. The historical actual transaction data can be historical actual sales volume or other data that can be used to count the historical sales of items. The historical forecast data is the forecast data corresponding to the historical actual transaction data, which is also the historical mean forecast data. For different lead times, the difference between the historical actual transaction data and the historical forecast data can be used as the historical forecast deviation. The smaller the historical forecast deviation, the more accurate the mean forecast; the larger the historical forecast deviation, the less accurate the mean forecast.

[0088] Different lead times can be lead times relative to the current moment. For example, to predict the volatility of sales on September 1, 2019, the historical transaction data may be the data on August 31, 2019 and n days before (n is the longest forecast period), and the lead time is the number of days from September 1, 2019. The significance of marking the lead time is to capture the risk of sales fluctuations due to incomplete information, such as delays in entering promotional information and measurement errors, which results in the fact that marketing, weather and other information are often more accurate the closer they are to the current moment, so the risk of sales fluctuations is relatively smaller relative to the corresponding lead time forecast.

[0089] In step S120, a non-parametric model and a parametric model are established based on historical forecast deviations of multiple different lead times.

[0090] In practical applications, there may be many factors that cause fluctuations in sales forecasts, such as incomplete information, promotional factors, holiday impacts, the impact of competitive / complementary product promotions, sudden weather changes, and other factors that affect commodity demand. The present disclosure can comprehensively consider the above factors and establish parametric models and non-parametric models with the help of classical parametric statistical learning algorithms and non-parametric methods to predict the volatility of sales.

[0091] Among them, the non-parametric model does not make any assumptions about the distribution of the population. It only knows that the population is a random variable and its distribution exists (there may also be parameters in the distribution), but it is impossible to know the form of its distribution, let alone the relevant parameters of the distribution. Only under the condition of given some samples can it be inferred based on non-parametric statistical methods. The parametric model usually assumes that the population (random variable) obeys a certain distribution, which is determined by some parameters (for example, the normal distribution is determined by the mean and variance). The model built on this basis is called a parametric model.

[0092] The following is an introduction to the method of establishing non-parametric models.

[0093] In one implementation of the present disclosure, a non-parametric model can be established through kernel density estimation based on historical forecast deviations of multiple different lead times. Kernel density estimation is used in probability theory to estimate unknown density functions and is a type of non-parametric estimation. It does not require strong assumptions about the overall parameter model and is a typical data-driven non-parametric method. Given a sample set, the distribution density function of the sample set can be obtained. Of course, other related methods for establishing non-parametric models are within the scope of protection of the present disclosure.

[0094] See also Figure 2 , Figure 2 A flow chart of a method for establishing a non-parametric model by kernel density estimation in an embodiment of the present disclosure is shown, which may include the following steps:

[0095] Step S210, dividing the historical forecast deviations of multiple different lead times according to the factors affecting the transaction data, and determining the local probability distribution of the historical forecast deviations of the groups corresponding to the single factor.

[0096] In the embodiments of the present disclosure, the factors affecting transaction data are different, and the corresponding historical forecast deviations may be different. The factors affecting transaction data may include one or more of the following: forecast lead time, calendar effect, weather changes, promotional activities, etc. In order to improve the accuracy of volatility forecast, the historical forecast deviations of multiple different lead times can be divided according to a single factor to obtain multiple groups. Figure 3According to the calendar effect, the historical forecast deviation can be divided into the historical forecast deviation corresponding to holidays and the historical forecast deviation corresponding to non-holidays, and two groups can be obtained. According to the promotional activities, the historical forecast deviation can be divided into the historical forecast deviation corresponding to promotions and the historical forecast deviation corresponding to non-promotions, and two groups can be obtained.

[0097] After that, by flexibly specifying multiple kernel functions, the Bootstrap technique can be used to simulate the local probability distribution of the forecast deviation of future transaction data for different groups such as lead time, promotion, non-promotion, holiday, non-holiday, etc. Among them, Bootstrap is an important statistical method in non-parametric statistics that estimates the variability of statistics and can perform statistical interval estimation, also known as the bootstrap method.

[0098] Step S220, based on the local probability distribution of the historical prediction deviations corresponding to a single group, determine the overall deviation distribution corresponding to the group, and use the overall deviation distribution corresponding to each group as a non-parametric model.

[0099] Specifically, a single group obtained by the above division method corresponds to a factor that affects the transaction data. Based on the local probability distribution of the historical forecast deviation corresponding to each group, the overall deviation distribution can be determined. That is, each group corresponds to an overall deviation distribution, which is a sub-nonparametric model, and the collection of overall deviation distributions corresponding to each group is the nonparametric model.

[0100] For example, the Gaussian kernel is used to estimate the kernel density of the following six sample points: X = {x1 = -2.1, x2 = -1.3, x3 = -0.4, x4 = 1.9, x5 = 5.1, x6 = 6.2}. The "fitting" results of the six sample points are as follows: Figure 4 As shown. Through the Gaussian distribution of the six local points (dashed line part), the overall deviation distribution (solid line part) is obtained, from which the mean and standard deviation estimates of the same overall forecast deviation at the next moment are obtained.

[0101] The following is an introduction to the method of establishing the parameter model.

[0102] Combined with actual business scenarios, on the one hand, the average prediction accuracy under promotional activities is usually lower than that on ordinary days, so replenishment requires a correspondingly higher standard deviation. On the other hand, after the promotion, the large fluctuations in sales during the promotion period will affect the standard deviation calculation of the following ordinary days, and a high standard deviation will be input. Similar business scenarios also include long holidays (such as the Spring Festival and National Day). The sales fluctuations on these special days are quite different from those on ordinary days, and it is necessary to distinguish standard deviation models with different mechanisms / states.

[0103] Traditional time series assumes that the fluctuation range (variance) of time series variables is fixed, which is not in line with reality. The ARCH (Autoregressive conditional heteroskedasticity) model can express the variance as an autoregressive model of its own history. The ARCH model solves the problem caused by the second assumption (constant variance) of traditional econometrics on time series variables. The model is one of the econometric achievements that won the 2003 Nobel Prize in Economics. If it is expanded to the ARMA (Autoregressive moving average) model, the ARCH model is transformed into the Generalized ARCH model, that is, the generalized autoregressive conditional heteroskedasticity model.

[0104] Therefore, according to the above business scenarios and the generalized autoregressive conditional heteroskedasticity model, the present disclosure can establish a nonlinear threshold generalized autoregressive conditional heteroskedasticity model for sales forecasting. Figure 5 , Figure 5 A flow chart of a method for establishing a parameter model in an embodiment of the present disclosure is shown, which may include the following steps:

[0105] Step S510, establishing a threshold generalized autoregressive conditional heteroskedasticity model according to the generalized autoregressive conditional heteroskedasticity model and the factors affecting the transaction data, wherein the threshold variable in the threshold generalized autoregressive conditional heteroskedasticity model is determined according to the factors affecting the transaction data.

[0106] In the embodiment of the present disclosure, it is assumed that the prediction deviation is a time series r i,n , r i,n =sale i,n -forecast i,n , sale i,n Represents the historical actual transaction data corresponding to the target item i in time series n, forecast i,n Represents the historical forecast data corresponding to the target item i in the time series n. The residual mean model can be expressed as follows:

[0107] r i,n =c 1i +c 2i s n +ε i,n ,

[0108] Among them, ε i,n Represents the residual corresponding to the target item i in the time series n. The residual represents the difference between the actual value and the fitted value of the prediction deviation; ε i,n =σ i,n z n , z nis a normal distribution with a mean of 0 and a standard deviation of 1, c 1i 、c 2i is the mean model parameter; It represents the variance of target item i in time series n. ω i ,α i ,β i ,γ represents the model parameters.

[0109] s n is the state variable,

[0110] tv represents the threshold variable, such as the promotion discount ratio; d represents the threshold value. For example, if the discount ratio is 0.7 (70% off), the conditional mean of the prediction residual will vary with the promotion intensity tv.

[0111] E(r i,n |tv) indicates that r i,n expectations.

[0112] That is, the expected forecast deviation is different when the promotion intensity is higher than or lower than d. In reality, when the promotion intensity is larger, the sales volume tends to fluctuate greatly and the forecast deviation is larger. Therefore, the theoretical expectation c 2i >0.

[0113] Residual ε i,n =σ i,n z n , where z n is a normal distribution with a mean of 0 and a standard deviation of 1. Then the conditional mean E(ε i,n |I n-1 )=0, conditional variance I n-1 Represents the historical transaction data of time series 1 to n-1.

[0114] Variance of time series n The forecast deviation of time series n-1 variance and the state variable s of the time series n n The prediction variance will vary with the promotion intensity tv. In each promotion intensity area, it is a generalized autoregressive conditional heteroskedasticity model. Therefore, the threshold generalized autoregressive conditional heteroskedasticity model can be specifically:

[0115]

[0116] At the same time, in each promotion scenario, it is usually assumed that β i >0, a larger volatility variance Often accompanied by a larger variance

[0117] Among them, the advantages of the threshold generalized autoregressive conditional heteroskedasticity model include: the ability to flexibly specify multiple threshold variables (such as factors affecting sales, etc.) to meet business scenarios. For example, holiday type judgments, promotion types, etc. can be added; in addition, the prediction results under different threshold values ​​can have different performances, and the threshold generalized autoregressive conditional heteroskedasticity model can be captured by state variables and their coefficients.

[0118] Step S520, based on the historical forecast deviations of multiple different lead times, the parameters in the threshold generalized autoregressive conditional heteroskedasticity model are solved by the maximum likelihood estimation method, and the solved threshold generalized autoregressive conditional heteroskedasticity model is used as a parameter model.

[0119] The model parameters in the present disclosure can be obtained by the maximum likelihood estimation method, that is, maximizing the log-likelihood function of the joint distribution of samples. Maximum likelihood estimation is one of the parameter estimation methods. It is known that a random sample satisfies a certain probability distribution, but the specific parameters are unclear. Maximum likelihood estimation is based on the idea that a certain parameter is known to maximize the probability of the occurrence of this sample. This estimation result is a consistent estimate of the parameter, that is, maximum likelihood estimation.

[0120] In addition, the optimal threshold value d can be obtained by hyperparameter search using Grid-search technology. Grid-search is a parameter adjustment method that can try every possibility among all candidate parameter selections by looping through them, and the best performing parameter is the final result.

[0121] In step S130, the non-parametric model and the parametric model are integrated to obtain an integrated model.

[0122] In the disclosed embodiment, the models generated by the above two methods can be integrated to give full play to the advantages of the two methods and make up for their shortcomings. When integrated, the measurement of the true standard deviation can be reversed in combination with the service level of replenishment. For example, the target replenishment condition can be the service level that is closest to the replenishment result required by the business.

[0123] Optionally, the non-parametric model and the parametric model can be integrated by a stacking method (a model fusion method), and the weights of the non-parametric model and the weights of the parametric model can be determined to obtain an integrated model. That is, by adjusting the weights of the non-parametric model and the weights of the parametric model, the obtained integrated model can achieve a better effect, for example, making the replenishment result closest to the service level required by the business. Among them, the replenishment result closest to the service level required by the business may include: the replenishment rate is lower than the replenishment rate threshold, etc. The replenishment rate threshold can be set according to actual needs and is not limited here.

[0124] In step S140, the transaction fluctuation data of the target item on the target future date is determined according to the integrated model, and the transaction fluctuation data is sent to the replenishment system, so that the replenishment system determines the replenishment quantity according to the transaction fluctuation data.

[0125] In the embodiment of the present disclosure, a prediction can be made based on the actual situation of the target future date. Optionally, the factors affecting the transaction data corresponding to the target future date can be determined first, for example, the target future date is a holiday, the target item is to be promoted on that day, and so on. Afterwards, the transaction fluctuation data of the target item on the target future date is determined based on the integrated model and the factors affecting the transaction data corresponding to the target future date. And the transaction fluctuation data is sent to the replenishment system so that the replenishment system determines the replenishment quantity based on the transaction fluctuation data. It should be noted that when determining whether to replenish, the replenishment system can obtain the mean prediction result of the target item in addition to the transaction fluctuation data, and predict the sales volume of the target future date based on the mean prediction result and the transaction fluctuation data. Then refer to the existing inventory to determine the replenishment quantity. Therefore, the present disclosure can improve the accuracy of replenishment.

[0126] In addition, the standard deviation is used as a strong input for the safety stock of the replenishment formula, and different dynamic forecast values ​​will be calculated every day. At the same time, based on the existing transaction fluctuation data, if the mean of the forecast residual is not 0, it means that there is a systematic deviation in the existing method, so the existing forecast results can be flexibly corrected twice.

[0127] In one implementation of the present disclosure, the prediction residuals of the target item on multiple future dates can be determined based on the parameter model. When the mean of the prediction residuals is not 0, the mean of the prediction residuals is sent to the replenishment system, so that the replenishment system corrects the predicted transaction data for multiple future dates based on the mean of the prediction residuals. The predicted transaction data can be obtained based on the method of the embodiment of the present disclosure in combination with the mean prediction result. For example, the mean of the prediction residuals is -5, which means that the predicted transaction data may be 5 smaller than the actual result. The replenishment system can add 5 to the predicted transaction data to obtain the final predicted transaction data, thereby further improving the accuracy of the replenishment recommendation quantity.

[0128] It should be noted that the present disclosure can also send the mean of the prediction residuals to the mean prediction system, so that the mean prediction system adjusts the mean prediction result according to the mean of the prediction residuals, thereby improving the accuracy of the mean prediction and further improving the accuracy of replenishment.

[0129] The replenishment method of the disclosed embodiment can be based on the dynamic change pattern of transaction data, the transaction data mean prediction results, and consider multiple factors that affect the transaction data. The prediction results are integrated with the help of classical parameter statistical learning algorithms and non-parametric methods to give full play to the respective advantages of the two methods, realize a scalable dynamic standard deviation prediction model, and provide an important basis for the accurate calculation of intelligent replenishment quantity. The present disclosure not only gives the dynamic prediction results of the standard deviation, but also gives the consistent estimator of the mean of the prediction deviation. In the case of systematic deviations in existing methods, flexible secondary corrections can be made to the existing prediction results to further improve the accuracy of replenishment.

[0130] It should be noted that although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0131] Furthermore, in this exemplary embodiment, a replenishment device 600 is also provided, referring to Figure 6 As shown, including:

[0132] A historical forecast deviation determination module 610 is used to obtain and determine the historical forecast deviation of the target item at multiple different lead times based on the historical actual transaction data and historical forecast data of the target item at multiple different lead times;

[0133] A model building module 620, for building a non-parametric model and a parametric model based on historical forecast deviations of multiple different lead times;

[0134] A model integration module 630 is used to integrate the non-parametric model and the parametric model to obtain an integrated model;

[0135] The data sending module 640 is used to determine the transaction fluctuation data of the target item on the target future date according to the integrated model, and send the transaction fluctuation data to the replenishment system so that the replenishment system determines the replenishment quantity according to the transaction fluctuation data.

[0136] Optional, model building module, including:

[0137] The non-parametric model building unit is used to build a non-parametric model through kernel density estimation based on historical forecast deviations of multiple different lead times.

[0138] Optionally, a non-parametric model building unit, specifically used to divide the historical forecast deviations of multiple different lead times according to factors affecting the transaction data, and determine the local probability distribution of the historical forecast deviations of the group corresponding to a single factor;

[0139] Based on the local probability distribution of the historical prediction deviations corresponding to a single group, the overall deviation distribution corresponding to the group is determined, and the overall deviation distribution corresponding to each group is used as a non-parametric model.

[0140] Optionally, the model building module also includes:

[0141] A parameter model building unit is used to build a threshold generalized autoregressive conditional heteroskedasticity model according to the generalized autoregressive conditional heteroskedasticity model and factors affecting the transaction data, wherein the threshold variable in the threshold generalized autoregressive conditional heteroskedasticity model is determined according to the factors affecting the transaction data;

[0142] Based on the historical forecast deviations of multiple different lead times, the parameters in the threshold generalized autoregressive conditional heteroskedasticity model are solved by the maximum likelihood estimation method, and the solved threshold generalized autoregressive conditional heteroskedasticity model is used as the parameter model.

[0143] Optionally, the threshold generalized autoregressive conditional heteroskedasticity model is specifically:

[0144]

[0145] Among them, ε i,n =σ i,n z n , ε i,n represents the residual corresponding to the target item i in time series n, represents the variance of target item i in time series n, z n is a normal distribution with a mean of 0 and a standard deviation of 1, ω i ,α i ,β i ,γ represents the model parameter, tv represents the threshold variable, and d represents the threshold value.

[0146] Optionally, a model integration module is specifically used to integrate the non-parametric model and the parametric model through a stacking method, and determine the weight of the non-parametric model and the weight of the parametric model to obtain an integrated model.

[0147] Optionally, the data sending module includes:

[0148] The transaction fluctuation data determination unit is specifically used to determine the factors affecting the transaction data corresponding to the target future date; based on the integrated model and the factors affecting the transaction data corresponding to the target future date, the transaction fluctuation data of the target item on the target future date is determined.

[0149] Optionally, the replenishment device of the embodiment of the present disclosure further includes:

[0150] A prediction residual determination module, for determining the prediction residuals of the target item at multiple future dates based on the parameter model;

[0151] When the mean of the prediction residuals is not 0, the mean of the prediction residuals is sent to the replenishment system, so that the replenishment system corrects the predicted transaction data of multiple future dates according to the mean of the prediction residuals.

[0152] Optionally, factors affecting the transaction data include one or more of the following: forecast lead time, calendar effect, weather change, and promotional activities.

[0153] The specific details of each module or unit in the above device have been described in detail in the corresponding method, so they will not be repeated here.

[0154] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0155] In an exemplary embodiment of the present disclosure, an electronic device is also provided, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute all or part of the steps of the replenishment method in this exemplary embodiment.

[0156] Figure 7 The structure diagram of the computer system of the electronic device used to implement the embodiment of the present disclosure is shown. It should be noted that: Figure 7 The computer system 700 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0157] like Figure 7As shown, the computer system 700 includes a central processor 701, which can perform various appropriate actions and processes according to the program stored in the read-only memory 702 or the program loaded from the storage part 708 to the random access memory 703. In the random access memory 703, various programs and data required for system operation are also stored. The central processor 701, the read-only memory 702 and the random access memory 703 are connected to each other through a bus 704. An input / output interface 705 is also connected to the bus 704.

[0158] The following components are connected to the input / output interface 705: an input section 706 including a keyboard, a mouse, etc.; an output section 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a local area network (LAN) card, a modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the input / output interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed so that a computer program read therefrom is installed into the storage section 708 as needed.

[0159] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through the communication part 709, and / or installed from a removable medium 711. When the computer program is executed by the central processor 701, various functions defined in the device of the present application are executed.

[0160] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor, any one of the above-mentioned methods is implemented.

[0161] It should be noted that the computer-readable storage medium shown in the present disclosure may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, radio frequency, etc., or any suitable combination of the above.

[0162] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0163] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A replenishment method, characterized in that: The method comprises: Obtaining and determining historical forecast deviations of the target item at the multiple different lead times based on historical actual transaction data and historical forecast data of the target item at the multiple different lead times; Based on the historical forecast deviations of the multiple different lead times, a non-parametric model and a parametric model are established respectively; the non-parametric model is established by kernel density estimation based on the historical forecast deviations of the multiple different lead times; the parametric model is a nonlinear threshold generalized autoregressive conditional heteroskedasticity model for predicting the sales volume of the target item; Integrating the non-parametric model and the parametric model to obtain an integrated model; According to the integrated model, transaction fluctuation data of the target item on a target future date is determined, and the transaction fluctuation data is sent to a replenishment system, so that the replenishment system determines a replenishment quantity according to the transaction fluctuation data.

2. The method according to claim 1, characterized in that The non-parametric model is established by kernel density estimation based on the historical forecast deviations of the multiple different lead times, including: According to factors affecting the transaction data, the historical forecast deviations of the plurality of different lead times are divided, and a local probability distribution of the historical forecast deviations of the groups corresponding to a single factor is determined; Based on the local probability distribution of the historical prediction deviations corresponding to a single group, the overall deviation distribution corresponding to the group is determined, and the overall deviation distribution corresponding to each group is used as a non-parametric model.

3. The method according to claim 1, characterized in that The method of establishing a parameter model based on the historical forecast deviations of the plurality of different lead times comprises: According to the generalized autoregressive conditional heteroskedasticity model and the factors affecting the transaction data, a threshold generalized autoregressive conditional heteroskedasticity model is established, wherein the threshold variable in the threshold generalized autoregressive conditional heteroskedasticity model is determined according to the factors affecting the transaction data; Based on the historical forecast deviations of the multiple different lead times, the parameters in the threshold generalized autoregressive conditional heteroskedasticity model are solved by the maximum likelihood estimation method, and the solved threshold generalized autoregressive conditional heteroskedasticity model is used as a parameter model.

4. The method according to claim 3, characterized in that The threshold generalized autoregressive conditional heteroskedasticity model is specifically: Among them, ε i,n =σ i,n z n , ε i,n represents the residual corresponding to the target item i in time series n, represents the variance of target item i in time series n, z n is a normal distribution with a mean of 0 and a standard deviation of 1, ω i ,α i ,β i ,γ represents the model parameter, tv represents the threshold variable, and d represents the threshold value.

5. The method according to claim 1, characterized in that The step of integrating the non-parametric model and the parametric model to obtain an integrated model comprises: The non-parametric model and the parametric model are integrated by a stacking method, and the weight of the non-parametric model and the weight of the parametric model are determined to obtain an integrated model.

6. The method according to claim 1, characterized in that Determining the transaction fluctuation data of the target item on a target future date according to the integrated model includes: Determine factors affecting transaction data corresponding to the target future date; The transaction fluctuation data of the target item on the target future date is determined according to the integrated model and the factors affecting the transaction data corresponding to the target future date.

7. The method according to claim 1, characterized in that The method further comprises: Determining, based on the parameter model, forecast residuals for the target item at multiple future dates; When the mean of the prediction residuals is not 0, the mean of the prediction residuals is sent to the replenishment system, so that the replenishment system corrects the predicted transaction data of the multiple future dates according to the mean of the prediction residuals.

8. The method according to claim 2 or 3, characterized in that: The factors affecting transaction data include one or more of the following: forecast lead time, calendar effect, weather changes, and promotional activities.

9. A replenishment device, characterized in that: The device comprises: A historical forecast deviation determination module, used to obtain and determine the historical forecast deviation of the target item at the multiple different lead times based on the historical actual transaction data and historical forecast data of the target item at the multiple different lead times; A model building module, for building a non-parametric model and a parametric model based on the historical forecast deviations of the multiple different lead times; the non-parametric model is built by kernel density estimation based on the historical forecast deviations of the multiple different lead times; the parametric model is a nonlinear threshold generalized autoregressive conditional heteroskedasticity model for predicting the sales volume of the target item; A model integration module, used for integrating the non-parametric model and the parametric model to obtain an integrated model; A data sending module is used to determine the transaction fluctuation data of the target item on a target future date according to the integrated model, and send the transaction fluctuation data to the replenishment system so that the replenishment system determines the replenishment quantity according to the transaction fluctuation data.

10. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to perform the method according to any one of claims 1 to 8 by executing the executable instructions.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images