Demand prediction method, demand prediction system, training method, and training system

Abstract

In a demand prediction method according to the present invention, a first demand amount of an article of interest before a prescribed time point is acquired, a first order amount of the article of interest at the prescribed time point is acquired, and a first weight α2 indicating the probability of the first order amount is acquired. In the demand prediction method, a first prediction amount indicating a demand amount of the article of interest after the prescribed time point is calculated on the basis of the first demand amount. In the demand prediction method, a second prediction amount indicating a demand amount of the article of interest after the prescribed time point is calculated on the basis of the first order amount, the first weight, and the first prediction amount.

Description

Demand forecasting method, demand forecasting system, learning method, and learning system

[0001] The present disclosure relates to a demand forecasting method for forecasting demand for an item.

[0002] Patent Document 1 discloses an ordering support method for supporting the ordering of parts used in the manufacture of a product.

[0003] JP 2018-73338 A

[0004] The present disclosure provides a demand forecasting method and the like that can easily improve the accuracy of forecasting demand for goods.

[0005] A demand forecasting method according to one aspect of the present disclosure includes: acquiring a first demand quantity for a target item prior to a predetermined time point; acquiring a first order quantity for the target item at the predetermined time point; and acquiring a first weight representing likelihood of the first order quantity. The demand forecasting method calculates a first forecast quantity indicating the demand quantity for the target item after the predetermined time point based on the first demand quantity. The demand forecasting method calculates a second forecast quantity indicating the demand quantity for the target item after the predetermined time point based on the first order quantity, the first weight, and the first forecast quantity.

[0006] A demand forecasting system according to one aspect of the present disclosure includes a first acquisition unit, a second acquisition unit, a third acquisition unit, a first prediction unit, and a second prediction unit. The first acquisition unit acquires a first demand quantity of a target item prior to a predetermined time point. The second acquisition unit acquires a first order quantity of the target item at the predetermined time point. The third acquisition unit acquires a first weight representing the likelihood of the first order quantity. The first prediction unit calculates a first predicted quantity indicating the demand quantity of the target item after the predetermined time point based on the first demand quantity. The second prediction unit calculates a second predicted quantity indicating the demand quantity of the target item after the predetermined time point based on the first order quantity, the first weight, and the first predicted quantity.

[0007] A learning method according to one aspect of the present disclosure includes: acquiring a second demand quantity of a target item at a predetermined time point; acquiring a second order quantity of the target item prior to the predetermined time point; and acquiring a predicted quantity of the second demand quantity based on a first demand quantity of the target item prior to the predetermined time point. The learning method includes learning a first weight representing a likelihood of the second order quantity based on the second demand quantity, the second order quantity, and the predicted quantity.

[0008] A learning system according to one aspect of the present disclosure includes a fourth acquisition unit, a fifth acquisition unit, a sixth acquisition unit, and a weight learning unit. The fourth acquisition unit acquires a second demand quantity of a target item at a predetermined time. The fifth acquisition unit acquires a second order quantity of the target item prior to the predetermined time. The sixth acquisition unit acquires a predicted quantity of the second demand quantity based on a first demand quantity of the target item prior to the predetermined time. The weight learning unit learns a first weight representing the likelihood of the second order quantity based on the second demand quantity, the second order quantity, and the predicted quantity.

[0009] The demand forecasting method according to one aspect of the present disclosure has the advantage of easily improving the accuracy of forecasting demand for goods.

[0010] FIG. 1 is a block diagram showing a schematic configuration of a demand forecasting system according to an embodiment. FIG. 2 is an explanatory diagram illustrating an overview of forecasting based on a first demand quantity. FIG. 3 is an explanatory diagram illustrating a specific example of a first operation of the demand forecasting system according to an embodiment. FIG. 4 is an explanatory diagram illustrating a specific example of a second operation of the demand forecasting system according to an embodiment. FIG. 5 is an explanatory diagram illustrating a specific example of a third operation of the demand forecasting system according to an embodiment. FIG. 6 is an explanatory diagram illustrating another specific example of the third operation of the demand forecasting system according to an embodiment. FIG. 7 is a block diagram showing a schematic configuration of a learning system according to an embodiment. FIG. 8 is an explanatory diagram illustrating a specific example of learning of a first weight and a second weight by the learning system according to an embodiment. FIG. 9 is an explanatory diagram illustrating a specific example of learning of a third weight by the learning system according to an embodiment. FIG. 10 is a flowchart showing a first operation of the demand forecasting system according to an embodiment. FIG. 11 is a flowchart showing a second operation of the demand forecasting system according to an embodiment. FIG. 12 is a flowchart showing a third operation of the demand forecasting system according to an embodiment. FIG. 13 is a flowchart showing an example of generating a trained model by the learning system according to an embodiment. FIG. 14 is a flowchart showing an example of learning of a first weight by the learning system according to an embodiment. FIG. 15 is a flowchart showing an example of learning of a second weight by the learning system according to an embodiment. Fig. 16 is a flowchart showing an example of learning of the third weight by the learning system according to the embodiment Fig. 17 is a diagram illustrating advantages of the demand forecasting system according to the embodiment.

[0011] Hereinafter, the embodiments will be specifically described with reference to the drawings.

[0012] The embodiments described below are all comprehensive or specific examples. The numerical values, shapes, components, arrangement positions and connection forms of the components, etc. shown in the following embodiments are merely examples and are not intended to limit the present disclosure. Furthermore, among the components in the following embodiments, components that are not recited in the independent claims are described as optional components.

[0013] Furthermore, each figure is a schematic diagram and is not necessarily an exact illustration. Therefore, for example, the scales and the like do not necessarily match in each figure. Furthermore, in each figure, substantially the same configurations are assigned the same reference numerals, and duplicate explanations are omitted or simplified.

[0014] Furthermore, in this specification, terms indicating relationships between elements such as the same, as well as numerical values ​​and numerical ranges, are not expressions that only express a strict meaning, but are expressions that also include a substantially equivalent range, for example, a difference of about several percent (e.g., about 10%).

[0015] (Embodiment) [1. Configuration of Demand Forecasting System] A demand forecasting system 1 according to an embodiment will be described below with reference to Fig. 1. Fig. 1 is a block diagram showing a schematic configuration of the demand forecasting system 1 according to an embodiment. The demand forecasting system 1 is a system for forecasting future demand for goods. Here, the demand may include, for example, demand for the sale of goods, the rental of goods, or the manufacture of goods. In the following description, it is assumed that the demand is demand for the sale of goods.

[0016] The demand forecasting system 1 has a memory and a processor, and various functions are realized by the processor executing programs stored in the memory. The demand forecasting system 1 is, for example, a laptop or desktop personal computer. Note that the demand forecasting system 1 may also be a portable information processing terminal such as a smartphone or tablet terminal, as long as it has a memory and a processor.

[0017] The demand forecasting system 1 includes a first acquisition unit 11, a second acquisition unit 12, a third acquisition unit 13, a first prediction unit 14, and a second prediction unit 15. The demand forecasting system 1 is configured to be able to communicate with the database 2 via a wide area network such as the Internet. The demand forecasting system 1 may be configured to be able to communicate with the database 2 via a local area network (LAN), for example, or may include the database 2.

[0018] Database 2 stores information about demand for a product at a predetermined time (the present time) and before the predetermined time. Here, the present time is the month to which the day the demand forecasting system 1 (or the learning system 3 described below) is used belongs. In the embodiment, database 2 stores, for each product, the actual sales volume of the product before the present time, the order volume of the product before the present time, the current order volume of the product, etc. Furthermore, if there are multiple customers for the product, database 2 also stores, for each customer, the actual sales volume of the product before the present time, the order volume of the product before the present time, the current order volume of the product, etc.

[0019] The first acquisition unit 11 acquires a first demand amount for a target item prior to a predetermined time point (the present time). Here, the target item is an item for which future demand is predicted by the demand forecasting system 1. In the embodiment, the first acquisition unit 11 acquires, as the first demand amount, the actual sales amount of the target item for the most recent m months (m is a natural number) prior to the present time of the target item from the database 2. Note that the first demand amount is not limited to the actual past sales amount of the target item on a monthly basis, but may also be the actual past sales amount of the target item on a daily, weekly, or yearly basis, for example.

[0020] The second acquisition unit 12 acquires a first order quantity of the target item at a predetermined time (current time). The first order quantity is the order quantity of the target item that has been confirmed at the predetermined time (current time). In the embodiment, the second acquisition unit 12 acquires the current order quantity of the target item from the database 2 as the first order quantity.

[0021] The third acquisition unit 13 acquires weights to be used in calculations by the second prediction unit 15, which will be described later. Specifically, the third acquisition unit 13 acquires the weights by reading out weights that are pre-stored in a memory. Note that the third acquisition unit 13 may acquire weights that are pre-stored in a storage device external to the demand forecasting system 1 by communicating with the storage device. In the embodiment, the weights are determined by machine learning using the learning system 3, which will be described later, but may also be set appropriately based on, for example, a user's rule of thumb.

[0022] In the embodiment, when the demand forecasting system 1 executes the first operation (described later), the third obtainer 13 obtains a first weight α2 representing the likelihood of the first order quantity. Here, the first order quantity is the order quantity of the target item that is determined at a predetermined time (the present time). However, due to factors such as the production yield of the target item or deficiencies in the equipment used to manufacture the target item, it is not necessarily possible to deliver the entire order quantity to the customer at a time after the predetermined time (the delivery date). In other words, there is uncertainty in the first order quantity at the predetermined time. Therefore, in the embodiment, the third obtainer 13 obtains the first weight α2 to predict the demand quantity of the target item at a time after the predetermined time, taking this uncertainty into account.

[0023] Furthermore, when the demand forecasting system 1 executes a second operation (described later), the third acquirer 13 acquires, in addition to the first weight α2, a second weight α1 representing the likelihood of the first predicted quantity, which indicates the demand quantity of the target item at a predetermined time point and thereafter. Here, the first predicted quantity is calculated by the first predictor 14 (described later) based on past sales records of the target item prior to the predetermined time point (the present time). However, the first predicted quantity is not an amount that takes into account factors other than the past sales records of the target item, and therefore, uncertainty exists. Therefore, in the embodiment, the third acquirer 13 acquires the second weight α1 when the second operation is executed in order to predict the demand quantity of the target item at a time point and thereafter, taking this uncertainty into account.

[0024] Furthermore, when the demand forecasting system 1 executes a third operation (described later), the third acquisition unit 13 acquires a third weight β in addition to the first weight α2 and the second weight α1. The third weight β represents a change in demand depending on whether the target item is a discontinued item whose supply will end. Here, when the target item is a discontinued item, demand increases from a predetermined point in time (the present time) until the supply of the target item ends, compared to when the target item is not a discontinued item, which may result in so-called last-minute demand. Therefore, in the embodiment, the third acquisition unit 13 acquires the third weight β when executing the third operation in order to predict the demand amount of the target item at a point in time after the predetermined point in time, taking this last-minute demand into account.

[0025] The first prediction unit 14 calculates a first predicted amount indicating the demand amount of the target item from a predetermined point in time onward, based on the first demand amount acquired by the first acquisition unit 11. In the embodiment, the first prediction unit 14 predicts the sales amount of the target item for each month from month N to month N+n (n is a natural number) based on the actual sales amount of the target item for the most recent m months prior to the current N months (N is a natural number between 1 and 12), as shown in Fig. 2. Fig. 2 is an explanatory diagram illustrating an overview of prediction based on the first demand amount.

[0026] For example, when m = 3, n = 12, and N = 6, the first prediction unit 14 predicts the sales volume of the target item for each month from June to June of the following year based on the actual sales volume of the target item from March to May. Note that when the first acquisition unit 11 acquires the past sales volume of the target item on a daily, weekly, or yearly basis, for example, the first prediction unit 14 predicts the sales volume of the target item on a daily, weekly, or yearly basis.

[0027] The first prediction unit 14 can calculate the first forecast amount by using, for example, a mean shift method, an autoregressive (AR) model, a time series method, a machine learning method, or a combination of these methods. The time series method may include, for example, an autoregressive integrated moving average (ARIMA) model, a seasonal autoregressive integrated moving average (SARIMA) model that adds a seasonal periodic pattern to the ARIMA model, a Holt-Winter's model, a Croston model, or a Prophet model. The machine learning method may include, for example, a random forest, eXtreme Gradient Boosting (XGBoost), a gradient boosting decision tree (GBDT), a support vector machine (SVM), or a neural network.

[0028] In the embodiment, the first prediction unit 14 calculates the first predicted amount using the trained model 141. The trained model 141 is a model trained by machine learning so as to input the first demand amount and output the first predicted amount.

[0029] The second prediction unit 15 calculates a second predicted quantity indicating the demand quantity of the target item after a predetermined time point, based on at least the first order quantity acquired by the second acquisition unit 12, the first weight α2 acquired by the third acquisition unit 13, and the first predicted quantity predicted by the first prediction unit 14. In other words, the second prediction unit 15 calculates the second predicted quantity by correcting the first predicted quantity based on at least the first order quantity, the first weight α2, and the first predicted quantity.

[0030] In the embodiment, the second prediction unit 15 calculates the second predicted amount by any one of the following first, second, and third operations. Note that which of the first, second, and third operations the second prediction unit 15 will execute may be set in advance or may be set appropriately by the user.

[0031] [1-1. First Operation] In the first operation, the second prediction unit 15 calculates the second predicted amount based only on the first order amount, the first weight α2, and the first predicted amount. Specifically, in the first operation, the second prediction unit 15 calculates the second predicted amount using the following equation (1):

[0032] <Second forecast quantity> = <First forecast quantity> + α2 × <First order quantity> (1) Figure 3 is an explanatory diagram of a specific example of the first operation of the demand forecasting system 1 according to an embodiment. Figure 3(a) is a diagram showing a calculation process of the second forecast quantity of the target item "KKK" for customer "A" with December 2022 as the current time. Figure 3(b) is a diagram showing the first order quantity of the target item "KKK" by customer "A" with December 2022 as the current time. In the example shown in Figure 3(a), "forecast quantity based on past performance" represents the first forecast quantity, "confirmed order quantity" represents the first order quantity, "α2" represents the first weight α2, and "forecast quantity" represents the second forecast quantity.

[0033] 3(a), for example, when the forecast target year and month is January 2023, the first order quantity for that month is 0, and therefore the second prediction unit 15 calculates the second predicted quantity as 30 + 0.5 × 0 = 30 units using formula (1). Also, in the example shown in FIG. 3(a), when the forecast target year and month is February 2023, the first order quantity for that month is 50 units, and therefore the second prediction unit 15 calculates the second predicted quantity as 40 + 0.5 × 50 = 65 units using formula (1).

[0034] As described above, in the first operation, the second forecasted quantity is calculated taking into account the uncertainty of the first order quantity, and therefore the forecast accuracy of the second forecasted quantity is likely to be improved compared to when the uncertainty of the first order quantity is not taken into account.

[0035] [1-2. Second Operation] In the second operation, the second prediction unit 15 calculates the second predicted amount based on the second weight α1 in addition to the first order amount, the first weight α2, and the first predicted amount. Specifically, in the second operation, the second prediction unit 15 calculates the second predicted amount using the following equation (2):

[0036] <Second forecast quantity> = α1 × <First forecast quantity> + α2 × <First order quantity> (2) Figure 4 is an explanatory diagram of a specific example of the second operation of the demand forecasting system 1 according to the embodiment. Figure 4(a) is a diagram illustrating the calculation process of the second forecast quantity of the target item "KKK" for customer "A" with December 2022 as the current time, similar to the example shown in Figure 3(a). Figure 4(b) is a diagram illustrating the first order quantity of the target item "KKK" by customer "A" with December 2022 as the current time, similar to the example shown in Figure 3(b). Note that in the example shown in Figure 4(a), unlike the example shown in Figure 3(a), "α1" represents the second weight α1.

[0037] 4(a), for example, when the forecast target year and month is January 2023, the first order quantity for that month is 0, and therefore the second prediction unit 15 calculates the second predicted quantity as 0.8×30+0.7×0=24 units using formula (2). Also, in the example shown in FIG. 4(a), when the forecast target year and month is February 2023, the first order quantity for that month is 50 units, and therefore the second prediction unit 15 calculates the second predicted quantity as 0.8×40+0.7×50=67 units using formula (2).

[0038] As described above, in the second operation, the second predicted amount is calculated taking into account the uncertainty of the first predicted amount compared to the first operation, and therefore the prediction accuracy of the second predicted amount is more likely to be improved than in the first operation.

[0039] [1-3. Third Operation] In the third operation, the second prediction unit 15 calculates the second predicted amount based on the first order volume, the first weight α2, the second weight α1, the first predicted amount, and the third weight β. Specifically, in the third operation, the second prediction unit 15 calculates the second predicted amount using the following equation (3):

[0040] <Second forecast quantity> = β × (α1 × <First forecast quantity> + α2 × <First order quantity>) (3) FIG. 5 is an explanatory diagram of a specific example of the third operation of the demand forecasting system 1 according to the embodiment. Unlike the example shown in FIG. 3(a), FIG. 5(a) is a diagram illustrating the calculation process of the second forecast quantity of the target item "AAA" for customer "J" with December 2022 as the current time. Unlike the example shown in FIG. 3(b), FIG. 5(b) is a diagram illustrating the first order quantity of the target item "AAA" by customer "J" with December 2022 as the current time. Unlike the example shown in FIG. 3(b), FIG. 5(c) is a diagram illustrating that the target item "AAA" is a discontinued item, the "order end date" when orders for the discontinued item will end, and the "production end date" when production of the discontinued item will end. Note that in the example shown in FIG. 5(a), unlike the example shown in FIG. 4(a), "β" represents the third weight β.

[0041] 5(a), for example, when the prediction target year and month is January 2023, the target item is a discontinued item and it is predicted that a rush of demand will occur, so the second prediction unit 15 calculates the second predicted amount as 2×(0.8×30+0.7×0)=48 units using formula (3). Also, in the example shown in FIG. 5(a), when the prediction target year and month is February 2023, the target item is also a discontinued item and it is predicted that a rush of demand will occur, so the second prediction unit 15 calculates the second predicted amount as 2×(0.8×40+0.7×50)=134 units using formula (3).

[0042] FIG. 6 is an explanatory diagram of another specific example of the third operation of the demand forecasting system 1 according to the embodiment. (a) of FIG. 6 is a diagram illustrating a calculation process of the second forecast quantity of the target item "BBB" for customer "J" with December 2022 as the current time, which is different from the example shown in (a) of FIG. 5. (b) of FIG. 6 is a diagram illustrating the first order quantity of the target item "BBB" by customer "J" with December 2022 as the current time, which is different from the example shown in (b) of FIG. 5. (c) of FIG. 6 is a diagram illustrating that the target item "BBB" is a discontinued item, which is different from the example shown in (c) of FIG. 5.

[0043] In the example shown in (a) of FIG. 6, unlike the example shown in (a) of FIG. 5, orders for the target item "BBB," which is currently a discontinued product, have ended. Therefore, in the example shown in (a) of FIG. 6, the second prediction unit 15 calculates 50 units, which is the first order quantity at the time of delivery (here, February 2023), as the second predicted quantity. Specifically, the second prediction unit 15 calculates the second predicted quantity as 1×(0×40+1×50)=50 units using equation (3).

[0044] 6A, the delivery deadline for the target item "BBB" is the point when a predetermined period (e.g., six months) has passed since the production end date, and there is no demand for the target item "BBB" after the delivery deadline. Therefore, in the example shown in FIG. 6A, the second prediction unit 15 calculates the second predicted quantity after the delivery deadline as 0 units.

[0045] Therefore, in the example shown in (a) of Figure 6, if the prediction target year and month is January 2023, the second prediction unit 15 calculates the second prediction amount as 1 x (0 x 30 + 1 x 0) = 0 using equation (3).

[0046] As described above, in the third operation, the second predicted quantity is calculated by further taking into account demand depending on whether the target item is a discontinued item, compared to the second operation, and therefore the prediction accuracy of the second predicted quantity is more likely to be improved than in the second operation.

[0047] 2. Configuration of the Learning System Next, a learning system 3 according to an embodiment will be described with reference to Fig. 7. Fig. 7 is a block diagram showing a schematic configuration of the learning system 3 according to an embodiment. The learning system 3 is a system for learning the trained model 141, the first weight α2, the second weight α1, and the third weight β used in the demand forecasting system 1 through machine learning.

[0048] The learning system 3 has a memory and a processor, and various functions are realized by the processor executing programs stored in the memory. The learning system 3 is, for example, a laptop or desktop personal computer. The demand forecasting system 1 only needs to have a memory and a processor, and may be, for example, a portable information processing terminal such as a smartphone or tablet terminal.

[0049] The learning system 3 includes a fourth acquisition unit 31, a fifth acquisition unit 32, a sixth acquisition unit 33, a seventh acquisition unit 34, a learning unit 35, and a weight learning unit 36. The learning system 3 is configured to be able to communicate with the database 2 via a wide area network such as the Internet. The learning system 3 may be configured to be able to communicate with the database 2 via a LAN, for example, or may include the database 2.

[0050] The fourth acquisition unit 31 acquires the second demand amount of the target item at a predetermined time (current time). Here, the target item is an item that is the target of the learning system 3. In the embodiment, the fourth acquisition unit 31 acquires the current actual sales volume of the target item from the database 2 as the second demand amount. Note that the second demand amount is not limited to the current actual sales volume of the target item on a monthly basis, but may also be the current actual sales volume of the target item on a daily, weekly, or yearly basis, for example.

[0051] The fifth acquisition unit 32 acquires a second order quantity for the target item prior to a predetermined time point (the current time point). The second order quantity is an order quantity for the target item that was confirmed at a time point prior to the current time point. In this embodiment, the fifth acquisition unit 32 acquires the order quantity for the target item prior to the current time point from the database 2 as the second order quantity.

[0052] The sixth acquisition unit 33 acquires a forecast amount of the second demand amount based on the first demand amount for the target item at a time before the predetermined time (the current time). In other words, the sixth acquisition unit 33 acquires the first forecast amount for a time before the current time by the first prediction unit 14 of the demand forecasting system 1. In the embodiment, the sixth acquisition unit 33 acquires the forecast amount of the second demand amount by acquiring the first forecast amount for a time before the current time from the database 2.

[0053] The seventh acquisition unit 34 acquires discontinued product information indicating whether the target product was a discontinued product at a time point prior to a predetermined time point (the current time point). In the embodiment, the seventh acquisition unit 34 acquires the discontinued product information from the database 2.

[0054] The learning unit 35 uses machine learning to learn the trained model 141 used by the first prediction unit 14 of the demand forecasting system 1. In the embodiment, the learning unit 35 generates the trained model 141 using any one or more of the methods described in the description of the first prediction unit 14 of the demand forecasting system 1. The trained model 141 receives as input the actual sales volume of the target item for the most recent m months prior to month N, which is a predetermined point in time (the current point in time), and outputs a predicted sales volume of the target item for each month from month N to month N+n.

[0055] The weight learning unit 36 ​​learns, by machine learning, the first weight α2, the second weight α1, and the third weight β used in the second forecasting unit 15 of the demand forecasting system 1. Below, a learning example of the first weight α2 and the second weight α1, and a learning example of the third weight β will be described.

[0056] [2-1. Example of Learning First Weight and Second Weight] The weight learning unit 36 ​​learns the first weight α2 and the second weight α1 based on the second demand amount acquired by the fourth acquisition unit 31, the second order amount acquired by the fifth acquisition unit 32, and the forecast amount of the second demand amount acquired by the sixth acquisition unit 33. In the embodiment, the weight learning unit 36 ​​learns the first weight α2 and the second weight α1 by machine learning using regression. Note that the machine learning method used by the weight learning unit 36 ​​is not limited to regression, and other methods may be used.

[0057] 8 is an explanatory diagram of a specific example of learning of the first weight α2 and the second weight α1 by the learning system 3 according to the embodiment. FIG. 8 is a diagram showing data used for learning the first weight α2 and the second weight α1 of "N+2" (in other words, the first weight α2 and the second weight α1 used for prediction two months from now) with December 2022 as the current time. In the example shown in FIG. 8 , the "forecast quantity" in the "forecast value based on past performance" represents the forecast quantity of the second demand quantity, the "confirmed order quantity" in the "confirmed order" represents the second order quantity, and the "sales quantity" in the "sales performance" represents the second demand quantity.

[0058] In the example shown in Fig. 8, the weight learning unit 36 ​​uses the data shown in Fig. 8 to learn the first weight α2 and the second weight α1 to be used for prediction two months later by regression, without distinguishing between customers and items. Note that the weight learning unit 36 ​​may learn the first weight α2 and the second weight α1 for each period to be predicted (e.g., three months later). Furthermore, the weight learning unit 36 ​​may learn the first weight α2 and the second weight α1 for each customer, for each type of item, or for each item.

[0059] [2-2. Learning Example of Third Weight] The weight learning unit 36 ​​learns the third weight β based on the second demand quantity, the second order quantity, the predicted second demand quantity, the first weight α2, and the second weight α1 of the discontinued item of the target item. That is, the weight learning unit 36 ​​learns the third weight β after learning the first weight α2 and the second weight α1. In the embodiment, the weight learning unit 36 ​​learns the third weight β by machine learning using regression. Note that the machine learning method used by the weight learning unit 36 ​​is not limited to regression, and other methods may also be used.

[0060] FIG. 9 is an explanatory diagram of a specific example of learning the third weight β by the learning system 3 according to the embodiment. FIG. 9 is a diagram showing data used to learn the third weight β of "N+2" (in other words, the third weight β used for predicting two months from now), with December 2022 as the current time. In the example shown in FIG. 9 , unlike the example shown in FIG. 8 , "α1" represents the second weight α1, and "α2" represents the first weight α2. Furthermore, the data shown in FIG. 9 is data in which the target items are limited to discontinued items in the past. The weight learning unit 36 ​​can narrow the target items to discontinued items in the past by referring to the discontinued item information acquired by the seventh acquisition unit 34.

[0061] 9 , the weight learning unit 36 ​​uses the data shown in FIG. 9 to learn the third weight β to be used for prediction two months later by regression, without distinguishing between customers and items. Note that the weight learning unit 36 ​​may learn the third weight β for each period to be predicted (e.g., three months later). Furthermore, the weight learning unit 36 ​​may learn the third weight β for each customer, for each type of item, or for each item, for example.

[0062] [3. Operation] An example of operation of the demand forecasting system 1 and an example of operation of the learning system 3 according to the embodiment will be described below.

[0063] [3-1. Example of Operation of Demand Forecasting System] First, a first operation of the demand forecasting system 1 according to the embodiment will be described. FIG. 10 is a flowchart showing the first operation of the demand forecasting system 1 according to the embodiment. First, the first acquisition unit 11 of the demand forecasting system 1 acquires a first demand amount for the target item prior to a predetermined time point (the current time) (S11). Next, the first prediction unit 14 of the demand forecasting system 1 acquires a trained model 141 (S12). Here, the first prediction unit 14 acquires the trained model 141 by reading the trained model 141 pre-stored in memory. Note that steps S11 and S12 may be executed in reverse order or in parallel. Then, the first prediction unit 14 calculates a first forecast amount indicating the demand amount for the target item after the predetermined time point (the current time) based on the first demand amount using the trained model 141 (S13), and stores the calculated first forecast amount in the database 2 (S14).

[0064] Next, the second acquisition unit 12 of the demand forecasting system 1 acquires the first order quantity of the target item at a predetermined time (current time) (S15). Furthermore, the third acquisition unit 13 of the demand forecasting system 1 acquires the first weight α2 (S16). Note that steps S15 and S16 may be executed before step S14. Then, the second prediction unit 15 of the demand forecasting system 1 calculates the second predicted quantity by substituting the first order quantity, the first weight α2, and the first predicted quantity into equation (1) (S17).

[0065] Next, a second operation of the demand forecasting system 1 according to the embodiment will be described. Fig. 11 is a flowchart showing the second operation of the demand forecasting system 1 according to the embodiment. Note that steps S21 to S25 in Fig. 11 are the same as steps S11 to S15 in the first operation (see Fig. 10), and therefore will not be described here.

[0066] The third acquisition unit 13 of the demand forecasting system 1 acquires the first weight α2 and the second weight α1 (S26). Note that step S26 may be executed before step S25. Then, the second forecasting unit 15 of the demand forecasting system 1 calculates the second forecasted amount by substituting the first order amount, the first weight α2, the second weight α1, and the first forecasted amount into equation (2) (S27).

[0067] Next, a third operation of the demand forecasting system 1 according to the embodiment will be described. Fig. 12 is a flowchart showing the third operation of the demand forecasting system 1 according to the embodiment. Note that steps S31 to S35 in Fig. 12 are the same as steps S11 to S15 in the first operation (see Fig. 10), and therefore will not be described here.

[0068] The third acquisition unit 13 of the demand forecasting system 1 acquires the first weight α2, the second weight α1, and the third weight β (S36). Note that step S36 may be executed before step S35. Then, the second forecasting unit 15 of the demand forecasting system 1 calculates the second forecasted amount by substituting the first order quantity, the first weight α2, the second weight α1, the third weight β, and the first forecasted amount into equation (3) (S37).

[0069] [3-2. Example of operation of the learning system] First, an example of generation of a trained model 141 by the learning system 3 according to the embodiment will be described. FIG. 13 is a flowchart showing an example of generation of a trained model 141 by the learning system 3 according to the embodiment. First, the learning unit 35 of the learning system 3 acquires a first demand amount for the target item prior to a predetermined time point (the current time point) (S41). Here, the learning unit 35 acquires the first demand amount from the database 2. Then, the learning unit 35 generates the trained model 141 by learning through machine learning based on the acquired first demand amount (S42).

[0070] Next, a learning example of the first weight α2 by the learning system 3 according to the embodiment will be described. FIG. 14 is a flowchart showing a learning example of the first weight α2 by the learning system 3 according to the embodiment. First, the sixth acquisition unit 33 of the learning system 3 acquires a predicted amount of the second demand based on the first demand prior to a predetermined time (the current time) for the target item (S51). The fifth acquisition unit 32 of the learning system 3 acquires a second order amount prior to the predetermined time (the current time) for the target item (S52). The fourth acquisition unit 31 of the learning system 3 acquires the second demand at the predetermined time (the current time) for the target item (S53). Note that steps S51 to S53 are not limited to this order and may be executed in another order or in parallel.

[0071] Then, the weight learning unit 36 ​​of the learning system 3 learns the first weight α2 by, for example, regression based on the acquired second demand, second order, and predicted second demand (S54).

[0072] Next, an example of learning the second weight α1 by the learning system 3 according to the embodiment will be described. FIG. 15 is a flowchart showing an example of learning the second weight α1 by the learning system 3 according to the embodiment. Note that steps S61 to S63 in FIG. 15 are the same as steps S51 to S53 (see FIG. 14) in the example of learning the first weight α2, and therefore a description thereof will be omitted here. The weight learning unit 36 ​​of the learning system 3 learns the second weight α1 by, for example, regression based on the acquired second demand quantity, second order quantity, and predicted second demand quantity (S64).

[0073] Next, an example of learning the third weight β by the learning system 3 according to the embodiment will be described. FIG. 16 is a flowchart showing an example of learning the third weight β by the learning system 3 according to the embodiment. Note that steps S71 to S73 in FIG. 16 are the same as steps S51 to S53 (see FIG. 14) in the example of learning the first weight α2, and therefore a description thereof will be omitted here. The weight learning unit 36 ​​of the learning system 3 learns the first weight α2 and the second weight α1, for example, by regression, based on the acquired second demand quantity, second order quantity, and predicted quantity of the second demand quantity (S74). Note that if the first weight α2 and the second weight α1 have already been learned, step S74 may not be executed.

[0074] Next, the seventh acquisition unit 34 of the learning system 3 acquires discontinued product information indicating whether the target product was a discontinued product at a time prior to a predetermined time (the current time) (S75). Note that step S75 may be executed before step S74. Next, the weight learning unit 36 ​​of the learning system 3 narrows down the target products to past discontinued products based on the acquired discontinued product information (S76). Then, the weight learning unit 36 ​​learns the third weight β based on the second demand quantity, second order quantity, predicted second demand quantity, first weight α2, and second weight α1 of the discontinued product among the target products (S77).

[0075] [4. Advantages] Advantages of the demand forecasting system 1 (demand forecasting method) according to the embodiment will be described below. As described above, the demand forecasting system 1 according to the embodiment calculates the second forecasted amount indicating the demand amount of the target item after the predetermined time point by using not only the first demand amount prior to the predetermined time point (the current time) for the target item and the first order amount for the target item at the predetermined time point, but also the first weight α2 representing the likelihood of the order amount. Therefore, the demand forecasting system 1 according to the embodiment calculates the second forecasted amount taking into account the uncertainty of the first order amount, which has the advantage of making it easier to improve the accuracy of predicting the demand amount of the target item after the predetermined time point, i.e., making it easier to improve the accuracy of predicting the demand for the item, compared to when the uncertainty of the first order amount is not taken into account.

[0076] FIG. 17 is an explanatory diagram of advantages of the demand forecasting system 1 according to the embodiment. FIG. 17 shows the forecast error of the demand quantity of the target item for each month from month N to month N+12, assuming that the current month is month N. Here, the forecast error is the difference between the actual demand quantity of the target item for the corresponding month and the forecast quantity (second forecast quantity) of the demand quantity of the target item for that month. Note that "AVG" in FIG. 17 represents the average value of the forecast error for each month from month N to month N+12.

[0077] In addition, in FIG. 17, the black graph represents the prediction error when prediction is made without using a weight, the white graph represents the prediction error when prediction is made using the first weight α2, and the hatched graph represents the prediction error when prediction is made using the first weight α2 and the second weight α1.

[0078] As shown in FIG. 17 , when prediction is made using the first weight α2 as in the demand forecasting system 1 according to the embodiment, the prediction error is smaller than when prediction is made without using a weight. In this way, the demand forecasting system 1 according to the embodiment improves the accuracy of predicting demand for goods compared to when prediction is made without using a weight. Also, as shown in FIG. 17 , when prediction is made using the second weight α1 in addition to the first weight α2, the prediction error is smaller than when prediction is made using only the first weight α2. Therefore, by making predictions using not only the first weight α2 but also the second weight α1, the accuracy of predicting demand for goods is further improved.

[0079] Furthermore, by further using the third weight β in the prediction, it is expected that the accuracy of predicting demand for goods will be further improved compared to when making a prediction using only the first weight α2 and the second weight α1.

[0080] (Other Embodiments) As described above, the embodiments have been described as examples of the technology according to the present disclosure. However, the technology according to the present disclosure is not limited to these, and can be applied to embodiments in which modifications, substitutions, additions, omissions, etc. are made as appropriate. For example, the following modifications are also included in one embodiment of the present disclosure.

[0081] In the above embodiment, the demand forecasting system 1 may output information indicating at least one of the first order quantity, the first forecast quantity, the first weight α2, the second weight α1, and the third weight β. For example, the demand forecasting system 1 may output the information by displaying the information on a display built into the information processing terminal including the demand forecasting system 1 or on an external display connected to the information processing terminal. In this case, the user can check the data used to forecast the demand quantity of the target item from a predetermined point in time (the present time) onward, which has the advantage of making it easier to verify the validity of the forecast.

[0082] In the above embodiment, the weight learning unit 36 ​​of the learning system 3 learns each weight using the second order quantity prior to the predetermined time point (the present time) and the predicted quantity of the second demand quantity based on the first demand quantity prior to the predetermined time point. However, the range of the first demand quantity and the second order quantity to be referenced may be narrowed. For example, both the first demand quantity and the second order quantity may be data from a predetermined period (e.g., several months) prior to the predetermined time point (the present time). In this case, each weight can be learned by referring to past data relatively close to the predetermined time point (the present time), which has the advantage of being able to expect improved prediction accuracy for the demand quantity of the target item after the predetermined time point.

[0083] In the above embodiment, the weight learning unit 36 ​​of the learning system 3 learns the first weight α2, the second weight α1, and the third weight β, but this is not limited to this. For example, if the demand forecasting system 1 executes only the first operation, the weight learning unit 36 ​​only needs to learn the first weight α2. Furthermore, for example, if the demand forecasting system 1 can also execute the second operation, the weight learning unit 36 ​​only needs to learn the first weight α2 and the second weight α1.

[0084] In the above embodiment, the second prediction unit 15 calculates the second predicted amount by multiplying a target variable (hereinafter referred to as the "target variable") by a weight, as shown in equations (1) to (3). For example, in each of equations (1) to (3), the first weight α2 is multiplied by the first order quantity, which is the target variable. Here, the second prediction unit 15 is not limited to multiplying the target variable by a weight, and may calculate the second predicted amount by performing any of the following operations, or may calculate the second predicted amount by combining one or more operations:

[0085] For example, the second prediction unit 15 may calculate the second predicted amount by adding a weight to the target variable. Alternatively, for example, the second prediction unit 15 may calculate the second predicted amount by dividing the target variable by the weight. Alternatively, for example, the second prediction unit 15 may calculate the second predicted amount by exponentiating the target variable with the weight as the base and the weight as the exponent. Alternatively, for example, the second prediction unit 15 may calculate the second predicted amount using a function with the weight and the target variable as variables. Alternatively, for example, the second prediction unit 15 may calculate the second predicted amount using a trained model that uses the weight and the target variable as inputs.

[0086] In the above embodiment, the demand forecasting system 1 may execute only the third operation. That is, the demand forecasting method may acquire a first demand quantity for the target item prior to a predetermined time point, acquire a first order quantity for the target item at the predetermined time point, and acquire a third weight β that represents a change in demand depending on whether the target item is a discontinued item whose supply will end. The demand forecasting method may also calculate a first forecast quantity that indicates the demand quantity for the target item after the predetermined time point, based on the first demand quantity. The demand forecasting method may then calculate a second forecast quantity that indicates the demand quantity for the target item after the predetermined time point, based on the first order quantity, the third weight β, and the first forecast quantity.

[0087] In the above embodiment, the learning system 3 may re-learn at least one of the trained model 141, the first weight α2, the second weight α1, and the third weight β during operation of the demand forecasting system 1. The timing of the re-learning may be, for example, periodically or at the time when the demand forecasting system 1 is used.

[0088] In the above embodiment, the learning system 3 includes both the learning unit 35 and the weight learning unit 36, but this is not limited to this. For example, the learning system 3 may include only the learning unit 35 or only the weight learning unit 36.

[0089] In the above embodiment, the learning system 3 may be incorporated into the demand forecasting system 1.

[0090] For example, the present disclosure can be realized as a program for causing a computer (processor) to execute steps included in a demand forecasting method. Furthermore, the present disclosure can be realized as a non-transitory computer-readable recording medium, such as a CD-ROM, on which the program is recorded. The same applies to steps included in a learning method.

[0091] For example, when the present disclosure is realized as a program (software), each step is performed by running the program using hardware resources such as a computer's CPU, memory, input / output circuitry, etc. In other words, each step is performed by the CPU acquiring data from memory or input / output circuitry, etc., performing calculations on the data, and outputting the calculation results to memory or input / output circuitry, etc.

[0092] In the above embodiment, the demand forecasting system is implemented by a single device, but the present invention is not limited to this and may be implemented by distributing the system across multiple devices. The same applies to the learning system.

[0093] Furthermore, in the above-described embodiments, each component included in the demand forecasting system may be configured with dedicated hardware, or may be realized by executing a software program suitable for each component. Each component may be realized by a program execution unit such as a CPU or processor reading and executing a software program recorded on a recording medium such as a hard disk or semiconductor memory. The same applies to each component included in the learning system.

[0094] Some or all of the functions of the demand forecasting system according to the above-described embodiments are typically realized as an LSI, which is an integrated circuit. These may be individually integrated into a single chip, or may be integrated into a single chip that includes some or all of the functions. Furthermore, the integrated circuit is not limited to an LSI, and may be realized using a dedicated circuit or a general-purpose processor. An FPGA (Field Programmable Gate Array), which can be programmed after the LSI is manufactured, or a reconfigurable processor, which can reconfigure the connections and settings of circuit cells within the LSI, may also be used. The same applies to the functions of the learning system.

[0095] Furthermore, if an integrated circuit technology that can replace LSIs emerges due to advances in semiconductor technology or other derivative technologies, that technology may naturally be used to integrate the components included in the demand forecasting system into integrated circuits. The same applies to the components included in the learning system.

[0096] In addition, this disclosure also includes forms obtained by making various modifications to the embodiments that a person skilled in the art would think of, and forms realized by arbitrarily combining the components and functions in each embodiment within the scope that does not deviate from the intent of this disclosure.

[0097] (Summary) As described above, the demand forecasting method according to the first aspect acquires a first demand amount for a target item prior to a predetermined time point (S11), acquires a first order amount for the target item at the predetermined time point (S15), and acquires a first weight α2 representing the likelihood of the first order amount (S16). The demand forecasting method calculates a first forecast amount indicating the demand amount for the target item after the predetermined time point based on the first demand amount (S13). The demand forecasting method calculates a second forecast amount indicating the demand amount for the target item after the predetermined time point based on the first order amount, the first weight α2, and the first forecast amount (S17).

[0098] This has the advantage that, since the second forecast quantity is calculated taking into account the uncertainty of the first order quantity, it is easier to improve the accuracy of predicting the demand quantity of the target item from a specified point in time onwards, i.e., it is easier to improve the accuracy of predicting the demand for the item, compared to when the uncertainty of the first order quantity is not taken into account.

[0099] In the demand forecasting method according to the second aspect, a second weight α1 representing the likelihood of the first forecasted amount calculated based on the first demand amount in the first aspect is further acquired (S26). In the demand forecasting method according to the first aspect, when calculating the second forecasted amount, the second forecasted amount is calculated based on the first order amount, the first weight α2, the first forecasted amount, and the second weight α1 (S27).

[0100] This has the advantage that the second forecast quantity is calculated taking into account the uncertainty of the first forecast quantity, making it easier to further improve the accuracy of predicting the demand quantity of the target item from a specified point in time onwards, compared to when the uncertainty of the first forecast quantity is not taken into account.

[0101] In the demand forecasting method according to the third aspect, the second aspect further acquires a third weight β that represents a change in demand depending on whether the target item is a discontinued item whose supply will end (S36). In the demand forecasting method, when calculating the second forecast amount, the second forecast amount is calculated based on the first order amount, the first weight α2, the first forecast amount, the second weight α1, and the third weight β (S37).

[0102] This has the advantage that the accuracy of predicting the demand for the target item after a specified point in time can be further improved compared to when demand depending on whether the target item is a discontinued item is not taken into account, since the second forecast quantity is calculated by further taking into account demand depending on whether the target item is a discontinued item or not.

[0103] In addition, in the demand forecasting method according to the fourth aspect, in the third aspect, information indicating at least one of the first order quantity, the first forecast quantity, the first weight α2, the second weight α1, and the third weight β is output.

[0104] This has the advantage that the user can check the data used to predict the demand for the target item from a predetermined point in time onward, making it easier to verify the validity of the prediction.

[0105] Furthermore, a demand forecasting system 1 according to a fifth aspect includes a first acquisition unit 11, a second acquisition unit 12, a third acquisition unit 13, a first prediction unit 14, and a second prediction unit 15. The first acquisition unit 11 acquires a first demand quantity of the target item prior to a predetermined time point. The second acquisition unit 12 acquires a first order quantity of the target item at the predetermined time point. The third acquisition unit 13 acquires a first weight α2 representing the likelihood of the first order quantity. The first prediction unit 14 calculates a first predicted quantity indicating the demand quantity of the target item after the predetermined time point, based on the first demand quantity. The second prediction unit 15 calculates a second predicted quantity indicating the demand quantity of the target item after the predetermined time point, based on the first order quantity, the first weight α2, and the first predicted quantity.

[0106] This has the advantage that, since the second forecast quantity is calculated taking into account the uncertainty of the first order quantity, it is easier to improve the accuracy of predicting the demand quantity of the target item from a specified point in time onwards, i.e., it is easier to improve the accuracy of predicting the demand for the item, compared to when the uncertainty of the first order quantity is not taken into account.

[0107] In addition, in a learning method according to a sixth aspect, a second demand amount of a target product at a predetermined time point is acquired (S53), a second order amount of the target product prior to the predetermined time point is acquired (S52), and a forecast amount of the second demand amount based on a first demand amount of the target product prior to the predetermined time point is acquired (S51). In the learning method, a first weight α2 representing the likelihood of the second order amount is learned based on the second demand amount, the second order amount, and the forecast amount (S54).

[0108] This has the advantage that the first weight α2 can be learned taking into account the order quantity of the target item prior to a specified time point, making it easier to obtain a first weight α2 that more accurately reflects the uncertainty of the order quantity of the target item at a specified time point.

[0109] In addition, in the learning method according to the seventh aspect, in the sixth aspect, a second weight α1 representing the likelihood of a prediction based on the first demand quantity is learned based on the second demand quantity, the second order quantity, and the forecast quantity (S64).

[0110] This has the advantage that the second weight α1 can be learned by taking into account the order quantity of the target item prior to a specified time point, making it easier to obtain a second weight α1 that more accurately reflects the uncertainty of the forecast based on the demand quantity of the target item prior to a specified time point.

[0111] In addition, in the learning method according to the eighth aspect, in the seventh aspect, a third weight β representing a change in demand depending on whether the target item is a discontinued item or not is learned based on the second demand quantity, second order quantity, forecast quantity, first weight α2, and second weight α1 of the discontinued item among the target items whose supply has ended (S77).

[0112] This has the advantage that the third weight β can be learned taking into account demand depending on whether the target item is a discontinued item or not, making it easier to obtain a third weight β that more accurately reflects demand depending on whether the target item is a discontinued item or not.

[0113] In addition, in the learning method according to the ninth aspect, in any one of the sixth to eighth aspects, the first demand quantity and the second order quantity are both data for a range of a specified period prior to a specified point in time.

[0114] This has the advantage that each weight can be learned by referring to past data relatively close to the specified point in time, which is expected to improve the accuracy of predicting the demand for the target item from the specified point in time onwards.

[0115] Furthermore, the learning system 3 according to the tenth aspect includes a fourth acquisition unit 31, a fifth acquisition unit 32, a sixth acquisition unit 33, and a weight learning unit 36. The fourth acquisition unit 31 acquires a second demand quantity of the target item at a predetermined time. The fifth acquisition unit 32 acquires a second order quantity of the target item prior to the predetermined time. The sixth acquisition unit 33 acquires a forecast quantity of the second demand quantity based on the first demand quantity of the target item prior to the predetermined time. The weight learning unit 36 ​​learns a first weight α2 representing the likelihood of the second order quantity based on the second demand quantity, the second order quantity, and the forecast quantity.

[0116] This has the advantage that the first weight α2 can be learned taking into account the order quantity of the target item prior to a specified time point, making it easier to obtain a first weight α2 that more accurately reflects the uncertainty of the order quantity of the target item at a specified time point.

[0117] The present disclosure can be applied to systems for predicting demand for goods, etc.

[0118] REFERENCE SIGNS LIST 1 Demand forecasting system 11 First acquisition unit 12 Second acquisition unit 13 Third acquisition unit 14 First prediction unit 141 Trained model 15 Second prediction unit 2 Database 3 Learning system 31 Fourth acquisition unit 32 Fifth acquisition unit 33 Sixth acquisition unit 34 Seventh acquisition unit 35 Learning unit 36 ​​Weight learning unit α1 Second weight α2 First weight β Third weight

Claims

1. A demand forecasting method comprising: acquiring a first demand quantity for a target item prior to a predetermined time; acquiring a first order quantity for the target item at the predetermined time; acquiring a first weight representing the likelihood of the first order quantity; calculating a first forecast quantity indicating the demand quantity for the target item after the predetermined time based on the first demand quantity; and calculating a second forecast quantity indicating the demand quantity for the target item after the predetermined time based on the first order quantity, the first weight, and the first forecast quantity.

2. The demand forecasting method according to claim 1, further comprising: obtaining a second weight representing the likelihood of the first forecasted quantity calculated based on the first demand quantity; and, when calculating the second forecasted quantity, calculating the second forecasted quantity based on the first order quantity, the first weight, the first forecasted quantity, and the second weight.

3. The demand forecasting method according to claim 2, further comprising: obtaining a third weight representing a change in demand depending on whether the target item is a discontinued item whose supply will end; and, when calculating the second forecasted quantity, calculating the second forecasted quantity based on the first order quantity, the first weight, the first forecasted quantity, the second forecasted quantity, and the third weight.

4. The demand forecasting method according to claim 3, further comprising outputting information indicating at least one of the first order quantity, the first forecast quantity, the first weight, the second weight, and the third weight.

5. A demand forecasting system comprising: a first acquisition unit that acquires a first demand quantity for a target item prior to a predetermined time point; a second acquisition unit that acquires a first order quantity for the target item at the predetermined time point; a third acquisition unit that acquires a first weight that represents the likelihood of the first order quantity; a first prediction unit that calculates a first predicted quantity that indicates the demand quantity for the target item after the predetermined time point based on the first demand quantity; and a second prediction unit that calculates a second predicted quantity that indicates the demand quantity for the target item after the predetermined time point based on the first order quantity, the first weight, and the first predicted quantity.

6. A learning method comprising: acquiring a second demand quantity for a target item at a predetermined time; acquiring a second order quantity for the target item prior to the predetermined time; acquiring a predicted quantity of the second demand quantity based on a first demand quantity for the target item prior to the predetermined time; and learning a first weight representing the likelihood of the second order quantity based on the second demand quantity, the second order quantity, and the predicted quantity.

7. The learning method according to claim 6, further comprising: learning a second weight representing the likelihood of a prediction based on the first demand quantity based on the second demand quantity, the second order quantity, and the prediction quantity.

8. The learning method described in claim 7, further comprising: learning a third weight representing a change in demand depending on whether the target item is a discontinued item or not, based on the second demand quantity, the second order quantity, the forecast quantity, the first weight, and the second weight of a discontinued item among the target items whose supply has ended.

9. A learning method according to any one of claims 6 to 8, wherein the first demand quantity and the second order quantity are both data for a range of a predetermined period prior to the predetermined time point.

10. A learning system comprising: a fourth acquisition unit that acquires a second demand quantity of a target item at a predetermined time; a fifth acquisition unit that acquires a second order quantity of the target item prior to the predetermined time; a sixth acquisition unit that acquires a predicted quantity of the second demand quantity based on a first demand quantity of the target item prior to the predetermined time; and a weight learning unit that learns a first weight that represents the likelihood of the second order quantity based on the second demand quantity, the second order quantity, and the predicted quantity.

Images