Demand forecasting device, demand forecasting method, and demand forecasting program

Abstract

This demand prediction device predicts a number of store visits on the basis of a store visit number prediction model for predicting the number of store visits and factor information relating to factors affecting a number of sales, predicts a ratio of the number of sales to the number of visits on the basis of the factor information and a ratio prediction model for predicting the ratio, and predicts the number of sales of a product on the basis of the predicted number of store visits and ratio.

Description

[Technical field]

[0001] The disclosed technology relates to a demand forecasting device, a demand forecasting method, and a demand forecasting program. [Background technology]

[0002] Demand forecasting techniques are known that predict the number of customers visiting a store such as a restaurant.

[0003] For example, Non-Patent Document 1 discloses a demand forecasting technology that predicts the number of visitors to a store by combining environmental information related to the external environment around the store, such as POS (Point of Sale) data, weather, and event data. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Takashi Tanizaki, Yuta Hanayama, Takeshi Shimmura "Forecasting Customers Visiting Using Machine Learning and Characteristics Analysis with Low Forecasting Accuracy Days", APMS 2020: Advances in Production Management Systems. Towards Smart and Digital Manufacturing pp 670-678, 25 August 2020. Summary of the Invention [Problem to be solved by the invention]

[0005] The problem with the technology described in Non-Patent Document 1 is that it is not possible to predict the number of products sold in a store, that is, the demand for each product.

[0006] The disclosed technology has been made in consideration of the above points, and aims to provide a demand forecasting device, a demand forecasting method, and a demand forecasting program that can predict the number of products sold in a store. [Means for solving the problem]

[0007] A first aspect of the present disclosure is a demand prediction device including: an acquisition unit that acquires factor information related to factors that affect sales volume of a product at a store; a visitor count prediction model for predicting the number of visits to the store in a prediction period that is learned based on visitor count information related to the number of visits to the store in the past and the factor information; a visitor prediction unit that predicts the number of visits to the store in the prediction period based on the factor information acquired by the acquisition unit; a proportion prediction model for predicting the proportion that is learned based on the past factor information and the proportion of the number of visits that the number of sales of the product accounts for, and a proportion prediction unit that predicts the proportion based on the factor information acquired by the acquisition unit; and a sales prediction unit that predicts the sales volume of the product based on the visitor count predicted by the store visit prediction unit and the proportion predicted by the proportion prediction unit.

[0008] A second aspect of the present disclosure is a demand forecasting method including: an acquisition unit acquires factor information regarding factors affecting sales volume of a product at a store; a store visit prediction unit predicts the number of visits to the store in a forecast period based on a store visit prediction model for predicting the number of visits to the store in a forecast period, the store visit prediction model having been learned based on store visit information regarding the number of past visits to the store and the factor information, and the factor information acquired by the acquisition unit; a proportion prediction unit predicts the proportion based on the factor information and a proportion prediction model for predicting the proportion, the proportion being accounted for by the number of visits to the store, and the factor information acquired by the acquisition unit; and a sales prediction unit predicts the sales volume of the product based on the number of visits predicted by the store visit prediction unit and the proportion predicted by the proportion prediction unit.

[0009] A third aspect of the present disclosure is a demand prediction program for causing a computer to function as each part of the demand prediction device of the first aspect. Effect of the Invention

[0010] The disclosed technology makes it possible to predict the number of products sold in a store. [Brief description of the drawings]

[0011] [Figure 1] FIG. 2 is a schematic diagram illustrating an example of a hardware configuration of the demand prediction device according to the embodiment. [Diagram 2] 1 is a block diagram showing an example of a functional configuration of a demand prediction device according to an embodiment. [Diagram 3] FIG. 4 is a schematic diagram illustrating an example of a product table according to the embodiment. [Figure 4] FIG. 2 is a schematic diagram illustrating an example of a configuration of a store visit number database according to the embodiment. [Diagram 5] FIG. 4 is a schematic diagram illustrating an example of a configuration of an environment database according to the embodiment. [Figure 6] FIG. 1 is a schematic diagram illustrating an example of a configuration of a sales number database according to an embodiment; [Figure 7] FIG. 2 is a schematic diagram illustrating an example of a configuration of a feature database according to the embodiment. [Figure 8] FIG. 2 is a schematic diagram illustrating an example of a configuration of a popularity database according to the embodiment. [Figure 9] FIG. 2 is a schematic diagram illustrating an example of a configuration of a popularity factor database according to the embodiment. [Figure 10] 4 is a flowchart illustrating an example of a demand learning process in the demand prediction device according to the embodiment. [Figure 11] 4 is a flowchart illustrating an example of a demand prediction process in the demand prediction device according to the embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, an example of an embodiment of the disclosed technology will be described with reference to the drawings. In each drawing, the same or equivalent components and parts are given the same reference numerals. In addition, the dimensional ratios of the drawings are exaggerated for the convenience of explanation and may differ from the actual ratios.

[0013] 1, the demand forecasting device 10 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a storage 14, a display unit 16, and a communication I / F (Interface) 17. Each component is connected to each other via a bus 19 so as to be able to communicate with each other.

[0014] The CPU 11 is a central processing unit, and executes various programs and controls each part. That is, the CPU 11 reads a program from the ROM 12 or the storage 14, and executes the program using the RAM 13 as a working area. The CPU 11 controls each of the above components and performs various arithmetic processing according to the program stored in the ROM 12 or the storage 14. In this embodiment, a demand learning program and a demand forecasting program are stored in the ROM 12 or the storage 14. The demand learning program and the demand forecasting program may be one program, or may be a group of programs composed of multiple programs or modules.

[0015] The ROM 12 stores various programs and various data. The RAM 13 temporarily stores programs or data as a working area. The storage 14 is configured with an HDD (Hard Disk Drive) or an SSD (Solid State Drive) and stores various programs including an operating system and various data.

[0016] The display unit 16 is, for example, a liquid crystal display, and displays the sales volume of the product predicted by the demand prediction process described later. The display unit 16 may adopt a touch panel system and function as an input unit.

[0017] The communication I / F 17 is an interface for communicating with other devices, and uses standards such as Ethernet (registered trademark), FDDI, and Wi-Fi (registered trademark).

[0018] Next, a functional configuration of the demand prediction device 10 will be described. As shown in Fig. 2, the demand prediction device 10 includes an acquisition unit 101, a store visit learning unit 102, a ratio learning unit 103, a store visit prediction unit 104, a ratio prediction unit 105, a sales prediction unit 106, and an output unit 107. Furthermore, the demand prediction device 10 includes a store visit number database 120, an environment database 121, a sales number database 122, a feature database 123, a popularity database 124, and a popularity factor database 125.

[0019] In this embodiment, a prediction period, which is a period for predicting the sales volume of each product, i.e., a prediction unit is set to one day, and a case where the sales volume of each product in the prediction period is predicted is described. Note that the sales volume of each classification (category) to which each product belongs in the prediction period may also be predicted. In addition, in this embodiment, a case where products of multiple classifications are sold in one day, as in the product table shown in FIG. 3, and one product corresponds to one classification, is described. Note that there may be multiple products corresponding to one classification. In addition, the prediction period is not limited to the current day, but may be the next day or later, and the length of the prediction period is not limited to one day, but may be in units of one hour, one week, etc. In addition, in this embodiment, the products are assumed to be menu items provided in a restaurant, but are not limited thereto.

[0020] The acquisition unit 101 acquires visit number information on the number of past visits to the store, factor information on factors that affect the number of products sold at the store, and sales number information on the number of products sold in the past. The factor information includes, for example, environmental information on the external environment around the store, and feature information on the features of the products. Note that, although the following describes a case where both the environmental information and the feature information are used as the factor information, it is also possible to use either one of them.

[0021] Specifically, the acquiring unit 101 acquires, as the visit number information during learning, from the visit number database 120, the number of visits to the store during a first period having the same length as the length of the prediction target period (here, one day), and the number of visits to the store until the day before the first period, for each of a plurality of first periods. Note that the acquiring unit 101 may acquire the number of visits on the day before the first period as the number of visits to the store until the day before the first period. Also, instead of the number of visits on the day before the first period, the acquiring unit 101 may acquire the number of visits on each day from a predetermined number of days before (e.g., one week before) to the day before the first period, the average number of visits per day until the day before which is the average of the number of visits on each day, the cumulative number of visits from a predetermined number of days before to the day before the day before, etc.

[0022] Moreover, the acquisition unit 101 acquires the number of visits up to the day before the prediction target period (here, the current day) as the visit number information at the time of prediction from the visit number database 120. Note that the number of visits up to the day before the prediction target period is the same as the number of visits up to the day before the first period described above.

[0023] As shown in FIG. 4, the visit number database 120 is a database in which the number of visits to a store for each predetermined period is stored. In the visit number database 120 shown in FIG. 4, the number of visits to a store in the past per hour is stored. In addition, the visit number database 120 may store the number of visits to a store in the past per day. As in the example of FIG. 4, when the number of visits per hour is stored and the length of the prediction target period and the first period is one day, the acquisition unit 101 may acquire the total number of visits to a store in the past per hour on the day before the prediction target period and the day before the first period as the number of visits for one day on the day before the prediction target period and the day before the first period, or may acquire a vector listing the number of visits per hour for one day.

[0024] Moreover, the acquiring unit 101 acquires, as environmental information at the time of learning, the temperature and the amount of precipitation for each first period from the environmental database 121. Moreover, the acquiring unit 101 acquires, as environmental information at the time of prediction, the temperature and the amount of precipitation for the prediction target period.

[0025] As shown in FIG. 5, the environment database 121 is a database in which the temperature and precipitation around the store for each predetermined date and time are stored. In the environment database 121 shown in FIG. 5, the temperature and precipitation around the store per hour are stored. The environment database 121 may store the temperature and precipitation around the store per day. The environment database 121 may also store other environmental information such as humidity and pollen information. In the environment database 121, the temperature and precipitation actually observed are stored in association with each other at past dates and times, and the temperature and precipitation based on the forecast of the Japan Meteorological Agency or the like are stored in association with each other at the date and time of the day. The acquisition unit 101 may acquire the average values ​​of the temperature and precipitation per hour for the prediction target period and the first period as the temperature and precipitation for one day for each prediction target period and the first period, or may acquire a vector in which the temperature and precipitation per hour for one day are listed.

[0026] The acquisition unit 101 also acquires, for multiple first periods, the sales number for each product in the first period, the total sales number of all products up to the day before the first period, and the sales number for each product, from the sales number database 122, as sales number information at the time of learning. Hereinafter, the total sales number of all products is referred to as the total sales number. The acquisition unit 101 also acquires, for multiple first periods, the total sales number and the sales number for each product up to the day before the prediction period, from the sales number database 122, as sales number information at the time of prediction.

[0027] 6, the sales number database 122 is a database that stores the classification (hereinafter simply referred to as "classification"), name (hereinafter also referred to as "product name"), and sales number of products sold for each predetermined period. The acquisition unit 101 may acquire the total of the total sales numbers per hour on the day before the prediction target period and the day before the first period as the total sales number for one day on the day before the prediction target period and the day before the first period, or may acquire a vector that lists the total sales numbers per hour for one day.

[0028] Moreover, the acquiring unit 101 acquires, as feature information at the time of learning, from the feature database 123, ingredients used in all products sold in the first period and flag information indicating whether a specific ingredient is used in the products. Moreover, the acquiring unit 101 acquires, as feature information at the time of prediction, ingredients and flag information used in all products scheduled to be sold in the prediction target period from the feature database 123. The acquiring unit 101 converts the acquired ingredients into vectors by multi-hot encoding. Moreover, the acquiring unit 101 may include the above-mentioned product classification in the feature information. When the classification is also included in the feature information, the acquiring unit 101 converts the classification into a vector by one-hot encoding.

[0029] As shown in Fig. 7, the feature database 123 stores the product name, category, ingredients, and flag information of the product sold on each specified date. In the example shown in Fig. 7, the flag information is information indicating whether meat or fish is used in the corresponding product. In the example shown in Fig. 7, the flag information is stored as "1" if the corresponding product uses a specific ingredient, and as "0" if the specific ingredient is not used. In addition, the feature database 123 may store in association with the product characteristics such as the price or place of origin of the product sold on the corresponding date.

[0030] The acquisition unit 101 passes various information acquired during learning to the store visit learning unit 102 and the ratio learning unit 103, and passes various information acquired during prediction to the store visit prediction unit 104 and the ratio prediction unit 105.

[0031] The store visit learning unit 102 learns a store visit number prediction model 130 for predicting the number of visits to a store in a prediction target period, based on the store visit number information, environmental information, sales number information, and feature information.

[0032] Specifically, the store visit learning unit 102 uses the number of visits in the first period acquired by the acquisition unit 101 as the objective variable, and the number of visits and total sales up to the day before the first period, the temperature and precipitation in the first period, and the ingredients and flag information converted into vectors of all products sold in the first period as explanatory variables to learn the store visit number prediction model 130 using machine learning. Note that the store visit learning unit 102 may learn the store visit number prediction model 130 using, for example, multiple regression analysis. The store visit learning unit 102 stores the learned store visit number prediction model 130 in a predetermined storage area.

[0033] The proportion learning unit 103 learns a proportion prediction model 131 for predicting a sales proportion, which is the proportion of the number of product sales to the number of visits to a store, based on the number of visits information, environmental information, sales number information, and feature information.

[0034] First, the ratio learning unit 103 calculates the sales ratio for the first period for each product by dividing the sales number for each product in the first period acquired by the acquisition unit 101 by the number of store visits in the corresponding first period acquired by the acquisition unit 101. Next, the ratio learning unit 103 calculates a simple moving average of the sales ratios by using the sales ratios of each product whose first period is included in the most recent five business days of the corresponding first period, for example.

[0035] Next, the ratio learning unit 103 calculates the popularity indicating the relative size of the sales ratio for each product. Specifically, if the value obtained by dividing the sales ratio of the product in the corresponding first period by the simple moving average of the sales ratio for the corresponding first period is equal to or greater than a first threshold (first threshold>1), the ratio learning unit 103 sets the popularity of the product in the corresponding first period to a value indicating relatively high popularity (for example, 1), and if the value is less than the first threshold, the ratio learning unit 103 sets the popularity of the product in the corresponding first period to a value indicating relatively low popularity (for example, 0). Then, the ratio learning unit 103 stores the calculated popularity in the popularity database 124. The ratio learning unit 103 may calculate the popularity of each product based on the value obtained by dividing the sales ratio for the first period by the weighted moving average or exponential moving average of the sales ratio for the first period.

[0036] As shown in FIG. 8, the popularity database 124 is a database in which the popularity of products is stored for each period (here, for each day corresponding to the first period).

[0037] Next, the ratio learning unit 103 learns a popularity prediction model for predicting the popularity based on the sales ratio, the environmental information, and the characteristic information.

[0038] Specifically, the ratio learning unit 103 acquires the popularity and period of each product from the popularity database 124. Hereinafter, the period corresponding to each product is referred to as the popular period. In addition, the ratio learning unit 103 acquires the temperature and precipitation of the popular period from the environment database 121.

[0039] In this embodiment, for each product, the oldest date in the popularity period is T1, the date on which the ratio learning unit 103 learns the popularity prediction model is T2, and the period from T1 to T2 is hereinafter referred to as the second period. Note that a date obtained by subtracting a predetermined number of days from T2 may be applied as T1. Also, the newest date in the popularity period may be applied as T2.

[0040] Then, the percentage learning unit 103 uses records in which the popular period is included in the second time period to learn a popularity prediction model for each product using, for example, multiple regression analysis, with the popularity of the corresponding product as the objective variable and the temperature and precipitation during the popular period as explanatory variables. For example, the percentage learning unit 103 uses the regression equation represented by formula (1) as the popularity prediction model, and calculates coefficients a1, a2, ..., and the intercept. Note that the percentage learning unit 103 may learn the popularity prediction model using other machine learning techniques.

[0041]

number

[0042] If the coefficient of determination of the learned popularity prediction model is less than a predetermined second threshold, the ratio learning unit 103 shortens the second period and re-learns the popularity prediction model. In this embodiment, if the coefficient of determination of the popularity prediction model is less than the second threshold, the ratio learning unit 103 adds a predetermined fixed value Tw (e.g., one month) to T1 and re-learns the popularity prediction model. The ratio learning unit 103 may change T1 so that the number of records related to the corresponding product included in the shortened second period is reduced by a fixed value, and re-learn the popularity prediction model.

[0043] When the coefficient of determination of the popularity prediction model becomes equal to or greater than the second threshold, the ratio learning unit 103 stores the calculated intercept of the popularity prediction model in equation (1), the coefficients a1, a2, ..., etc., T1, T2, and T1_init, which is the initial value of T1, in the popularity factor database 125.

[0044] In addition, when the proportion learning unit 103 cannot learn the popularity prediction model due to a reason such as a small number of applicable products included in the second period, the proportion learning unit 103 learns the popularity prediction model according to the popularity of the applicable product in the most recent predetermined number. For example, when the proportion of "1" in the popularity of the applicable product in the most recent predetermined number is equal to or greater than a predetermined threshold, the proportion learning unit 103 sets the intercept of the popularity prediction model of the applicable product to 1 and the coefficient to 0. Furthermore, the oldest date in the popularity of the applicable product is set to T1, and the date on which the popularity prediction model is learned is set to T2. Note that the latest date in the popularity period may be applied as T2. On the other hand, when the proportion of "1" in the popularity of the applicable product in the most recent predetermined number is less than a predetermined threshold, the proportion learning unit 103 sets the intercept of the popularity prediction model of the applicable product to 0 and the coefficient to 0. Furthermore, the oldest date in the popularity of the applicable product is set to T1, and the date on which the popularity prediction model is learned is set to T2. Note that the latest date in the popularity period may be applied as T2.

[0045] 9, the popularity factor database 125 stores the product name, the corresponding intercept of the popularity prediction model, the coefficients T1, T2, and T1_init for each product. The product name is converted into a vector by one-hot encoding and stored.

[0046] Next, the ratio learning unit 103 learns the ratio prediction model 131 based on the sales ratio, the popularity, the environmental information, and the feature information.

[0047] Specifically, first, the ratio learning unit 103 acquires vectorized ingredients, flag information, and a popularity prediction model for each product for each first period from the feature database 123. In addition, the ratio learning unit 103 acquires the popularity of each product for each first period from the popularity database 124.

[0048] Then, the ratio learning unit 103 uses the sales ratio of each product in the first period calculated as described above as the objective variable, and the popularity, vectorized ingredients, and flag information of each product in the corresponding first period, the temperature and precipitation in the corresponding first period, and the sales volume of the product up to the day before the corresponding first period as explanatory variables to learn a ratio prediction model 131 for each product using machine learning. Note that the ratio learning unit 103 may learn the ratio prediction model 131 using, for example, multiple regression analysis. The ratio learning unit 103 stores the learned ratio prediction model 131 in a predetermined storage area.

[0049] The store visit prediction unit 104 predicts the number of visits to a store in a prediction period based on the store visit number prediction model 130, the store visit number information, the environmental information, the sales number information, and the feature information.

[0050] Specifically, the store visit prediction unit 104 loads the number of visits and total sales up to the day before the prediction period acquired by the acquisition unit 101, the temperature and precipitation for the prediction period, and the raw material and flag information converted into vectors for all products scheduled to be sold in the prediction period into the store visit number prediction model 130 learned by the store visit learning unit 102. Then, the store visit prediction unit 104 predicts the number of visits for the prediction period (hereinafter referred to as the "predicted number of visits").

[0051] The ratio prediction unit 105 predicts the sales ratio in the prediction period based on the ratio prediction model 131, the number of store visits information, the environmental information, the sales volume information, and the characteristic information.

[0052] Specifically, first, the ratio prediction unit 105 loads the temperature and precipitation for the prediction target period acquired by the acquisition unit 101 into a popularity prediction model related to the target product for which sales volume is to be predicted, read from the popularity factor database 125. Then, the ratio prediction unit 105 predicts the popularity of the target product for the prediction target period (hereinafter referred to as "predicted popularity").

[0053] Next, the ratio prediction unit 105 loads the predicted popularity, the vectorized ingredients and flag information of the target product in the prediction period, the temperature and precipitation in the prediction period, and the sales volume of the target product up to the day before the prediction period into the ratio prediction model 131 of the target product for the prediction period learned by the ratio learning unit 103. Then, the ratio prediction unit 105 predicts the sales ratio of the target product in the prediction period (hereinafter referred to as the "predicted sales ratio").

[0054] The sales prediction unit 106 predicts the number of sales of the target product based on the predicted number of visitors predicted by the visit prediction unit 104 and the predicted sales ratio predicted by the ratio prediction unit 105. Specifically, the sales prediction unit 106 outputs the product of the predicted number of visitors and the predicted sales ratio to the output unit 107 as the predicted sales number of the target product.

[0055] The output unit 107 displays on the display unit 16 the predicted sales volume of the target product predicted by the sales prediction unit 106 .

[0056] Next, the operation of the demand forecasting device 10 will be described.

[0057] 10 is a flowchart showing the flow of the demand learning process by the demand forecasting device 10. The CPU 11 reads out a demand learning program from the ROM 12 or the storage 14, loads it into the RAM 13, and executes it, thereby performing the demand learning process.

[0058] In step S101, the CPU 11, as the acquisition unit 101, acquires the number of visits, temperature, amount of precipitation, and total sales for the first period, and the number of visits, total sales, and sales for each product up until the day before the first period. The CPU 11, as the acquisition unit 101, also acquires ingredients and flag information used for all products sold in the first period.

[0059] In step S102, the CPU 11, as the store visit learning unit 102, uses machine learning to learn a store visit prediction model 130, with the number of visits in the first period as the objective variable, and the number of visits and total sales up to the day before the first period, the temperature and precipitation in the first period, and the ingredients and flag information of all products sold in the first period as explanatory variables.

[0060] In step S103, the CPU 11, functioning as the rate learning unit 103, calculates the sales rate for the first period by dividing the number of sales of the relevant product in the first period by the corresponding number of visits to the store in the first period.

[0061] In step S104, the CPU 11, as the ratio learning unit 103, determines whether the value obtained by dividing the sales ratio of the relevant product calculated in step S103 by the simple moving average of the sales ratio for the corresponding first period is equal to or greater than the first threshold. If the value obtained by dividing the sales ratio by the simple moving average of the sales ratio is equal to or greater than the first threshold (step S104: YES), the CPU 11 proceeds to step S105. On the other hand, if the value obtained by dividing the sales ratio by the simple moving average of the sales ratio is less than the first threshold (step S104: NO), the CPU 11 proceeds to step S106.

[0062] In step S105, the CPU 11, functioning as the ratio learning unit 103, sets the popularity of the relevant product to 1 and stores the popularity in the popularity database .

[0063] In step S106, the CPU 11, functioning as the ratio learning unit 103, sets the popularity of the corresponding product to 0 and stores the popularity in the popularity database 124.

[0064] In step S107, the CPU 11, functioning as the ratio learning unit 103, acquires, from the popularity factor database 125, the temperature and precipitation during the popular period, and the ingredients and flag information converted into vectors corresponding to the popular period and the relevant product, as popularity factors.

[0065] In step S108, the CPU 11, as the ratio learning unit 103, uses multiple regression analysis to learn a popularity prediction model for the second period, using the popularity of the corresponding product read from the popularity database 124 as the objective variable and the temperature and precipitation during the popular period as the explanatory variables.

[0066] In step S109, CPU 11, functioning as ratio learning unit 103, determines whether the coefficient of determination of the popularity prediction model learned in step S108 is equal to or greater than the second threshold. If the coefficient of determination of the popularity prediction model is equal to or greater than the second threshold (step S109: YES), CPU 11 proceeds to step S111. On the other hand, if the coefficient of determination of the popularity prediction model is less than the second threshold (step S109: NO), CPU 11 proceeds to step S110.

[0067] In step S110, the CPU 11, functioning as the rate learning unit 103, shortens the second period, and the process returns to step S109. Specifically, the CPU 11 adds a predetermined fixed value Tw to T1, which is the start time of the second period.

[0068] In step S111, the CPU 11, as the ratio learning unit 103, learns a ratio prediction model 131 using the sales ratio of the corresponding product in the first period as the objective variable, and the popularity of the corresponding product, the vectorized ingredient and flag information, the temperature and precipitation in the corresponding first period, and the sales volume of the corresponding product up to the day before the corresponding first period as explanatory variables.

[0069] In step S112, the CPU 11, functioning as the proportion learning unit 103, determines whether or not proportion prediction models for all products sold in the first period have been learned. If the CPU 11 has learned proportion prediction models for all products sold in the first period (step S112: YES), the CPU 11 ends this demand learning process. On the other hand, if the CPU 11 has not learned proportion prediction models for all products sold in the first period (step S112: NO), the CPU 11 returns to step S103.

[0070] 11 is a flowchart showing the flow of the demand forecasting process by the demand forecasting device 10. The CPU 11 reads out a demand forecasting program from the ROM 12 or the storage 14, loads it into the RAM 13, and executes it, thereby performing the demand forecasting process.

[0071] In step S201, the CPU 11, as the acquisition unit 101, acquires the number of store visits, the total sales amount, and the sales amount of the target product up to the day before the prediction target period. The CPU 11, as the acquisition unit 101, also acquires the temperature and precipitation amount for the prediction target period, and the ingredients and flag information for all products to be sold in the prediction target period.

[0072] In step S202, the CPU 11, functioning as the store visit prediction unit 104, loads the store visit number and total sales number on the day before the prediction period, the temperature and precipitation amount for the prediction period, and the raw material and flag information converted into vectors for all products scheduled to be sold in the prediction period into the store visit prediction model 130 for the prediction period. Then, the store visit prediction unit 104 predicts the predicted number of store visits.

[0073] In step S203, the CPU 11, functioning as the ratio prediction unit 105, reads the temperature and the amount of precipitation in the prediction target period into the popularity prediction model of the target product in the prediction target period. Then, the CPU 11 predicts the predicted popularity.

[0074] In step S204, the CPU 11, functioning as the ratio prediction unit 105, causes the ratio prediction model 131 of the target product in the prediction target period to read the predicted popularity, the temperature and precipitation in the prediction target period, the sales volume of the target product on the day before the prediction target period, and the ingredients and flag information of the target product in the prediction target period. Then, the ratio prediction unit 105 predicts the predicted sales ratio.

[0075] In step S205, the CPU 11, functioning as the sales forecasting unit 106, forecasts the product of the forecasted number of store visits and the forecasted sales ratio as the forecasted sales volume of the target product.

[0076] In step S206, the CPU 11, functioning as the output unit 107, displays the predicted sales volume of the target product on the display unit 16, and ends this demand forecasting process.

[0077] As described above, the demand prediction device according to this embodiment acquires factor information relating to factors that affect the number of product sales at a store. The demand prediction device predicts the number of visits to the store during a prediction period based on the acquired factor information and a visit number prediction model for predicting the number of visits to the store during a prediction period that has been learned based on visit number information and factor information relating to the number of visits to the store in the past. The demand prediction device predicts a ratio based on the acquired factor information and a ratio prediction model for predicting a ratio that has been learned based on the factor information and the ratio of the number of visits to the store that the number of product sales accounts for. The demand prediction device predicts the number of product sales based on the predicted number of visits and the predicted ratio. This makes it possible to predict the number of product sales at the store.

[0078] Here, in order to predict the sales volume for each product, a method of predicting the number of store visitors in the prior art is applied to each product, and data related to the same product is collected to create a prediction model for each product or for each classification (category) to which the product belongs. However, while the sales volume of a product is strongly influenced by the number of store visitors, the sales volume is also influenced by the popularity of the product. In other words, the reason for a high sales volume of a product is not limited to a high number of store visitors, and the sales volume may be high because the product is popular. In this way, the increase or decrease in the sales volume includes the influence of the increase or decrease in the number of store visitors and the influence of the popularity of each product. As described above, when the method of predicting the number of store visitors in the prior art is applied to each product, it is difficult to distinguish the influence of the number of store visitors from the influence of the popularity, and there is a possibility that the prediction accuracy will decrease when the number of store visitors increases or decreases. On the other hand, according to this embodiment, as described above, by combining the prediction of the number of store visitors with the prediction of the sales ratio equivalent to the popularity, the influence of the increase or decrease in the number of store visitors in the prediction of the sales volume can be separated as the prediction of the number of store visitors, and the accuracy of the sales volume prediction is improved.

[0079] In addition, the demand forecasting device of this embodiment predicts sales ratios using a model that takes into account environmental information, product characteristics, and changes in popularity over time, thereby improving the accuracy of forecasting sales numbers using the sales ratios.

[0080] In addition, since the demand forecasting device according to this embodiment performs learning and predictions using the characteristics of ingredients and other elements that make up the product, it can make predictions even when there are changes in the elements that make up the product, such as when the same menu item has some different ingredients. A secondary effect can be obtained, such as utilizing the characteristics of popular products obtained in the process of predicting sales ratios for the development of new products.

[0081] <Modification> The present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the spirit and scope of the present invention.

[0082] For example, the store visit learning unit 102 and the ratio learning unit 103 may be configured separately from the demand prediction device 10. In other words, the store visit prediction unit 104 may predict the number of store visits to a store in a prediction period based on a store visit number prediction model 130 learned by a device configured separately from the demand prediction device 10 and the factor information acquired by the acquisition unit 101. Furthermore, the ratio prediction unit 105 may predict a sales ratio based on a ratio prediction model learned by a device configured separately from the demand prediction device 10 and the factor information acquired by the acquisition unit 101.

[0083] In the above embodiment, the ratio learning unit 103 learns a popularity prediction model for each product as a popularity prediction model for the product. However, a popularity prediction model for each category to which the product belongs or for each ingredient used in the product may be learned as a popularity prediction model for the product. For example, when learning a popularity prediction model for each ingredient, the ratio learning unit 103 learns a popularity prediction model for each ingredient as in the above embodiment, using the popularity of the product related to the ingredient as the objective variable. Then, the ratio prediction unit 105 integrates the output results of the popularity prediction models for each ingredient corresponding to all ingredients used in the target product during the prediction target period by weighted averaging or the like, and predicts the popularity of the target product. The ratio learning unit 103 may learn a popularity prediction model for some or all of the ingredients included in the product as a popularity prediction model for each ingredient. In this case, the ratio learning unit 103 converts the ingredients corresponding to the product obtained from the feature database 123 into vectors by multi-hot encoding, and adds flag information to the vectors of ingredients. Then, the proportion learning unit 103 may learn a popularity prediction model in the same manner as described above, using some or all of the vectors of the ingredients and the popularity of the product related to the ingredients as objective variables.

[0084] In addition, the ratio prediction unit 105 is not limited to predicting the popularity of the target product from a popularity prediction model of a product or raw material that matches the target product or its raw materials, but may also predict the popularity of the target product from a popularity prediction model of a product or raw material that is similar to the target product or its raw materials.

[0085] In the above embodiment, the environmental information and the characteristic information are used as an example of the factor information, but other factor information may be used. For example, information about an event that has been held or is scheduled to be held in the vicinity of the store may be used as the factor information.

[0086] In the above embodiment, each model may be trained for each day of the week to which a day belongs during the first period. In this case, the number of store visits and sales ratios may be predicted using each model corresponding to the day of the week to which a day belongs during the prediction target period. This makes it possible to predict sales figures that reflect the trends for each day of the week.

[0087] In addition, in the above embodiment, a case has been described in which the number of visits and the number of sales during the prediction period and up until the day before the first period are used for learning and prediction, but it is not essential to use this information for learning and prediction.

[0088] In addition, various processes that the CPU reads and executes software (programs) in the above embodiment may be executed by various processors other than the CPU. Examples of processors in this case include PLDs (Programmable Logic Devices) such as FPGAs (Field-Programmable Gate Arrays) whose circuit configuration can be changed after manufacture, and dedicated electric circuits such as ASICs (Application Specific Integrated Circuits) that are processors having circuit configurations designed exclusively for executing specific processes. The demand forecasting process may be executed by one of these various processors, or may be executed by a combination of two or more processors of the same or different types (e.g., multiple FPGAs, and a combination of a CPU and an FPGA). The hardware structure of these various processors is, more specifically, an electric circuit that combines circuit elements such as semiconductor elements.

[0089] In the above embodiment, the demand forecasting program is stored (installed) in advance in the storage 14, but the present invention is not limited to this. The program may be provided in a form stored in a non-transitory storage medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a USB (Universal Serial Bus) memory. The program may also be downloaded from an external device via a network.

[0090] The following supplementary notes are further disclosed regarding the above embodiment.

[0091] (Additional note 1) A demand forecasting device, Memory, at least one processor coupled to the memory; Including, The processor, Obtaining factor information regarding factors that affect the sales volume of products in a store; predicting the number of visits to the store during the prediction period based on a visit number prediction model for predicting the number of visits to the store during the prediction period, the visit number prediction model being learned based on visit number information regarding the number of visits to the store in the past and the factor information, and the factor information acquired by the acquisition unit; predicting the ratio based on a ratio prediction model for predicting the ratio, the ratio being learned based on the factor information and the ratio of the number of sales of the product to the number of store visits, and the factor information acquired by the acquisition unit; The number of sales of the product is predicted based on the number of store visits predicted by the store visit prediction unit and the ratio predicted by the ratio prediction unit. The demand forecasting device is configured as follows.

[0092] (Additional note 2) A non-transitory storage medium storing a program executable by a computer to execute a demand forecasting process, The demand forecasting process includes: Obtaining factor information regarding factors that affect the sales volume of products in a store; predicting the number of visits to the store during the prediction period based on a visit number prediction model for predicting the number of visits to the store during the prediction period, the visit number prediction model being learned based on visit number information regarding the number of visits to the store in the past and the factor information, and the factor information acquired by the acquisition unit; predicting the ratio based on a ratio prediction model for predicting the ratio, the ratio being learned based on the factor information and the ratio of the number of sales of the product to the number of store visits, and the factor information acquired by the acquisition unit; The number of sales of the product is predicted based on the number of store visits predicted by the store visit prediction unit and the ratio predicted by the ratio prediction unit. Non-transitory storage medium. [Explanation of symbols]

[0093] 10 Demand forecasting device 11 CPU 12 ROM 13 RAM 14. Storage 16 Display 17 Communication I / F 19 Bus 101 Acquisition Department 102 Visiting Learning Department 103 Proportion Learning Section 104 Store Visitor Forecast Department 105 Percentage Prediction Section 106 Sales Forecast Department 130 Store visitor forecast model 131 Percentage Prediction Model

Claims

1. an acquisition unit for acquiring factor information relating to factors that affect the number of products sold at a store; a store visit prediction unit that predicts the number of visits to the store in a prediction target period based on a store visit prediction model for predicting the number of visits to the store in a prediction target period, the store visit prediction model being learned based on store visit number information regarding the number of visits to the store in the past and the factor information, and the factor information acquired by the acquisition unit; a ratio prediction unit that predicts the ratio based on a ratio prediction model for predicting the ratio, the ratio being learned based on the past factor information and the ratio of the number of sales of the product to the number of visits to the store, and the factor information acquired by the acquisition unit; a sales prediction unit that predicts the number of sales of the product based on the number of visitors predicted by the store visit prediction unit and the ratio predicted by the ratio prediction unit; Equipped with The ratio prediction unit predicts a popularity indicating the relative size of the ratio based on the factor information acquired by the acquisition unit, and predicts the ratio based on the factor information acquired by the acquisition unit, the predicted popularity, and the ratio prediction model.

2. The demand forecasting device according to claim 1 , wherein the factor information includes at least one of environmental information relating to an external environment around the store and characteristic information relating to characteristics of the products.

3. The acquisition unit acquires the past visit number information, The store visit prediction unit predicts the number of store visits based on the factor information and the store visit number information acquired by the acquisition unit and the store visit number prediction model. The demand forecasting device according to claim 1 or 2.

4. A demand prediction device as described in any one of claims 1 to 3, wherein the percentage prediction unit predicts the popularity based on a popularity prediction model for predicting the popularity learned based on the past factor information and the popularity, and the factor information acquired by the acquisition unit, and predicts the percentage based on the factor information acquired by the acquisition unit, the predicted popularity, and the percentage prediction model.

5. A store visit learning unit that learns the store visit number prediction model based on the store visit number information and the factor information; a ratio learning unit that calculates the ratio based on the past visit number information and the past sales number information related to the sales number of the product, and learns the ratio prediction model based on the past factor information and the calculated ratio; The demand forecasting device according to any one of claims 1 to 4, comprising:

6. A store visit learning unit that learns the store visit number prediction model based on the store visit number information and the factor information; a ratio learning unit that calculates the ratio based on the store visitor information during a specified period and sales number information related to past sales numbers of the product, learns the popularity prediction model based on the past factor information and the popularity corresponding to the calculated ratio, and if the learned popularity prediction model does not satisfy a predetermined condition, shortens the specified period and re-learns the popularity prediction model, and learns the ratio prediction model based on the calculated ratio, the past factor information, and the popularity corresponding to the ratio; The demand forecasting device according to claim 4 .

7. The acquisition unit acquires factor information regarding factors that affect the sales volume of the product at the store, a store visit prediction unit predicts the number of store visits to the store in the prediction period based on a store visit prediction model for predicting the number of store visits to the store in the prediction period, the store visit prediction model being learned based on store visit number information regarding the number of past visits to the store and the factor information, and based on the factor information acquired by the acquisition unit; a ratio prediction unit predicts the ratio based on a ratio prediction model for predicting the ratio, the ratio being learned based on the factor information and the ratio of the number of sales of the product to the number of store visits, and the factor information acquired by the acquisition unit; a sales prediction unit predicts the number of sales of the product based on the number of store visits predicted by the store visit prediction unit and the ratio predicted by the ratio prediction unit; Including, The prediction by the ratio prediction unit predicts a popularity indicating a relative size of the ratio based on the factor information acquired by the acquisition unit, and predicts the ratio based on the factor information acquired by the acquisition unit, the predicted popularity, and the ratio prediction model. Demand forecasting methods.

8. A program for causing a computer to function as each unit of the demand forecasting device according to any one of claims 1 to 6.

Images