JP7481382B2 - Information processing device, method, and computer program - Google Patents

Description

The technology disclosed in this application relates to an information processing device, method, and computer program used in logistics.

JP 2020-38535 A (Patent Document 1) shows an overall picture of logistics in Figure 8. Looking at this overall picture from the perspective of a wholesaler, the wholesaler places an order for a product with a manufacturer that produces the product, receives the product from the manufacturer, and stores the product as inventory. The wholesaler also receives an order for the product from a retailer, and ships the product stored as inventory to the retailer.

JP 2020-38535 A

In general, it is often desirable for wholesalers to reduce the inventory of a product as much as possible while reducing the probability of the occurrence of an event in which the product runs out of stock (stock-out). To achieve this, it is desirable to predict the number of products that will be shipped from the wholesaler to the retailer in response to an order from the retailer.

Therefore, the technology disclosed in this application provides an information processing device, method, and computer program for predicting the number of products to be shipped from a wholesaler to a retailer.

An information processing device according to one embodiment is "an information processing device having at least one processor, used in association with a wholesaler that purchases goods from manufacturers and ships them to retailers, and generating information regarding the number of goods to be shipped from the wholesaler to the retailer on a target date, wherein the at least one processor acquires as input information for the goods, the category of the goods, the number of goods shipped from the wholesaler to the retailer from one day before to seven days before the target date, and the number of days of inventory, which is a value obtained by dividing the number of goods in stock one day before the target date by the average sales number, which is an average of the number of goods sold per day at the retailer, or the average sales number of the goods, and inputs the input information into a learning model generated by performing supervised learning, thereby causing the learning model to output predictive information regarding the number of goods to be shipped from the wholesaler to the retailer on the target date."
A method according to one embodiment can be "a method executed by at least one processor executing computer readable instructions for use in association with a wholesaler that receives merchandise from a manufacturer and ships it to a retailer, and for generating information regarding the number of the merchandise to be shipped from the wholesaler to the retailer on a target date, wherein the at least one processor executes the instructions to obtain, as input information for the merchandise, a category of the merchandise, the number of the merchandise shipped from the wholesaler to the retailer from one day prior to the target date through seven days prior to the target date, and a number of days of inventory, which is a value obtained by dividing the number of inventory of the merchandise one day prior to the target date by an average sales number, which is an average of the number of sales of the merchandise per day at the retailer, or the average sales number of the merchandise, and inputting the input information into a learning model generated by performing supervised learning, thereby causing the learning model to output predictive information regarding the number of the merchandise to be shipped from the wholesaler to the retailer on the target date."
A computer program according to one embodiment can be used in association with a wholesaler that purchases products from manufacturers and ships them to retailers, and can be executed by at least one processor to generate information regarding the number of the products to be shipped from the wholesaler to the retailer on a target date, thereby causing the at least one processor to function as follows: obtain, as input information, for the products, a category of the products, the number of the products shipped from the wholesaler to the retailer from one day before to seven days before the target date, and a number of days of inventory, which is a value obtained by dividing the number of products in stock one day before the target date by an average sales number, which is an average of the number of products sold per day at the retailer, or the average sales number of the products, and input the input information into a learning model generated by performing supervised learning, thereby causing the learning model to output predictive information regarding the number of the products to be shipped from the wholesaler to the retailer on the target date."

FIG. 1 is a block diagram illustrating an example of an overview of a prediction system according to an embodiment. FIG. 2 is a block diagram showing an example of the hardware configuration of each of one or more mutually connectable information processing devices constituting the prediction device 100 (retail system 200, wholesale system 300) included in the prediction system 1. As shown in FIG. FIG. 3 is a block diagram showing an example of functions of the prediction device 100 included in the prediction system 1 shown in FIG. FIG. 4 is a diagram showing an example of information stored in the storage unit 130 of the prediction device 100 included in the prediction system 1 shown in FIG. FIG. 5 is a diagram showing another example of information stored in the storage unit 130 of the prediction device 100 included in the prediction system 1 shown in FIG. FIG. 6 is a flow diagram showing an example of an operation related to generation of a learning model performed by the prediction device 100 included in the prediction system 1 shown in FIG. FIG. 7A is a diagram for explaining generation of a learning model and prediction using the generated learning model performed by the prediction device 100 included in the prediction system 1 shown in FIG. FIG. 7B is a diagram for explaining generation of a learning model and prediction using the generated learning model performed by the prediction device 100 included in the prediction system 1 shown in FIG. FIG. 8 is a flow diagram showing an example of an operation related to prediction performed by the prediction device 100 included in the prediction system 1 shown in FIG. FIG. 9 is a diagram for explaining a learning model generated by the prediction device 100 included in the prediction system 1 shown in FIG. 1 to obtain the total sales volume used to calculate the inventory days of a product. FIG. 10 is a block diagram illustrating an example of an outline of an order point calculation system according to an embodiment. FIG. 11 is a block diagram showing an example of functions of an order point calculation device 500 included in the order point calculation system 2 shown in FIG. FIG. 12 is a diagram showing an example of information stored in the storage unit 530 of the order point calculation device 500 included in the order point calculation system 2 shown in FIG. FIG. 13 is a flow diagram showing an example of the overall operation performed by the order point calculation device 500 included in the order point calculation system 2 shown in FIG. FIG. 14 is a flow diagram showing an example of an optimization calculation performed by the order point calculation device 500 included in the order point calculation system 2 shown in FIG. FIG. 15 is a flow diagram showing an example of a simulation performed at the time of optimization calculation by the order point calculation device 500 included in the order point calculation system 2 shown in FIG.

Various embodiments of the present invention will now be described with reference to the accompanying drawings. Note that common components in the drawings are given the same reference numerals. It should also be noted that components depicted in one drawing may be omitted in another drawing for ease of explanation. Furthermore, it should also be noted that the accompanying drawings are not necessarily drawn to scale.

1. Overview of the Prediction System Fig. 1 is a block diagram showing an example of an overview of a prediction system according to an embodiment. As shown in Fig. 1, the prediction system 1 may mainly include a prediction device 100, a retail system 200 that may be operated and managed by a retailer, and a wholesale system 300 that may be operated and managed by a wholesaler. The prediction device 100, the retail system 200, and the wholesale system 300 may be connected to each other via a network 400.

The network 400 may include, but is not limited to, a mobile phone network, a wireless LAN, a landline telephone network, the Internet, an intranet, and/or Ethernet (registered trademark), etc.

In the example shown in FIG. 1, the prediction device 100 is provided independently of the retail system 200 and the wholesale system 300, but it is also possible for the prediction device 100 or a part of it to be included in the retail system 200, the wholesale system 300, and/or other devices.

In the example shown in FIG. 1, the prediction device 100 is provided as a single device, but it may be composed of multiple devices that operate in cooperation with each other. In this case, some of the multiple devices that make up the prediction device 100 may also be included in the retail system 200, the wholesale system 300, and/or other devices.

In brief, in the prediction system 1 shown in FIG. 1, the prediction device 100 can predict information regarding the number of products shipped from a wholesaler to a retailer using a learning model, using only the information stored by the wholesale system 300, or using the information stored by the wholesale system 300 and the information stored by the retail system 200. The prediction result by the prediction device 100 (e.g., information regarding the number of products shipped from a wholesaler to a retailer) can be provided to the wholesale system 300 and/or the retail system 200.

In this application, "goods" refers to any item that can be bought and sold, including (but not limited to) food, clothing, stationery, beverages, books, cosmetics, medicines, iron ore, petroleum, automobile parts, and electrical products.

In addition, in this application, "retailer" refers to any business that provides goods to consumers, and can include not only businesses that exclusively sell goods to consumers, but also businesses that manufacture goods insofar as they sell them to consumers (e.g., parts manufacturers).

1-1. Retail system 200
The retail system 200 may be configured by one or more information processing devices (not shown) that can be connected to each other. The hardware configuration of each information processing device is similar to the hardware configuration of each of the one or more information processing devices that configure the prediction device 100. This hardware configuration will be described later in "1-3(1)".

The retail system 200 is, for example, a system owned by a retailer, and can store various information including sales history information, inventory history information, and sale information. At least a portion of this information can be provided to and stored in the prediction device 100 at any time or periodically. Details of this information will be explained later in "1-3(2)". The retail system 200 does not necessarily have to be operated and managed by the retailer, and may be operated and managed by a company related to the retailer or another company.

1-2. Wholesale system 300
The wholesale system 300 may be configured by one or more information processing devices (not shown) that can be connected to each other. The hardware configuration of each information processing device is the same as the hardware configuration of each of the one or more information processing devices constituting the prediction device 100. This hardware configuration will be described later in "1-3(2)". Note that the wholesale system 300 does not necessarily need to be operated and managed by the wholesaler, and may be operated and managed by a company related to the wholesaler or another company.

The wholesale system 300 is, for example, a system owned by a wholesaler, and can store various information including receipt/shipment history information and product information. At least a portion of this information can be provided to and stored in the prediction device 100 at any time or periodically. Details of this information will be explained later in "1-3(2)".

1-3. Prediction device 100
The prediction device (information processing device) 100 may be operated and managed by any company. The any company may include trading companies, wholesalers, retailers, manufacturers, companies related to these, and other companies.
(1) Hardware configuration
The prediction device 100 may be configured by one or more information processing devices (not shown) that can be connected to each other. Fig. 2 is a block diagram showing an example of a hardware configuration of each of one or more information processing devices that can be connected to each other and that configure the prediction device 100 (retail system 200, wholesale system 300) included in the prediction system 1.

As shown in FIG. 2, each information processing device can mainly include a central processing unit 11, a main memory device 12, an input/output interface device 13, an input device 14, an auxiliary memory device 15, and an output device 16. These devices are connected to each other by a data bus and/or a control bus.

The central processing unit 11 is called a "CPU" and can perform calculations on the instructions and data stored in the main memory device 12 and store the results of the calculations in the main memory device 12. Furthermore, the central processing unit 11 can control the input device 14, the auxiliary memory device 15, the output device 16, etc. via the input/output interface device 13. Each information processing device can include one or more such central processing units 11.

The main memory device 12 is referred to as "memory" and can store instructions and data received from the input device 14, the auxiliary memory device 15, and the network 400 via the input/output interface device 13, as well as the results of calculations by the central processing unit 11. The main memory device 12 can include, but is not limited to, RAM (random access memory), ROM (read only memory), and/or flash memory.

The auxiliary storage device 15 is a storage device having a larger capacity than the main storage device 12. The auxiliary storage device can store instructions and data (computer programs) that constitute an operating system and specific applications, etc. These specific applications can be executed by each information processing device, causing each information processing device as a whole to function as the prediction device 100. Furthermore, the auxiliary storage device 15 can transmit these instructions and data (computer programs) to the main storage device 12 via the input/output interface device 13 by being controlled by the central processing device 11. The auxiliary storage device 15 can include, without being limited to, a magnetic disk device and/or an optical disk device, etc.

The input device 14 is a device that inputs data from the outside and can include, but is not limited to, a touch panel, buttons, a keyboard, and/or a mouse.

The output device 16 may include, but is not limited to, a display device, a touch panel, and/or a printer device.

In such a hardware configuration, the central processing unit 11 can sequentially load instructions and data (computer programs) constituting a specific application stored in the auxiliary storage device 15 into the main storage device 12 and perform operations on the loaded instructions and data. This allows the central processing unit 11 to control the output device 16 via the input/output interface device 13, and also to send and receive various information to and from other devices (e.g., the retail system 200 and/or the wholesale system 300) via the input/output interface device 13 and the network 400.

In addition, each information processing device may include one or more microprocessors and/or a graphics processing unit (GPU) instead of or in addition to the central processing device 11.

The hardware configuration of each of the one or more information processing devices constituting the retail system 200 and/or the wholesale system 300 may also be similar to the hardware configuration of each of the one or more information processing devices constituting the prediction device 100 described above with reference to FIG. 2.

(2) Functions FIG. 3 is a block diagram showing an example of functions of the prediction device 100 included in the prediction system 1 shown in FIG. 3. As shown in FIG. 3, the prediction device 100 can mainly include a communication unit 110, a control unit 120, and a storage unit 130.

The communication unit 110 is connected to the network 400 and can transmit and receive various information between the retail system 200 and/or the wholesale system 300.

The control unit 120 can control and execute the processes necessary to generate information regarding the number of products shipped from the wholesaler to the retailer. To achieve this, the control unit 120 can mainly include an acquisition unit 122, a generation unit 124, a prediction unit 126, and a provision unit 128.

The acquisition unit 122 can acquire information necessary to generate information regarding the number of products shipped from the wholesaler to the retailer from the retail system 200 and/or the wholesale system 300 and store it in the memory unit 130.

The generation unit 124 can generate a learning model using the information acquired by the acquisition unit 122 (stored in the storage unit 130) and store the model in the storage unit 130.

The prediction unit 126 can generate information regarding the number of products to be shipped from the wholesaler to the retailer using the information acquired by the acquisition unit 122 (stored in the memory unit 130) and the learning model generated by the generation unit 124 (stored in the memory unit 130).

The providing unit 128 can transmit the information generated by the prediction unit 126 to the wholesale system 300.

The storage unit 130 can store information necessary to generate information regarding the number of products shipped from wholesalers to retailers. To achieve this, the storage unit 130 can include a retail information storage unit 132, a wholesale information storage unit 134, a model information storage unit 136, and a provision information storage unit 138.

The retail information storage unit 132 can store information about retailers received from the retail system 200. FIG. 4 is a diagram showing an example of information stored by the storage unit 130 of the prediction device 100 included in the prediction system 1 shown in FIG. 1. As illustrated in FIG. 4, the retail information storage unit 132 can store the wholesaler's identification information (center ID) and the date on which the tally was performed (tally date) in association with the retailer's identification information (store ID).

Furthermore, the retail information storage unit 132 can store sales history information, inventory history information, sale information, etc., in association with the retail store's identification information (store ID). The retail information storage unit 132 can store product identification information (product ID), the product's category, and sales of the product, etc., in association with the retail store's identification information (store ID), as sales history information.

Furthermore, the retail information storage unit 132 can store, as inventory history information, the number of items in stock (item identified by the item ID in the sales history information), the number of items sold, the number of items received, and the number of items discarded, in association with the retail store's identification information (store ID). Furthermore, the retail information storage unit 132 can store, as special sale information, the date on which a special sale began (start date) for the item (item identified by the item ID in the sales history information), and the date on which the special sale ended (end date).

Next, the wholesale information storage unit 134 can store information about wholesalers received from the wholesale system 300. FIG. 5 is a diagram showing another example of information stored by the storage unit 130 of the prediction device 100 included in the prediction system 1 shown in FIG. 1. As illustrated in FIG. 5, the wholesaler's identification information (center ID), the date on which the tally was performed (tally date), etc. can be stored in association with the retailer's identification information (retailer ID).

Furthermore, the wholesale information storage unit 134 can store receipt/shipment history information and product information, etc., in association with the retailer's identification information (retailer ID). The wholesale information storage unit 134 can store, as receipt/shipment history information, the number of products received and the number of products shipped, etc., in association with the retailer's identification information (retailer ID).

Furthermore, the wholesale information storage unit 134 can store, as product information, the identification information of the supplier of the product (supplier ID), the category of the product, the lot size of the product, and the date on which sales of the product ended (last sale date), etc.

Returning to FIG. 3, the model information storage unit 136 can store the learning model etc. generated by the generation unit 124.

The provided information storage unit 138 can store information generated by the prediction unit 126 and provided to the wholesale system 300.

2. Behavior of Prediction System 1 The prediction device 100 included in the prediction system 1 can generate a learning model and then execute prediction using the thus generated learning model.

2-1. Generation of Learning Model FIG. 6 is a flow diagram showing an example of an operation related to generation of a learning model performed by the prediction device 100 included in the prediction system 1 shown in FIG.

First, in step (hereinafter referred to as "ST") 602, the acquisition unit 122 in the control unit 120 of the prediction device 100 can acquire retail information relating to past retailing (e.g., the information shown in FIG. 4, etc.) from the retail system 200 and store it in the retail information storage unit 132, and can acquire wholesale information relating to past wholesale (e.g., the information shown in FIG. 5, etc.) from the wholesale system 300 and store it in the wholesale information storage unit 134.

Next, in ST604, the generation unit 124 of the prediction device 100 can generate a learning model based on the retail information and wholesale information acquired in ST602.

Figures 7A and 7B are diagrams for explaining the generation of a learning model performed by the prediction device 100 included in the prediction system 1 shown in Figure 1 and prediction using the generated learning model. In Figures 7A and 7B, the "Learning" column shows examples of information used as training data when generating a learning model, and the "Prediction" column shows examples of information used as input information when making a prediction using the learning model.

For simplicity, we will focus here only on the case where a learning model is generated using the information shown in the "Required parameters" row. "Change of objective variable" and "Options" will be described later with reference to Figures 7A and 7B.

The generation unit 124 of the prediction device 100 can generate a learning model by performing supervised learning. Specifically, as shown in Fig. 7A, the generation unit 124 can generate a learning model by using the following information as teacher data (question data and answer data).
(1) Category of a certain product (question data)
(2) The number of items shipped from the wholesaler to the retailer one day before the specific date (today) (question data)
(3) The number of items shipped from the wholesaler to the retailer two days prior to a specific date (question data)
(4) The number of items shipped from the wholesaler to the retailer three days prior to a specific date (question data)
(5) The number of items shipped from the wholesaler to the retailer four days prior to a specific date (question data)
(6) The number of items shipped from the wholesaler to the retailer five days prior to a specific date (question data)
(7) The number of items shipped from the wholesaler to the retailer six days prior to a specific date (question data)
(8) The number of products shipped from the wholesaler to the retailer seven days prior to a specific date (question data)
(9) The number of days the product is in stock or the average number of sales of the product (question data)
(10) The number of products shipped from the wholesaler to the retailer on a specific day (the same day) (answer data)

For the above (1), the generation unit 124 can use the "category" in the "product information" stored in the wholesale information storage unit 134. Since the characteristics differ depending on the product category, such as sweets or ice cream, information about the category is essential. Note that the information about the category can be processed using a technique called One Hot Encoding, as an example.

For the above (2) to (8), the generation unit 124 can use the "shipment volume" from one day to seven days before the specific date in the "shipment history information" stored in the wholesale information storage unit 134. It is important to look at the weekly shipment volume of the product. Also, in order to grasp the trend by day of the week, the shipment volume of the product for the past week is essential.

For the above (10), the generation unit 124 can use the "number of shipments" for a specific day in the "shipment history information" stored in the wholesale information storage unit 134. This number of shipments for a specific day is sometimes called the objective variable.

Regarding (9) above, the number of days that a product is in stock is the number of products in stock at a retail store linked to the wholesaler one day before a specific date divided by the average number of products sold per day. The number of days that a product is in stock can be obtained, for example, by the following method. The average number of products sold can also be used as a parameter to replace the number of days that a product is in stock. The average number of products sold is the average number of products sold at a retail store linked to the wholesaler over a certain period of time. The period for calculating the average can be determined arbitrarily.

(1) Statistical Processing Method The generation unit 124 can calculate the number of days of inventory of the product by using the inventory quantity and sales quantity in the "inventory history information" (FIG. 4) stored in the retail information storage unit 132, and by store and product, by "inventory quantity/average daily sales quantity." Here, the period targeted for calculating the "average daily sales quantity" (total sales quantity in a certain period/total number of days included in the period) may overlap with at least one of the days corresponding to (2) to (8) above, or may be different from the at least one day. In addition, the period may be any period including one day, one week, one month, three months, half a year, one year, two years, etc., and may be appropriately determined depending on the corresponding product.

(2) Machine Learning Method The inventory days of a product can be calculated using a learning model generated by supervised learning. Fig. 9 is a diagram for explaining a learning model generated by the prediction device 100 included in the prediction system 1 shown in Fig. 1 to obtain a total value of sales volume used to calculate the inventory days of a product.

First, the prediction device 100 can generate a learning model using the required parameters shown in FIG. 9 (the parameters shown in the "optional" row can also be used as additional parameters). The learning model can be generated by supervised learning using gradient boosting trees, linear regression, or neural networks in a similar manner to the generation of the learning model shown in FIG. 4.

Next, the prediction device 100 can obtain the total number of sales for that product over n days (output from the learning model as the objective variable) by making a prediction using the generated learning model. Furthermore, the prediction device 100 can calculate a predicted value for the average daily sales by calculating "obtained total sales over n days/n". Using the inventory quantity in the "inventory history information" (Figure 4) stored in the retail information storage unit 132 and the calculated predicted value for the average daily sales, the number of days the product will be in stock can be calculated by "inventory quantity/predicted value of average daily sales".

In addition, instead of calculating this product inventory days, the generation unit 124 can also receive this product inventory days from the retail system 200 and use it.

The generation unit 124 can generate a learning model by learning using the above (1) to (10) as training data.

In one embodiment, the generating unit 124 can generate a learning model using a gradient boosting tree. The generating unit 124 creates a decision tree and adds it to the learning model so as to improve the objective function calculated from the objective variable and the predicted value. The generating unit 124 repeats this process for the number of decision trees determined by the hyperparameter. For the second and subsequent trees, the generating unit 124 can learn about the difference between the objective variable and the prediction by the decision tree created up to that point. In this way, the generating unit 124 can determine the branching and leaf weights of each decision tree. That is, in the gradient boosting tree, each decision tree uses the question data in the teacher data as input data and the value of the leaf node reached by tracing each branch of the decision tree as output data (objective variable). Furthermore, the method of each branch of the decision tree and the value of each leaf node can be optimized so that the error between this data (objective variable) and the answer data (objective variable) in the teacher data (however, for the second and subsequent trees, the difference between the answer data (objective variable) and the predicted value by the decision tree created up to that point) is reduced.

In another embodiment, the generation unit 124 can generate a learning model using linear regression. In this linear regression, the generation unit 124 can learn the parameters of the linear regression model so as to improve the error between the objective variable and the predicted value. That is, the linear regression model is divided into an input layer and an output layer, and the parameters included in the linear regression model can be optimized using the least squares method so as to reduce the error between the data (objective variable) output from the output layer by inputting question data of the teacher data as input data into the input layer and the answer data (objective variable) of the teacher data.

In yet another embodiment, the generation unit 124 can use a neural network including an input layer, an intermediate layer, and an output layer. Specifically, the generation unit 124 can optimize (learn) each parameter (weight) included in the neural network using an error backpropagation method so as to reduce the error between the data (objective variable) output from the output layer by inputting question data of the teacher data into the input layer and the answer data (objective variable) of the teacher data.

Note that, as a preferred embodiment, a case has been described in which the prediction device 100 itself generates a learning model. However, instead of generating a learning model by itself, the prediction device 100 can also acquire a learning model generated by another device using a similar technique via the network 400 and store the acquired learning model in the model information storage unit 136.

2-2. Prediction FIG. 8 is a flow diagram showing an example of an operation related to prediction performed by the prediction device 100 included in the prediction system 1 shown in FIG.

First, in ST702, the acquisition unit 122 in the control unit 120 of the prediction device 100 can acquire information to be input to the learning model that has already been generated (or acquired from another device via the network 400) from the retail system 200 and the wholesale system 300. For example, in the example shown in FIG. 7A, the required parameters listed in the "Prediction" column need to be input to the learning model, so these required parameters can be acquired.

In ST704, the prediction unit 126 of the prediction device 100 can input the information acquired by the acquisition unit 122 in ST702 to the learning model. For example, in the example shown in FIG. 7A, the prediction unit 126 can input (a) the product category, (b) the number of products shipped one day before the target date to the number of products shipped seven days before the target date, and (c) the product inventory days of the product to the learning model. Note that the product inventory days in (c) above is a value obtained by dividing the product inventory number one day before the target date by the average sales number, which is the average sales number of products per day at the retailer, as described above. Also, as described above, it is possible to use the average sales number of products instead of the product inventory days.
In addition, as in the case of learning, in the case of prediction, the period used to calculate (c) the number of days of product inventory and the target date corresponding to (b) the number of products shipped may overlap or may be different.

As a result, in ST706, the prediction unit 126 can obtain the number of shipments (the number of products shipped from the wholesaler to the retailer on the target date) output from the learning model as the objective variable.

Furthermore, in ST708, the provision unit 128 can provide the shipment quantity predicted by the prediction unit 126 in ST706 to the wholesale system 300.

3. Modifications of Prediction System 1 In the above-described embodiment, the number of shipments of the product on the day is used as the objective variable during learning and prediction. However, in another embodiment, as shown in the row of "Change 1 of Objective Variable" in FIG. 7A, a value calculated by "Number of shipments on the day / average number of shipments for X months up to the previous day" can be used as the objective variable during learning and prediction. Here, X can be set to any period.

As an example, if the target date (today) is May 1, 2020, and the average number of shipments for the past three months is used, the calculation can be made as follows: "number of shipments on May 1, 2020 / average number of shipments for the period from February 1, 2020 to April 30, 2020."

This means the percentage change in the number of shipments of that product on that day from the average number of shipments in the past (i.e., the percentage change in the number of products shipped from the wholesaler to the retailer on the target day relative to the average number of products shipped from the wholesaler to the retailer up to the previous day, which is the dependent variable).

By using such a response variable, the accuracy of predictions can be improved when there is a large difference in the average sales volume between products.

In yet another embodiment, as shown in the "Change in Objective Variable 2" row in Figure 7A, the number of items shipped from the wholesaler to the retailer on each of multiple target dates can be used as the objective variable during learning and prediction.

In yet another embodiment, as shown in the "Change in Objective Variable 3" row in Figure 7A, the total number of products shipped from the wholesaler to the retailer on all of the multiple target dates (N days) can be used as the objective variable during learning and prediction.

In the above-described embodiment, the number of product shipments from 1 day to 7 days before the target date is used as a required parameter during learning and prediction. However, in another embodiment, in addition to this, as shown in the "Option 1" row in Figure 7A, the number of product shipments T+7 days before and the total number of product shipments for U weeks counting from T weeks before can be used as additional parameters (input data). Here, T can be defined as any natural number equal to or greater than 1. U is any natural number that satisfies T ≥ U.

In another embodiment, as shown in the "Option 2" row in FIG. 7A, information on the shipping difference can be used as an additional parameter. For example, the number of days elapsed from the last shipping date to the target date and/or the difference between the total number of shipped items (T-1) weeks ago and the total number of shipped items T weeks ago can be used as an additional parameter (input data) for the shipping difference information. Here, T can be defined as any natural number equal to or greater than 2.

In addition, in another embodiment, as additional parameters (input data), the average, variance, and median of the number of products shipped over the last X days can be used as parameters, as shown in the "Option 3" row of Figure 7B, where X can be any natural number.

In yet another embodiment, an additional parameter (input data) can be a value obtained by applying a moving window function of 7, 14, 28, 56 or 128 days to the total number of product shipments over the manufacturer's lead time, as shown in the "Option 4" row of Figure 7B.

Here, let's say that order A was placed with the manufacturer on May 1, 2020. If the lead time is three days, then calculate A, the total number of products shipped between May 1 and May 4 of the same year. Similarly, let's say that order B was placed on May 2 of the same year. If the lead time is three days, then calculate B, the total number of products shipped between May 2 and May 5 of the same year. Repeat the process in the same way for order C on May 3 to order G on May 7. In this case, the sum of totals A through G is "the value obtained by applying a seven-day moving window function to the total number of products shipped over the manufacturer's lead time."

Furthermore, in another embodiment, as an additional parameter (input data), a statistical value calculated for a group (sum, average, median of shipments of items in the group) can be used, as shown in the "Option 5" row of FIG. 7B. A group can include a center (wholesaler) or a category of items. It is also possible to use such a statistical value calculated per center (wholesaler), per category of items, or per item.

Furthermore, in another embodiment, as an additional parameter (input data), it is also possible to use a value obtained by converting the number of days from January 1st to the target date/365 into Sin or Cos to capture the annual cycle of the number of product shipments, as shown in the "Option 6" row in Figure 7B.

により計算することができる。 The above converted values are, for example,

It can be calculated as follows:

Furthermore, in another embodiment, as an additional parameter (input data), it is also possible to use a value obtained by converting the number of days from Monday to the target day/7 into Sin or Cos, in order to capture the cycle of days of the week related to the number of product shipments, as shown in the "Option 7" row in Figure 7B.

により計算することができる。 The above converted values are, for example,

It can be calculated as follows:

Furthermore, in another embodiment, the presence or absence of a sale and the category of the sale can be used as additional parameters (input data), as shown in the "Option 8" row of Figure 7B.

As described above, the prediction device 100 disclosed in the present application can use information about retail sales and inventory of a product to more accurately predict the number of that product that will be shipped from a wholesaler to a retailer.

4. Overview of the order point calculation system Fig. 10 is a block diagram showing an example of an overview of an order point calculation system according to an embodiment. As shown in Fig. 10, the order point calculation system 2 can further include an order point calculation device 500 connectable to the prediction device 100, retail system 200, and wholesale system 300 included in the above-mentioned prediction system 1 (Fig. 1) via a network 400. The prediction device 100, retail system 200, wholesale system 300, and network 400 are as described above by calculating Fig. 1 etc., so detailed description thereof will be omitted.

In FIG. 10, as an example, the order point calculation device 500 is provided independently of the retail system 200, the wholesale system 300, and the prediction device 100, but the order point calculation device 500 or a part thereof may be included in the prediction device 100, the retail system 200, the wholesale system 300, and/or other devices.

In the example shown in FIG. 10, the order point calculation device 500 is provided as a single device, but it may be configured from multiple devices that operate in cooperation with each other. In this case, some of the multiple devices that configure the order point calculation device 500 may be included in the prediction device 100, the retail system 200, the wholesale system 300, and/or other devices.

The order point calculation device 500 can calculate an order point for a product. The order point is a threshold for the inventory quantity of a product at a wholesaler, and means a threshold used to place an order for that product with a manufacturer when the inventory quantity falls below this threshold. The order point calculation device 500 can provide the calculated order point to the wholesale system 300 via the network 400. In this way, the wholesale system 300, which is operated and managed by the wholesaler, can order that product from the manufacturer when the inventory quantity of that product falls below the order point for that product provided by the order point calculation device 500.

4-1. Order point calculation device 500
The order point calculation device (information processing device) 500 may be operated and managed by any company. The arbitrary company may include a trading company, a wholesaler, a retailer, a manufacturer, a company related to these, and other companies.
(1) Hardware configuration
The order point calculation device 500 may be configured by one or more information processing devices (not shown) that can be connected to each other, similar to the prediction device 100. Each information processing device may have the hardware configuration exemplified in FIG.

(2) Functionality FIG. 11 is a block diagram showing an example of functions of an order point calculation device 500 included in the order point calculation system 2 shown in FIG. 10. As shown in FIG. 11, the order point calculation device 500 can mainly include a communication unit 510, a control unit 520, and a storage unit 530.

The communication unit 510 is connected to the network 400 and can transmit and receive various information between the prediction device 100, the retail system 200, and/or the wholesale system 300.

The control unit 520 can control and execute the processes required to calculate the order points for products used by wholesalers. To achieve this, the control unit 520 can mainly include an acquisition unit 522, a calculation unit 524, and a provision unit 526.

The acquisition unit 522 can acquire information necessary to calculate the order point for products used by wholesalers from the prediction device 100, the retail system 200, and/or the wholesale system 300, and store the information in the memory unit 530.

The calculation unit 524 can calculate the order point for a product using the information acquired by the acquisition unit 522 (and stored in the memory unit 530) and store it in the memory unit 530.

The providing unit 526 can transmit the order points and other information regarding the products calculated by the calculation unit 524 (and stored by the memory unit 530) to the wholesale system 300.

The storage unit 530 can store information necessary to calculate the order point for products used by wholesalers. To achieve this, the storage unit 530 can include a shipping forecast information storage unit 532 and a provided information storage unit 534.

The shipping forecast information storage unit 532 can store shipping forecast information etc. received from the prediction device 100.

Figure 12 is a diagram showing an example of information stored in the storage unit 530 of the order point calculation device 500 included in the order point calculation system 2 shown in Figure 10. As shown in Figure 12, the shipping forecast information storage unit 532 can include, for example, wholesaler identification information (center ID), manufacturer information, product information, center information, and shipping forecast information in association with product identification information (product ID).

The manufacturer information is information about the manufacturer that produces the product identified by the product ID, and may include, for example, information about the manufacturer lead time (the time it takes from when a wholesaler places an order for a product with a manufacturer until the product is received from the manufacturer) and non-delivery days.

The product information is information that indicates the lot number for the product identified by the above product ID.

The center information may include information indicating the wholesaler's non-delivery days for the product identified by the product ID.

The shipment prediction information may include information obtained from the prediction device 100. For example, the shipment prediction information may store the number of products identified by the product ID to be shipped from the wholesaler to the retailer on the target date (one day later) (the number of product shipments predicted by the prediction device 100), the standard deviation of the number of shipments, etc.

The provided information storage unit 534 can store the order points and other information related to the products calculated by the calculation unit 524.

5. Operation of the Order Point Calculation System FIG. 13 is a flow diagram showing an example of the overall operation performed by the order point calculation device 500 included in the order point calculation system 2 shown in FIG.

Here, we will explain the operations performed by the order point calculation device 500 when calculating the order point for product P at a wholesaler on June 1, 2020, which is the target date.

First, in ST802, the order point calculation device 500 can obtain the number of shipments of product P on the day before June 1, 2020, which is the target date, predicted by the prediction device 100 (for example, "number of shipments one day later" in FIG. 12).

Next, in ST804, the order point calculation device 500 can calculate optimal parameters by performing an optimization calculation that minimizes the inventory quantity and the number of out-of-stocks using the number of shipments of product P on the day before the target date obtained in ST802. The order point calculation device 500 can calculate the order point for product P on the target date using the parameters. Details of the optimization calculation will be described later with reference to Figures 14 and 15.

Finally, in ST806, the order point calculation device 500 can transmit the order point calculated in ST804 to the wholesale system 300.

Next, a specific example of the operation performed by the order point calculation device 500 in ST804 will be described with reference to Figs. 14 and 15. Fig. 14 is a flow diagram showing an example of an optimization calculation performed by the order point calculation device 500 included in the order point calculation system 2 shown in Fig. 10. Fig. 15 is a flow diagram showing an example of a simulation performed during optimization calculation by the order point calculation device 500 included in the order point calculation system 2 shown in Fig. 10.

Referring to FIG. 14, in ST902, the order point calculation device 500 can obtain data for the period for calculating the optimal parameters. Here, as described above, the target date is June 1, 2020, so the order point calculation device 500 can obtain data for any period before this target date, for example, the four-month period from February 1 to May 31, 2020 (note that data before February 1 may also be obtained). The data for this four-month period may include, for example, the number of products P in stock at the wholesaler and the number of products P shipped from this wholesaler to retailers for each day (each reference date) included in this period. This data is receivable by the order point calculation device 500 from the wholesale system 300 via the network 400.

Next, in ST904, the order point calculation device 500 can generate multiple parameter candidates.

In one embodiment, the number of shipments of product P at a wholesaler is modeled using a Gaussian distribution. Since the mean and variance are unknown, a Gaussian-Gamma distribution is assumed as the prior distribution of the mean and variance. The variance and standard deviation of the predicted distribution obtained by performing a Bayesian update on this prior distribution using the past shipments of product P are calculated.

For example, any period of the above four-month period (here, for example, May 1st to May 31st) can be set as the period to be simulated (hereinafter referred to as the "simulation period"). First, the standard deviation corresponding to each day included in the simulation period is calculated as follows.

と定める。
シミュレーション期間のうち、例えば５月１日に対応する標準偏差を算出する場合（過去が2020年2月1日～同年4月30日という3ヶ月の期間のデータの場合）、過去の出荷の回数とは上記3か月の期間に商品Pが出荷された回数を指す。商品Pが出荷された回数は、商品Pの出荷が発生した日数に一致し、出荷数を意味しない。上記3か月の期間において2020年2月1日及び、同年2月2日、及び同年2月3日のみ出荷を行なった場合は、nは3となる。 Specifically, the variance of a Gaussian distribution whose mean and variance are unknown can be found as follows.
here,

It is stipulated that:
For example, when calculating the standard deviation corresponding to May 1st during the simulation period (when past data is for the three-month period from February 1st, 2020 to April 30th of the same year), the number of past shipments refers to the number of times product P was shipped during the above three-month period. The number of times product P was shipped corresponds to the number of days on which shipments of product P occurred, and does not mean the number of shipments. If shipments were made only on February 1st, February 2nd, and February 3rd, 2020 during the above three-month period, n will be 3.

と定める。上記3か月の期間において2020年2月1日及び、同年2月2日、及び同年2月3日のみ出荷を行なった場合、1回目の出荷日は2020年2月1日となる。すなわち、d₁は2020年2月1日を表し、2回目の出荷日であるd₂は2020年2月2日を表し、d₃は2020年2月3日を指す。ここで2020年2月1日(すなわちd₁)に出荷した商品Ｐの数が10であった場合、s₁は10である。 Furthermore, here

If shipments were made only on February 1, 2020, February 2, and February 3, 2020 during the three-month period mentioned above, the first shipment date would be February 1, 2020. In other words, d ₁ represents February 1, 2020, d ₂ , the second shipment date, represents February 2, 2020, and d ₃ represents February 3, 2020. Here, if the number of products P shipped on February 1, 2020 (i.e., d ₁ ) was 10, then s ₁ is 10.

分散であるσ²は以下の通り算出することができる。

また、標準偏差σは分散σ²の平方根を計算することで算出できる。 In this case, using n and the value b calculated as follows:

The variance, ^σ2 , can be calculated as follows:

In addition, the standard deviation σ can be calculated by taking the square root of the variance ^σ2 .

In this way, the standard deviation corresponding to May 1st, the first day of the simulation period, can be calculated as above using data from February 1st to April 30th. Furthermore, the standard deviation corresponding to May 2nd, which is included in the simulation period, can be calculated as above using data from February 1st to May 1st, and the standard deviation corresponding to May 31st, the last day of the simulation period, can be calculated as above using data from February 1st to May 30th. In this way, the standard deviation corresponding to each day included in the simulation period can be calculated in a similar manner using data from February 1st to the day before that day.

Then, the number of shipments (=order point) of product P that will be smaller than the specified value with a probability of x% on each day included in the simulation period can be calculated by the number of shipments on that day (the number of shipments on that day predicted in ST706 by the prediction device 100 using data from the past week with that day as the base) + "parameter" x standard deviation corresponding to that day. For example, when calculating a confidence interval for x = 95%, 1.64 can be used as the parameter. Similarly, by changing x%, the parameter corresponding to the changed x can be obtained.

In one embodiment, in ST904, the order point calculation device 500 can use numbers from 1 to 10 incremented by 0.01 (1, 1.01, 1.02, ..., 9.98, 9.99, 10) as candidates for multiple parameters.

Next, the reorder point calculation device 500 can execute ST906 to ST910, which will be described later, for each of the multiple parameters. For simplicity of explanation, it is assumed here that the total number of the multiple parameters is N. In this case, the reorder point calculation device 500 can execute ST906 to ST910, which will be described later, for each of parameters 1 to N.

In ST906, the order point calculation device 500 can calculate the order point used in a simulation described below, using a corresponding parameter from among the multiple parameters. For example, in ST906 ₁ corresponding to parameter 1, the order point calculation device 500 can calculate the order point for each day included in the simulation period (here, for example, May 1 to May 31) using parameter 1. Similarly, in ST906 _N corresponding to parameter N, the order point calculation device 500 can calculate the order point for each day included in the simulation period (here, for example, May 1 to May 31) using parameter N.

As described above, the reorder point is calculated by the number of shipments for each day included in the simulation period (the number of shipments for that day predicted in ST706 by the prediction device 100 using data for the past week with that day as the base day) + "parameter" x standard deviation corresponding to that day.
Specifically, for example, the order point for May 1, which is the first day of the simulation period, can be calculated by the number of shipments of goods on May 1 (the number of shipments for May 1 predicted in ST706 by the prediction device 100 using data for the past week (April 25 to April 30) based on May 1) + the parameter × the standard deviation corresponding to May 1. Also, the order point for May 2, which is included in the simulation period, can be calculated by the number of shipments of goods on May 2 (the number of shipments predicted in ST706 by the prediction device 100 using data for the past week based on May 2) + the parameter × the standard deviation corresponding to May 2. Similarly, the order point for May 31, which is the last day of the simulation period, can be calculated by the number of shipments of goods on May 31 (the number of shipments predicted in ST706 by the prediction device 100 using data for the past week based on May 31) + the parameter × the standard deviation corresponding to May 31.

In ST908 ₁ to ST908 _N , the order point calculation device 500 can execute a simulation using the order points of each day included in the simulation period calculated in ST906 ₁ to ST906 _N. This simulation will be described later with reference to FIG.

In ST910 ₁ to ST910 _N , the order point calculation device 500 can extract the remaining stock and the number of stock-outs obtained in the simulations in ST908 ₁ to ST908 _N , respectively.

Finally, in ST912, the reorder point calculation device 500 can use the remaining stock and the number of stock-outs obtained in ST910 ₁ to ST910 _N for each of the multiple parameters to extract, from the multiple parameters, the parameter for which the reference value (α × "remaining stock on the last day of the simulation period (May 31)" + "the number of stock-outs that occurred during the simulation period (May 1 to May 31) (number of stock-outs)") is the smallest, as the final parameter. In other words, the reorder point calculation device 500 can extract, from the simulations in ST908 ₁ to ST908 _N , the parameter used in the specific simulation for which the reference value is the smallest, as the final parameter. It should be noted that α can be any number that the system user can appropriately determine in consideration of the balance between the number of stock-outs and the remaining stock.

Furthermore, the order point calculation device 500 can calculate the order point (final threshold value) on June 1, which is the target date calculated using the parameters extracted in this manner, using the following formula.
"The number of shipments of products on June 1 predicted in ST706 by the prediction device 100 based on the data for the past week (May 25 to May 31) with June 1 as the target date as the base date" + "Parameter extracted in ST912" x "Standard deviation corresponding to June 1 (standard deviation calculated as above based on the data from February 1 to May 31)"

In addition, the order point calculation device 500 can calculate the order point (final threshold value) on the target date, June 2nd, using the following formula.
"The number of shipments of products on June 2 predicted in ST706 by the prediction device 100 based on the data for the past week (May 26 to June 1) with June 2 as the target date as the base date" + "Parameter extracted in ST912" x "Standard deviation corresponding to June 2 (standard deviation calculated as above based on the data from February 1 to June 1)"

Furthermore, the reorder point calculation device 500 can similarly calculate the reorder point for the target date, June 3rd and after.

In order to reduce stockouts of product P at the wholesaler, it is necessary to increase the quantity of product P in stock at the wholesaler. However, if the quantity of product P in stock is increased too much, there is a high possibility that this will lead to an increase in the number of product P discarded. Therefore, in ST912, the reorder point calculation device 500 can select parameters, and therefore the reorder point, that minimize the remaining stock and the number of stockouts by extracting parameters that minimize "α x remaining stock on the last day of the simulation period + number of stockouts that occurred during the simulation period (number of stockouts)."

In addition, in ST912, the reorder point calculation device 500 can also add the number of orders for product P to select parameters that optimize "α x remaining stock on the last day of the simulation period + β x number of stockouts that occurred during the simulation period (number of stockouts) + number of orders that occurred during the simulation period (number of orders)" and therefore the reorder point. Note that α and β can be any numbers that the system user can appropriately determine in consideration of the balance between the number of stockouts, remaining stock, and number of orders.

Next, a specific example of the simulation performed in ST908 will be described with reference to FIG. 15. Note that the order point calculation device 500 can execute the processes in ST1002 to ST1024 described below for each of the multiple parameters extracted in ST904 (FIG. 14) described above.

First, in ST1002, the order point calculation device 500 can obtain the inventory quantity and shipment quantity on the first day (May 1, 2020) of the simulation period (here, as described above, from May 1, 2020 to May 31, 2020).

In ST1004, the reorder point calculation device 500 can determine whether the inventory quantity (inventory quantity on May 1 acquired in ST1002) is lower than the shipment quantity (shipment quantity on May 1 acquired in ST1002). If it is determined that the inventory quantity is lower than the shipment quantity, the reorder point calculation device 500 can increment the number of out-of-stocks by 1 and update the inventory quantity to 0 in ST1006. On the other hand, if it is determined that the inventory quantity is not lower than the shipment quantity, the reorder point calculation device 500 can update the "inventory quantity - shipment quantity" as the new inventory quantity in ST1008. After ST1006 and ST1008, the process proceeds to ST1010.

In ST1010, the reorder point calculation device 500 can determine whether the inventory quantity is below the reorder point on May 1. Here, the reorder point on May 1 may be calculated using a corresponding parameter from among the multiple parameters, as described above, in ST906. For example, if the process shown in FIG. 15 is performed for parameter N, the reorder point on May 1 may be the one calculated in _ST906N using parameter N.

If it is determined that the inventory quantity is below the order point on May 1, the order point calculation device 500 can calculate the incoming quantity as "order point on May 1 - inventory quantity (updated inventory quantity)" in ST1012. On the other hand, if it is determined that the inventory quantity is not below the order point on May 1, the order point calculation device 500 can set the incoming quantity to 0 in ST1014. After ST1012 and ST1014, the process proceeds to ST1016.

In ST1016, the reorder point calculation device 500 can determine whether data for the next day (here, May 2, 2020) exists. If it is determined that data for the next day (May 2) exists, the reorder point calculation device 500 can set the "inventory quantity (updated inventory quantity) + arrival amount" as the inventory quantity for the next day (May 2) in ST1018. After ST1018, the process proceeds to ST1022.

On the other hand, if it is determined that no data exists for the next day, the reorder point calculation device 500 can set the inventory quantity (updated inventory quantity) as the remaining inventory on the last day of the simulation period in ST1020. The reorder point calculation device 500 can use the remaining inventory on the last day of the simulation period and the number of stockouts that occurred during the simulation period (number of stockouts) as output data for the corresponding parameters (after which, the process returns to ST910 described above).

In ST1022, the order point calculation device 500 obtains the shipment quantity for May 2, which is the second day of the simulation period, and can set the inventory quantity for the next day (May 2) (the "next day's inventory quantity" set in ST1018) as the inventory quantity for May 2, which is the second day of the period.

In ST1024, the reorder point calculation device 500 can similarly execute ST1004 to ST1020 described above for the second day of the period (May 2). In this case, the reorder point used in ST1010 and ST1012 is the reorder point for May 2. The reorder point for May 2 may be calculated in ST906 using the corresponding parameter from among the multiple parameters described above, as described above.

Furthermore, the reorder point calculation device 500 can execute the same process from the third day of the period (May 3, 2020) to the last day of the period (May 31, 2020). As a result, ultimately, in ST1020, the reorder point calculation device 500 can use the remaining inventory on the last day of the simulation period (May 31) and the number of stockouts (number of stockouts) that occurred during the simulation period (May 1 to May 31) as output data for the corresponding parameters, as described above (after which, the process returns to ST910 described above).

As described above, the reorder point calculation device 500 disclosed in the present application prepares multiple parameters corresponding to each of multiple probabilities that the shipment quantity of a product will fall within the confidence interval on a target date, executes a simulation using each parameter to calculate the remaining inventory and number of stockouts for that product, and calculates the reorder point (final threshold) for that product using the parameter among the multiple parameters that minimizes the remaining inventory and number of stockouts. This makes it possible to obtain a more optimized reorder point for that product. Therefore, a wholesaler who is provided with this reorder point can more effectively reduce the inventory and number of stockouts for that product by ordering the product from the manufacturer when the inventory quantity of that product falls below the reorder point (final threshold).

7. Various Aspects An information processing device according to a first aspect is an information processing device including at least one processor, used in association with a wholesaler that receives merchandise from a manufacturer and ships it to a retailer, and generating information regarding the number of the merchandise to be shipped from the wholesaler to the retailer on a target date, wherein the at least one processor acquires, as input information, for the merchandise, a category of the merchandise, the number of the merchandise shipped from the wholesaler to the retailer from one day before to seven days before the target date, and a number of days of inventory obtained by dividing the number of inventory of the merchandise one day before the target date by an average sales number that is an average of the number of sales of the merchandise per day at the retailer, or the average sales number of the merchandise, and inputs the input information into a learning model generated by performing supervised learning, thereby causing the learning model to output prediction information regarding the number of the merchandise to be shipped from the wholesaler to the retailer on the target date.

The information processing device according to the second aspect can be the same as in the first aspect above, in which "the at least one processor outputs, as the forecast information, the number of the products to be shipped from the wholesaler to the retailer on the target date."

The information processing device according to the third aspect can be configured in the first aspect as follows: "The at least one processor outputs, as the forecast information, the rate of change in the number of the products to be shipped from the wholesaler to the retailer on the target date relative to the average number of the products previously shipped from the wholesaler to the retailer."

The information processing device according to the fourth aspect can be the same as in the first aspect above, in that "the at least one processor uses a plurality of target dates as the target date, and outputs, as the forecast information, the number of the products to be shipped from the wholesaler to the retailer on each of the plurality of target dates."

The information processing device according to the fifth aspect can be the same as in the first aspect above, in that "the at least one processor uses a plurality of target dates as the target dates, and outputs, as the forecast information, the total number of the products to be shipped from the wholesaler to the retailer on all of the plurality of target dates."

The information processing device according to the sixth aspect can generate the learning model in any one of the first to fifth aspects by the at least one processor performing learning using, as training data, the category of the product, the number of the product shipped from the wholesaler to the retailer in each of the past seven days before the specific day, and the number of days of inventory, which is the value obtained by dividing the inventory number of the product one day before the target day by the average sales volume, which is the average of the number of sales of the product per day at the retailer, or the average sales volume of the product, and the number of the product shipped from the wholesaler to the retailer on the specific day.

The information processing device according to the seventh aspect can be the one according to the sixth aspect, in which "the at least one processor generates the learning model by performing supervised learning using gradient boosting trees, linear regression, or a neural network."

The information processing device according to the eighth aspect is the same as that according to the second aspect, and "the at least one processor obtains a standard deviation using the number of the products shipped from the wholesaler to the retailer for each of a plurality of past reference dates prior to the target date, and for each of a plurality of parameters, obtains a threshold value for the number of the products in stock at the wholesaler on the reference date, which is used to place an order for the products with the manufacturer when the number of the products in stock falls below the threshold value, using the parameters, the number of the products shipped from the wholesaler to the retailer on each reference date output as the prediction information, and the standard deviation obtained for the reference date, and executes a simulation over the entirety of the plurality of past reference dates using the threshold value obtained for each of the plurality of parameters. and obtain the number of stockouts of the product at the wholesaler on the last day of the multiple past reference dates and the number of times stockouts occur at the wholesaler across the multiple past reference dates, and determine as final parameters the parameters used in a specific simulation in which a reference value calculated based on the number of stockouts of the product at the wholesaler on the last day and the number of stockouts is the smallest among the simulations, and calculate a final threshold for the number of stockouts of the product at the wholesaler on the target date using the final parameters, the number of products shipped from the wholesaler to the retailer on the target date output as the forecast information, and the standard deviation obtained for the target date, which is a final threshold for the number of stockouts of the product at the wholesaler on the target date and which is used to place an order for the product with the manufacturer if the number of stockouts falls below the final threshold.

The information processing device according to the ninth aspect can be the same as the eighth aspect above, in which "the at least one processor calculates the reference value according to the formula: α x the number of items in stock at the wholesaler on the final day + the number of times."

The information processing device according to the tenth aspect is the same as that according to the ninth aspect, and "the at least one processor, in each simulation, for each of the multiple past reference dates, if the inventory quantity of the product at the wholesaler on the reference date is lower than the number of the product shipped from the wholesaler to the retailer, increases the number of times that stockouts occur at the wholesaler by one and updates the inventory quantity of the product at the wholesaler on the reference date to 0, and if the inventory quantity of the product at the wholesaler on the reference date is equal to or greater than the number of the product shipped from the wholesaler to the retailer, calculates the inventory quantity of the product at the wholesaler on the reference date minus the number of the product shipped from the wholesaler to the retailer as the previous number of the reference date. the number of items in stock at the wholesaler is updated as the number of items in stock at the wholesaler on the day after the reference date, and if the updated number of items in stock at the wholesaler is below the threshold corresponding to the simulation, the difference between the updated number of items in stock at the wholesaler and the threshold is set as the number of items scheduled to arrive at the wholesaler, and if the updated number of items in stock at the wholesaler is equal to or greater than the threshold corresponding to the simulation, the number of items scheduled to arrive at the wholesaler is set to 0, and the sum of the updated number of items in stock at the wholesaler and the number of items scheduled to arrive is set as the number of items in stock at the wholesaler on the day after the reference date. This process can be executed sequentially from older past reference dates to newer past reference dates.

The information processing device according to the eleventh aspect can be any of the first to tenth aspects, in which "the at least one processor includes a central processing unit (CPU), a microprocessor, and/or a graphics processing unit (GPU)."

The method according to the twelfth aspect is a method executed by at least one processor executing computer-readable instructions for use in association with a wholesaler that receives merchandise from a manufacturer and ships it to a retailer, and for generating information regarding the number of the merchandise to be shipped from the wholesaler to the retailer on a target date, the at least one processor executing the instructions to obtain, as input information, the category of the merchandise, the number of the merchandise shipped from the wholesaler to the retailer from one day before to seven days before the target date, and the number of days of inventory, which is a value obtained by dividing the number of inventory of the merchandise one day before the target date by the average sales number, which is the average number of sales of the merchandise at the retailer per day, or the average sales number of the merchandise, and inputting the input information into a learning model generated by performing supervised learning, thereby causing the learning model to output prediction information regarding the number of the merchandise to be shipped from the wholesaler to the retailer on the target date.

The method according to the thirteenth aspect can be the method according to the twelfth aspect above, where "the at least one processor includes a central processing unit (CPU), a microprocessor, and/or a graphics processing unit (GPU)."

The computer program according to the fourteenth aspect can be used in association with a wholesaler that purchases products from manufacturers and ships them to retailers, and can be executed by at least one processor that generates information regarding the number of the products to be shipped from the wholesaler to the retailer on a target date, and can cause the at least one processor to function as follows: for the products, the category of the products, the number of the products shipped from the wholesaler to the retailer from one day before the target date to seven days before the target date, and the number of days of inventory, which is the value obtained by dividing the number of products in stock one day before the target date by the average sales number, which is the average number of sales of the products at the retailer per day, or the average sales number of the products, as input information; and inputting the input information into a learning model generated by performing supervised learning, thereby causing the learning model to output prediction information regarding the number of the products to be shipped from the wholesaler to the retailer on the target date.

The computer program of the fifteenth aspect can be the same as that of the fourteenth aspect, in which "the at least one processor includes a central processing unit (CPU), a microprocessor, and/or a graphics processing unit (GPU)."

As described above, according to various aspects, it is possible to provide an information processing device, method, and computer program for predicting the number of products to be shipped from a wholesaler to a retailer.

REFERENCE SIGNS LIST 1 Prediction system 2 Order point calculation system 100 Prediction device 110 Communication unit 120 Control unit 122 Acquisition unit 124 Generation unit 126 Prediction unit 128 Provision unit 130 Storage unit 132 Retail information storage unit 134 Wholesale information storage unit 136 Model information storage unit 138 Provision information storage unit 200 Retail system 300 Wholesale system 400 Network

Claims (8)

An information processing device including at least one processor, for use in association with a wholesaler that purchases merchandise from a manufacturer and ships it to a retailer, the information processing device generating information relating to a quantity of the merchandise to be shipped from the wholesaler to the retailer on a target date,
the at least one processor:
Acquiring, as input information, the category of the product, the number of the product shipped from the wholesaler to the retailer from one day before to seven days before the target date, and the number of days of inventory, which is the value obtained by dividing the inventory number of the product one day before the target date by the average sales number, which is the average number of sales of the product per day at the retailer, or the average sales number of the product; and
inputting the input information into a learning model generated by performing supervised learning, and outputting from the learning model prediction information regarding the number of the products to be shipped from the wholesaler to the retailer on the target date, wherein the learning model outputs, as the prediction information, the number of the products to be shipped from the wholesaler to the retailer on each day between the target date and a day after the target date, or the learning model outputs, as the prediction information, the total number of the products to be shipped from the wholesaler to the retailer on all days between the target date and a day after the target date;
23. An information processing device configured to execute the above-mentioned. the at least one processor:
2. The information processing device of claim 1, wherein the learning model is generated by performing learning using as training data the category of the product, the number of the product shipped from the wholesaler to the retailer in each of at least the past seven days prior to a specific date, the number of days of inventory which is the value obtained by dividing the inventory number of the product by the average sales volume which is the average sales volume per day of the product at the retailer, and the number of the product shipped from the wholesaler to the retailer on the specific date. the at least one processor:
The information processing device according to claim 2 , wherein the learning model is generated by performing supervised learning using a gradient boosting tree, a linear regression, or a neural network. The information processing device according to claim 1 , wherein the at least one processor comprises a central processing unit (CPU), a microprocessor, and/or a graphics processing unit (GPU). 1. A method executed by at least one processor executing computer readable instructions for use in connection with a wholesaler that receives merchandise from a manufacturer and ships it to a retailer, the method generating information regarding a quantity of the merchandise to be shipped by the wholesaler to the retailer on a target date, the method comprising:
Execution of the instructions by the at least one processor results in:
Acquiring, as input information, the category of the product, the number of the product shipped from the wholesaler to the retailer from one day before to seven days before the target date, and the number of days of inventory, which is the value obtained by dividing the inventory number of the product one day before the target date by the average sales number, which is the average number of sales of the product per day at the retailer, or the average sales number of the product; and
inputting the input information into a learning model generated by performing supervised learning, and outputting from the learning model prediction information regarding the number of the products to be shipped from the wholesaler to the retailer on the target date, wherein the learning model outputs, as the prediction information, the number of the products to be shipped from the wholesaler to the retailer on each day between the target date and a day after the target date, or the learning model outputs, as the prediction information, the total number of the products to be shipped from the wholesaler to the retailer on all days between the target date and a day after the target date;
The method of claim 1 , further comprising : The method of claim 5 , wherein the at least one processor comprises a central processing unit (CPU), a microprocessor, and/or a graphics processing unit (GPU). a method for generating information relating to a quantity of the product to be shipped from the wholesaler to the retailer on a target date, the method being executed by at least one processor;
Acquiring, as input information, the category of the product, the number of the product shipped from the wholesaler to the retailer from one day before to seven days before the target date, and the number of days of inventory, which is the value obtained by dividing the inventory number of the product one day before the target date by the average sales number, which is the average number of sales of the product per day at the retailer, or the average sales number of the product; and
inputting the input information into a learning model generated by performing supervised learning, and outputting from the learning model prediction information regarding the number of the products to be shipped from the wholesaler to the retailer on the target date, wherein the learning model outputs, as the prediction information, the number of the products to be shipped from the wholesaler to the retailer on each day between the target date and a day after the target date, or the learning model outputs, as the prediction information, the total number of the products to be shipped from the wholesaler to the retailer on all days between the target date and a day after the target date;
23. A computer program product comprising: a computer program product configured to cause the at least one processor to execute a program; The computer program product of claim 7 , wherein the at least one processor comprises a central processing unit (CPU), a microprocessor, and/or a graphics processing unit (GPU).