Demand prediction device, demand prediction system, demand prediction model generation method, and program

Abstract

To allow demand prediction which more easily follows a demand fluctuation.SOLUTION: A demand prediction device 1 for predicting the demand for a product comprises a customer information generation unit 11 which acquires and aggregates company information of an end customer from a supply chain network for the product that includes sales channels for 0-th products to N-th products and generates customer information indicating an aggregation result, a statistical information extraction unit 12 which extracts statistical information about the end customer on the basis of the customer information, a feature amount generation unit 13 which generates a feature amount about the end customer on the basis of the statistical information about the end customer, a demand prediction model generation unit 14 which generates a demand prediction model by learning a demand fluctuation with the feature amount as an explanatory variable and the demand of the product as an objective variable, and a demand prediction model application unit 15 which predicts the demand for the product by applying the demand prediction model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a demand forecasting device, a demand forecasting system, a demand forecasting model generation method, and a program. [Background technology]

[0002] Product demand forecasting requires predicting future demand as accurately as possible. To achieve this, it is common to use not only past performance and seasonality of product demand, but also external indicators related to the product for which demand is being predicted. Selecting these external indicators requires deep product knowledge. In particular, in cases where there are a wide range of uses, such as processing machines and semiconductors, human product knowledge alone cannot cover everything, making the selection of external indicators extremely difficult.

[0003] Patent Document 1 discloses a method for calculating a product demand plan in a supply chain of sales, manufacturing, and supply, using product demand forecast information on the sales side and parts delivery schedule information on the supply side. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2022-32437 Summary of the Invention [Problem to be solved by the invention]

[0005] The technology described in Patent Document 1 uses a supply chain network, but the downstream components of that supply chain network are only sales channels that sell the company's own products. In addition, while the demand forecast takes into account the sales performance of retail stores to end customers, this is a prediction of future sales figures based on past trends, and is not information that is directly linked to demand fluctuations, so it is difficult to make the demand forecast follow demand fluctuations.

[0006] The present disclosure has been made to solve the above-mentioned problems, and aims to enable demand forecasting that is more responsive to demand fluctuations. [Means for solving the problem]

[0007] In order to achieve the above object, a demand forecasting device according to the present disclosure is a demand forecasting device that forecasts product demand. The demand forecasting device includes a customer information generation unit, a statistical information extraction unit, a feature generation unit, a demand forecasting model generation unit, and a demand forecasting model application unit. The customer information generation unit acquires and aggregates company information related to end customer companies from a product supply chain network including sales channels for zero-order products to nth-order products, and generates customer information indicating the aggregation results. The statistical information extraction unit extracts statistical information related to the end customer based on the customer information. The feature generation unit generates feature information related to the end customer based on the statistical information related to the end customer. The demand forecasting model generation unit uses the feature information as explanatory variables and product demand as a target variable, and generates a demand forecasting model by learning demand fluctuations. The demand forecasting model application unit applies the demand forecasting model to forecast product demand. [Effects of the Invention]

[0008] According to the present disclosure, by generating a demand forecasting model using a supply chain network including sales channels for 0th to Nth order products, it becomes possible to make demand forecasts that are more responsive to demand fluctuations. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram illustrating an example of a functional configuration of a demand prediction device according to an embodiment. [Figure 2] FIG. 10 is a diagram illustrating an example of a functional configuration of a customer information generation unit according to an embodiment. [Figure 3] FIG. 1 is a diagram illustrating an example of a supply chain network according to an embodiment. [Figure 4] FIG. 10 is a diagram showing an example of customer information according to an embodiment. [Figure 5] FIG. 10 is a diagram illustrating an example of a functional configuration of a statistical information extraction unit according to an embodiment. [Figure 6] FIG. 10 is a diagram showing an example of a statistical information master according to an embodiment; [Figure 7] FIG. 10 is a diagram showing an example of time-series data for different time resolutions according to an embodiment; [Figure 8] FIG. 1 is a diagram illustrating an example of a functional configuration of a feature generating unit according to an embodiment. [Figure 9] FIG. 10 is a diagram illustrating an example of a functional configuration of a demand forecasting model generation unit according to an embodiment. [Figure 10] FIG. 1 is a diagram showing an example of a neural network according to an embodiment; [Figure 11] FIG. 10 is a diagram illustrating an example of a functional configuration of a demand forecasting model application unit according to an embodiment. [Figure 12] 1 is a flowchart showing a demand forecasting model generation process according to an embodiment. [Figure 13] Flowchart showing demand forecasting processing according to an embodiment [Figure 14] FIG. 1 is a diagram illustrating a configuration example of a demand forecasting system according to an embodiment. [Figure 15] FIG. 1 is a diagram illustrating an example of a hardware configuration of a demand prediction device according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] A demand forecasting device, a demand forecasting system, a demand forecasting model generating method, and a program according to the present embodiment will be described in detail below with reference to the drawings. Note that identical or corresponding parts in the drawings are designated by the same reference numerals.

[0011] First, a description will be given of the supply chain network according to this embodiment. In a typical supply chain network, the only downstream flow is the sales channels (sales agents, retail stores, sales offices, etc.) through which a company's products are sold. In contrast, the supply chain network according to this embodiment first connects from the company's own node to the company that sells the product sold by the company (hereinafter referred to as the zero-order product). Next, if the sales company uses the zero-order product to manufacture another product (hereinafter referred to as the primary product), it connects to the company that sells the primary product. Also, if the sales company sells the zero-order product, it connects to the company that sells it. This process is repeated until the product reaches the end consumer, forming a supply chain network. In other words, the supply chain network according to this embodiment is a product supply chain network that includes the flow of the zero-order product from the production of the primary product to the final N-order product, and the types of companies through which the zero-order product passes before reaching the end consumer, that is, the sales channels for the zero-order product to the N-order product. Furthermore, the supply chain network according to this embodiment is a graph expressed as computer-handleable data, with nodes representing companies involved in the manufacture and supply of products and edges representing relationships between companies (supply / demand of raw materials and intermediate products, provision / use of services). This supply chain network cannot be generated using only information about a company's sales destinations, but can be generated by asking sales destinations of zero-order products and primary products to sales destination companies of zero-order products, and then asking companies that handle primary products to ask sales destinations of secondary products. Hereinafter, the supply chain network will be abbreviated as SC network.

[0012] The configuration of a demand forecasting device 1 according to an embodiment will be described with reference to Fig. 1. As shown in Fig. 1, the demand forecasting device 1 includes a storage unit 10 that stores various types of information, a customer information generation unit 11 that generates customer information by aggregating company information on end-customer companies included in the SC network, a statistical information extraction unit 12 that extracts statistical information on end customers, a feature generation unit 13 that generates feature amounts on end customers, a demand forecasting model generation unit 14 that uses the feature amounts as explanatory variables and product demand as a target variable, learns demand fluctuations, and generates a demand forecasting model, and a demand forecasting model application unit 15 that applies the demand forecasting model to predict product demand.

[0013] The storage unit 10 includes an SC network storage unit 101 that stores the nodes and edges of the SC network, a customer information storage unit 102 that stores customer information indicating the aggregation result of corporate information of end customers included in the SC network, a statistical information storage unit 103 that stores corporate management information on corporate management and industry economic information on industry economic trends as statistical information, an intermediate information storage unit 104 that stores the corporate management information and industry economic information on end customers as intermediate information, a dependent variable storage unit 105 that stores time-series data of dependent variables, a feature storage unit 106 that stores explanatory variables used for demand forecasting, and a demand forecasting model storage unit 107 that stores a demand forecasting model generated as a result of learning demand fluctuations. The time-series data of dependent variables stored in the dependent variable storage unit 105 is time-series data that represents actual demand for products.

[0014] The customer information generation unit 11 acquires and aggregates company information of downstream end customers in the SC network to generate customer information. An example of the functional configuration of the customer information generation unit 11 will be described with reference to Fig. 2. The customer information generation unit 11 includes an SC network acquisition unit 111 that acquires nodes and edges of the SC network, an end customer node acquisition unit 112 that extracts end customer nodes, and an aggregation unit 113 that aggregates company information held by the end customer nodes.

[0015] The SC network acquisition unit 111 acquires the nodes and edges of the SC network from the SC network storage unit 101. An example of an SC network is shown in Figure 3. The SC network has a company's own node at the center, with the supplier side connected to the company's own node as upstream and the customer side branching off from the company's own node as downstream. The node one node away from the company's own central node is Tier 1, the node two nodes away is Tier 2, and the node M nodes away from the company's own node will hereinafter be referred to as a Tier M node.

[0016] Each node in the SC network is not a base unit such as a company's warehouse or sales outlet, but a company unit, and holds company information such as company name, address, and industry code. Downstream Tier 1 nodes indicate companies to which the company's products are sold. Downstream Tier 2 and above nodes indicate companies that purchase the company's products (0th-order products) from the company at the Tier 1 node, or companies that purchase products (1st-order products) manufactured using the company's products (0th-order products) purchased by the company at the Tier 1 node. Nodes that are not connected to the next downstream node in the SC network are end customers. An SC network such as the one shown in Figure 3 is stored in the SC network storage unit 101, and the SC network stored in the SC network storage unit 101 is updated periodically or when there is a change.

[0017] Returning to Figure 2, the end customer node acquisition unit 112 extracts, as end customers, nodes that are not connected to the next node downstream of the SC network acquired by the SC network acquisition unit 111. The aggregation unit 113 aggregates the company information of the end customer nodes extracted by the end customer node acquisition unit 112. The aggregation of company information may involve, for example, creating a list of company names included in the company information of the end customer nodes and aggregating industry codes. The aggregation unit 113 generates customer information by aggregating the company information and stores it in the customer information storage unit 102. Figure 4 shows an example of customer information. In the example of Figure 4, the customer information includes a list of company names and the aggregation results of industry codes, and the aggregation results of industry codes are expressed as the number of companies for each industry code.

[0018] Returning to Fig. 1, the statistical information extraction unit 12 extracts statistical information related to end customers. An example of the functional configuration of the statistical information extraction unit 12 will be described with reference to Fig. 5. The statistical information extraction unit 12 includes a customer information acquisition unit 121 that acquires customer information, a business management information acquisition unit 122 that acquires business management information related to end customers based on the customer information, and an industry economic information acquisition unit 123 that acquires industry economic information related to end customers based on the customer information.

[0019] The customer information acquisition unit 121 acquires customer information stored in the customer information storage unit 102. The corporate management information acquisition unit 122 acquires corporate management information from the statistical information storage unit 103 using as a key the company name included in the customer information acquired by the customer information acquisition unit 121. The corporate management information is time-series data in which the values ​​of indices that represent the value and management status of a company, such as stock prices and financial information, are linked to a time axis.

[0020] The industry economic information acquisition unit 123 acquires industry economic information from the statistical information storage unit 103 using the industry code included in the customer information acquired by the customer information acquisition unit 121 as a key. The industry economic information is time-series data in which the values ​​of indicators representing the competitiveness, growth rate, and market trends of an industry are linked to a time axis. For example, in the case of the semiconductor industry, the industry economic information is time-series data in which the values ​​of indicators such as the number of semiconductors shipped domestically and the number of semiconductor materials imported are linked to a time axis.

[0021] The statistical information storage unit 103 stores, for example, a statistical information master including corporate management information and industry economic information as shown in Fig. 6 and time-series data by time resolution as shown in Fig. 7. In the example of Fig. 6, the statistical information master has the following items: "Information ID" which is a number identifying the corporate management information and industry economic information; "Information Name" which is the name of the corporate management information and industry economic information; "Company Name" which is the name of the company in the corporate management information; "Industry Code" which identifies the industry of the industry economic information; "Time Resolution" which indicates the time resolution of the corporate management information and industry economic information; "Country" which indicates the country in which the corporate management information and industry economic information are from; and "Update Date" which indicates the date the information was updated. In the example of Fig. 7, the time-series data by time resolution is time-series data in which the time axis and index value for each "Information ID" stored in a table by time resolution are linked.

[0022] When acquiring corporate management information from the statistical information storage unit 103, the corporate management information acquisition unit 122 identifies corporate management information whose company name in the statistical information master matches the company name included in the customer information acquired by the customer information acquisition unit 121, and acquires it together with time series data whose "information ID" matches. When acquiring industry economic information from the statistical information storage unit 103, the industry economic information acquisition unit 123 identifies industry economic information whose industry code in the statistical information master matches the industry code included in the customer information acquired by the customer information acquisition unit 121, and acquires it together with time series data whose "information ID" matches.

[0023] Returning to FIG. 5, the intermediate information storage unit 124 stores the business management information acquired by the business management information acquisition unit 122 and the industry economic information acquired by the industry economic information acquisition unit 123 in the intermediate information storage unit 104 as intermediate information.

[0024] Returning to Fig. 1, the feature generation unit 13 generates features related to end customers. An example of the functional configuration of the feature generation unit 13 will be described with reference to Fig. 8. The feature generation unit 13 includes an intermediate information acquisition unit 131 that acquires intermediate information, a dependent variable acquisition unit 132 that acquires time series data of a dependent variable, a time resolution unifying unit 133 that unifies the time series data of the intermediate information to the time resolution of the dependent variable, a time shift unit 134 that time-shifts the time series data of the intermediate information unified to the time resolution of the dependent variable to match the time series of the dependent variable, and a statistical information prediction unit 135 that predicts and compensates for missing corporate management information and industry economic information using autoregression.

[0025] The intermediate information acquisition unit 131 acquires intermediate information from the intermediate information storage unit 104. The objective variable acquisition unit 132 acquires time-series data of objective variables such as the number of units shipped, the number of units ordered, and sales, which represent product demand, from the objective variable storage unit 105.

[0026] The time resolution unifying unit 133 unifies the time series data of the corporate management information and the industry economic information, which are the intermediate information acquired by the intermediate information acquiring unit 131, to the time resolution of the objective variable. This is because although both the corporate management information and the industry economic information are information linked to the time axis, their time resolutions are not basically unified.

[0027] Specifically, for corporate management information and industry economic information with a higher time resolution than the objective variable, for example, when the objective variable is monthly, the daily corporate management information and industry economic information are unified into monthly. To unify daily data into monthly data, for example, the sum of the daily corporate management information and industry economic information from January 1, 2023 to January 31, 2023 is calculated and the resulting value is defined as the value for January 2023. It should be noted that statistics such as the average, maximum, and minimum values ​​for one month can also be used, rather than just the sum. Furthermore, for corporate management information and industry economic information with a lower time resolution than the objective variable, for example, when the objective variable is monthly, the annual corporate management information and industry economic information are unified into monthly data. To unify annual data into monthly data, for example, the values ​​for the annual corporate management information and industry economic information for 2023 are used as values ​​from January 2023 to December 2023. Alternatively, the annual value can be divided by 12 to obtain a monthly value. For the corporate management information and industry economic information whose time resolution is equal to that of the objective variable, the process of unifying the time resolution is not required. The time resolution unifying unit 133 stores the corporate management information and industry economic information whose time resolution is unified to that of the objective variable as first features in the feature storage unit 106.

[0028] The time shift unit 134 time-shifts the first feature quantity to match the time series of the dependent variable. For example, when forecasting the demand for the company's products for April 2023, corporate management information and industry economic information for April 2023 have not yet been released and cannot be used. Therefore, corporate management information and industry economic information from L months ago must be used to forecast the demand for the company's products. In this case, the first feature quantity is time-shifted by L months. For example, if L = 3 months, the first feature quantity for January 2023 is time-shifted by 3 months to forecast the demand for the company's products for April 2023. The method for determining L is, for example, determined by intuition and experience by someone with deep product knowledge. Alternatively, the correlation coefficient and distance between the time series data of the dependent variable and the time series data of the first feature quantity may be calculated by time-shifting by K months, and the number K with the strongest correlation and shortest distance between the time series data of the dependent variable and the time series data of the first feature quantity may be set as the number of months for the time shift L. The maximum value of K may be, for example, predetermined or set by the user via the time shift unit 134. It is also possible to select not to perform time shifting (L=0). The time shifting unit 134 stores the time-shifted corporate management information and industry economic information as second features in the feature storage unit 106. When time shifting is not performed, the first feature and the second feature are the same.

[0029] The statistical information prediction unit 135 performs a process of predicting corporate management information and industry economic information using autoregression to compensate for insufficient corporate management information and industry economic information when predicting a dependent variable for L months or more. For example, when the time shift unit 134 sets L = 1 month, the dependent variable can only be predicted for one month. Therefore, when predicting the dependent variable for J months, corporate management information and industry economic information corresponding to JL months are predicted using autoregression. Autoregression may use ARIMA (Autoregressive Integrated Moving Average) or SARIMA (Seasonal Autoregressive Integrated Moving Average). J may be predetermined or user-configurable. The statistical information prediction unit 135 stores the corporate management information and industry economic information, which compensate for the insufficient second feature through autoregressive prediction, as the third feature in the feature storage unit 106. When J = L, the second feature and the third feature are the same. The first feature, the second feature, and the third feature are examples of features related to end customers.

[0030] Returning to Fig. 1, the demand forecasting model generation unit 14 learns demand fluctuations based on the time series data of the third feature amount and the time series data of the dependent variable, and generates a demand forecasting model. An example of the functional configuration of the demand forecasting model generation unit 14 will be described with reference to Fig. 9. The demand forecasting model generation unit 14 includes a dependent variable acquisition unit 141 that acquires time series data of the dependent variable, a learning feature acquisition unit 142 that acquires time series data of the third feature amount, a feature selection unit 143 that selects explanatory variables to be used for learning demand fluctuations from the third feature amount, and a learning unit 144 that learns demand fluctuations for products using the time series data of the explanatory variables and the time series data of the dependent variable.

[0031] The objective variable acquisition unit 141 acquires time-series data of objective variables such as the number of units shipped, the number of units ordered, and sales, which represent product demand, from the objective variable storage unit 105. The learning feature acquisition unit 142 acquires time-series data of third features from the feature storage unit 106. The feature selection unit 143 selects explanatory variables to be used for learning of demand fluctuations by the learning unit 144 from the time-series data of the third features acquired by the learning feature acquisition unit 142. If there are too many explanatory variables, demand forecasting may not work correctly due to over-learning, constraints caused by the curse of dimensionality, etc., so they are narrowed down here.

[0032] For example, explanatory variables may be selected by calculating the Variance Inflation Factor (VIF), which indicates the multicollinearity of time-series data, selecting and excluding the explanatory variable with the highest VIF, and then recalculating the VIF, until the maximum VIF reaches a certain value. Alternatively, demand may be forecast using all explanatory variables once using techniques such as random forest or gradient boosting, and the number of explanatory variables may be narrowed down to a predetermined number in descending order of Feature Importance in the forecast. Alternatively, the number of companies, which is the aggregated result of industry codes included in customer information, may be used to narrow down the explanatory variables to be used. Alternatively, instead of having the feature selection unit 143 calculate the explanatory variables, the user may narrow down the explanatory variables to be used via the feature selection unit 143 based on product knowledge.

[0033] The learning unit 144 uses the time-series data of the explanatory variables selected by the feature selection unit 143 and the time-series data of the objective variable acquired by the objective variable acquisition unit 141 to learn, for example, by so-called supervised learning according to a neural network model, in which the explanatory variables are input and the objective variable is output, i.e., to learn product demand fluctuations. Here, supervised learning refers to a technique in which pairs of input (numerical values) and output (numerical values) data are provided to the learning unit 144, and the learning unit learns the features of the learning data and infers the output from the input. The neural network is composed of an input layer having multiple neurons, an intermediate layer (hidden layer) having multiple neurons, and an output layer having one neuron. The intermediate layer may be one layer or two or more layers.

[0034] For example, in a three-layer neural network as shown in FIG. 10, when the numerical values ​​of each explanatory variable at a certain time are input to the input layer (X1-X3), the value is multiplied by a weight W1 (w11-w16) and input to the middle layer (Y1-Y2), and the result is further multiplied by a weight W2 (w21, w22) and output from the output layer (Z1). This result varies depending on the values ​​of weights W1 and W2. In this embodiment, a neural network is used to learn product demand fluctuations through so-called supervised learning, using combinations of explanatory variables and response variables as learning data. That is, the neural network learns by inputting explanatory variables to the input layer and adjusting weights W1 and W2 so that the value output from the output layer approaches the response variable.

[0035] Although the processing of the learning unit 144 has been described using a neural network as an example, other deep learning methods can also be used. For example, learning may be performed using multiple regression prediction, random forest, gradient boosting, support vector machine, etc. The learning unit 144 performs the above-mentioned learning of demand fluctuations to generate a demand forecasting model. The learning unit 144 stores the generated demand forecasting model in the demand forecasting model storage unit 107.

[0036] Returning to Fig. 1, the demand forecasting model application unit 15 predicts the dependent variable by applying the demand forecasting model generated by the demand forecasting model generation unit 14. An example of the functional configuration of the demand forecasting model application unit 15 will be described with reference to Fig. 11. The demand forecasting model application unit 15 includes a forecast feature acquisition unit 151 that acquires feature values ​​to be used for demand forecasting, a demand forecasting model acquisition unit 152 that acquires a demand forecasting model, a demand forecasting unit 153 that performs demand forecasting using the demand forecasting model, and a forecast result output unit 154 that outputs the results of the demand forecasting.

[0037] The prediction feature acquisition unit 151 acquires third feature values ​​to be used for demand forecasting from the feature storage unit 106, that is, third feature values ​​narrowed down by year and month for predicting future dependent variables that have not been learned by the learning unit 144 of the demand forecast model generation unit 14. For example, if the time resolution of the dependent variable is monthly, learning is performed up to February 2023, and predictions are made for March 2023 and after, the prediction feature acquisition unit 151 acquires third feature values ​​for March 2023 and after.

[0038] The demand forecasting model acquisition unit 152 acquires the demand forecasting model stored in the demand forecasting model storage unit 107. The demand forecasting unit 153 inputs the third feature acquired by the prediction feature acquisition unit 151 as an explanatory variable into the demand forecasting model acquired by the demand forecasting model acquisition unit 152, and obtains an output of a dependent variable. The output dependent variable is time-series data indicating the result of the demand forecast for the product.

[0039] The prediction result output unit 154 outputs time-series data indicating the result of the demand prediction for the product generated by the demand prediction unit 153. The method of outputting the result of the demand prediction may be, for example, to display the result on the screen as a file in a table format, a graph format, or the like, or to send the result to a user terminal used by the user.

[0040] Here, the flow of the demand forecasting model generation process executed by the demand forecasting device 1 will be explained using Fig. 12. The demand forecasting model generation process shown in Fig. 12 starts, for example, when a demand forecasting model generation instruction is input to the demand forecasting device 1. The SC network acquisition unit 111 of the customer information generation unit 11 of the demand forecasting device 1 acquires the nodes and edges of the SC network from the SC network storage unit 101 (step S11). Fig. 3 shows an example of an SC network. The SC network has a company's own node at the center, with the supplier side connected to the company's own node as the upstream side and the customer side branching out from the company's own node as the downstream side.

[0041] 12, the end customer node acquisition unit 112 extracts end customers that are nodes not connected to the next node downstream of the SC network acquired by the SC network acquisition unit 111 (step S12). The aggregation unit 113 generates customer information by aggregating the company information of the end customer nodes extracted by the end customer node acquisition unit 112 (step S13), and stores the information in the customer information storage unit 102. In the example of customer information shown in FIG. 4, the customer information includes a list of company names and the aggregation results of industry codes.

[0042] Returning to FIG. 12 , the customer information acquisition unit 121 acquires customer information stored in the customer information storage unit 102 (step S14). The corporate management information acquisition unit 122 acquires corporate management information from the statistical information storage unit 103 using the company name included in the customer information acquired by the customer information acquisition unit 121 as a key (step S15). The corporate management information is, for example, time-series data in which the values ​​of indicators representing the value and management status of a company are linked to a time axis. The industry economic information acquisition unit 123 acquires industry economic information from the statistical information storage unit 103 using the industry code included in the customer information acquired by the customer information acquisition unit 121 as a key (step S16). The industry economic information is, for example, time-series data in which the values ​​of indicators representing the competitiveness, growth rate, and market trends of an industry are linked to a time axis. The intermediate information storage unit 124 stores the corporate management information acquired by the corporate management information acquisition unit 122 and the industry economic information acquired by the industry economic information acquisition unit 123 as intermediate information in the intermediate information storage unit 104.

[0043] 6, the statistical information storage unit 103 stores a statistical information master including corporate management information and industry economic information, and when acquiring corporate management information from the statistical information storage unit 103, the corporate management information acquisition unit 122 identifies and acquires corporate management information that matches the company name included in the customer information acquired by the customer information acquisition unit 121. When acquiring industry economic information from the statistical information storage unit 103, the industry economic information acquisition unit 123 identifies and acquires industry economic information whose industry code in the statistical information master matches the industry code included in the customer information acquired by the customer information acquisition unit 121.

[0044] 12, the intermediate information acquisition unit 131 of the feature generation unit 13 acquires intermediate information from the intermediate information storage unit 104 (step S17). The objective variable acquisition unit 132 acquires time-series data of objective variables such as the number of units shipped, the number of units ordered, and sales, which represent product demand, from the objective variable storage unit 105 (step S18). The time resolution unifying unit 133 unifies the time-series data of the corporate management information and the industry economic information acquired by the intermediate information acquisition unit 131 to the time resolution of the objective variables (step S19). The time resolution unifying unit 133 stores the corporate management information and the industry economic information unified to the time resolution of the objective variables in the feature storage unit 106 as first features.

[0045] The time shift unit 134 time-shifts the first feature by L months (step S20). The time shift unit 134 stores the time-shifted corporate management information and industry economic information as second feature in the feature storage unit 106. The statistical information prediction unit 135 predicts the corporate management information and industry economic information by autoregression to compensate for the lack of corporate management information and industry economic information when predicting the dependent variable for L months or more (step S21). The statistical information prediction unit 135 stores the corporate management information and industry economic information, which have been compensated for by autoregressive prediction to compensate for the lack of the second feature, as third feature in the feature storage unit 106.

[0046] 12, the objective variable acquisition unit 141 of the demand forecasting model generation unit 14 acquires time-series data of objective variables such as the number of units shipped, the number of units ordered, and sales, which represent demand for the product, from the objective variable storage unit 105 (step S22). The learning feature acquisition unit 142 acquires time-series data of the third feature from the feature storage unit 106 (step S23). The feature selection unit 143 selects explanatory variables to be used in the learning process of the learning unit 144 from the time-series data of the third feature acquired by the learning feature acquisition unit 142 (step S24).

[0047] The learning unit 144 learns the demand fluctuation of the product using the explanatory variables selected by the feature selection unit 143 and the objective variables acquired by the objective variable acquisition unit 141, and generates a demand forecasting model (step S25). The learning unit 144 stores the generated demand forecasting model in the demand forecasting model storage unit 107, and ends the process.

[0048] For example, in the case of learning in which explanatory variables are input and a target variable is output using so-called supervised learning according to a three-layer neural network such as the one shown in Figure 10, when the value of each explanatory variable at a certain point in time is input to the input layer (X1-X3), that value is multiplied by a weight W1 (w11-w16) and input to the middle layer (Y1-Y2), and the result is further multiplied by a weight W2 (w21, w22) and output from the output layer (Z1). The neural network learns fluctuations in product demand by inputting explanatory variables to the input layer and adjusting the weights W1 and W2 so that the value output from the output layer approaches the target variable.

[0049] Next, the flow of the demand forecasting model generation process executed by the demand forecasting device 1 will be described with reference to FIG. 13. The demand forecasting model generation process shown in FIG. 13 starts, for example, when a demand forecast instruction is input to the demand forecasting device 1. The forecast feature acquisition unit 151 of the demand forecasting model application unit 15 acquires third feature values ​​to be used for demand forecasting from the feature storage unit 106, that is, third feature values ​​by narrowing down the search by year and month for predicting a future dependent variable that has not been learned by the learning unit 144 of the demand forecasting model generation unit 14 (step S31). For example, if the time resolution of the dependent variable is monthly, learning is performed up to February 2023, and predictions are made for March 2023 and after, the forecast feature acquisition unit 151 acquires third feature values ​​for March 2023 and after.

[0050] The demand forecasting model acquisition unit 152 acquires the demand forecasting model stored in the demand forecasting model storage unit 107 (step S32). The demand forecasting unit 153 inputs the third feature acquired by the prediction feature acquisition unit 151 as an explanatory variable into the demand forecasting model acquired by the demand forecasting model acquisition unit 152, and obtains an output of a dependent variable (step S33). The output dependent variable is time-series data indicating the result of the demand forecast for the product. The forecast result output unit 154 outputs time-series data indicating the result of the demand forecast for the product generated by the demand forecasting unit 153 (step S34), and ends the processing.

[0051] A demand forecasting system 100 using the demand forecasting device 1 will now be described with reference to FIG. 14. As shown in FIG. 14, the demand forecasting system 100 includes the demand forecasting device 1, an SC network generation system 2 that generates an SC network, and a sales management system 3 that manages product sales. As described above, the SC network generation system 2 generates an SC network by confirming with the sales destinations of the zero-order products and the primary products from the sales destination companies of the zero-order products and by confirming with the sales destination companies of the primary products from the sales destination companies of the secondary products. In the SC network generation system 2, a person in charge collects publicly available corporate management information and industry economic information, for example, via the Internet, and stores them in a data server. Based on the corporate management information and industry economic information stored in the data server, a computer generates a statistical information master as shown in FIG. 6 and time-series data by time resolution as shown in FIG. 7. The SC network storage unit 101 of the storage unit 10 of the demand forecasting device 1 stores the SC network generated by the SC network generation system 2, and the statistical information storage unit 103 stores the statistical information master and time-series data by time resolution generated by the SC network generation system 2. The sales management system 3 generates time-series data of the objective variable based on sales performance such as the number of units shipped, the number of units ordered, and sales, which represent the demand for the product. The objective variable storage unit 105 stores the time-series data of the objective variable generated by the sales management system 3.

[0052] According to the demand forecasting device 1 of the embodiment, by generating a demand forecasting model using a supply chain network including sales channels for 0th to Nth order products, it becomes possible to perform demand forecasting that is more responsive to demand fluctuations.

[0053] The hardware configuration of the demand prediction device 1 will be described with reference to Fig. 15. As shown in Fig. 15, the demand prediction device 1 includes a temporary storage unit 401, a storage unit 402, a calculation unit 403, an input unit 404, a transmission / reception unit 405, and a display unit 406. The temporary storage unit 401, the storage unit 402, the input unit 404, the transmission / reception unit 405, and the display unit 406 are all connected to the calculation unit 403 via a BUS.

[0054] The calculation unit 403 is, for example, a CPU (Central Processing Unit). The calculation unit 403 executes the processes of the customer information generation unit 11, the statistical information extraction unit 12, the feature generation unit 13, the demand forecasting model generation unit 14, and the demand forecasting model application unit 15 in accordance with the control program stored in the storage unit 402.

[0055] The temporary storage unit 401 is, for example, a RAM (Random-Access Memory). The temporary storage unit 401 loads the control program stored in the storage unit 402 and is used as a work area for the calculation unit 403.

[0056] The storage unit 402 is a non-volatile memory such as a flash memory, a hard disk, a DVD-RAM (Digital Versatile Disc - Random Access Memory), or a DVD-RW (Digital Versatile Disc - Rewritable). The storage unit 402 pre-stores a program for causing the calculation unit 403 to perform the processing of the demand forecasting device 1, and also supplies information stored by this program to the calculation unit 403 in accordance with instructions from the calculation unit 403, and stores the information supplied from the calculation unit 403. The storage unit 10 is configured in the storage unit 402.

[0057] The input unit 404 is an interface device that connects input devices such as a keyboard, a pointing device, and a voice input device to the BUS. Information input by the user is supplied to the calculation unit 403 via the input unit 404. In a configuration in which the user sets the maximum value of K via the time shift unit 134, the input unit 404 functions as the time shift unit 134. In a configuration in which the user narrows down the explanatory variables to be used via the feature selection unit 143, the input unit 404 functions as the feature selection unit 143.

[0058] The transmitting / receiving unit 405 is a network termination device or a wireless communication device that connects to the network, and a serial interface or a LAN (Local Area Network) interface that connects to them. When the demand forecasting model application unit 15 is configured to transmit time-series data indicating the results of product demand forecasting to a user terminal, the transmitting / receiving unit 405 functions as the demand forecasting model application unit 15.

[0059] The display unit 406 is a display device such as an LCD (Liquid Crystal Display), an organic EL (Electroluminescence) display, etc. When the demand forecasting model application unit 15 is configured to display time-series data indicating the results of product demand forecasting on a screen, the transmission / reception unit 405 functions as the demand forecasting model application unit 15.

[0060] The processing of the memory unit 10, customer information generation unit 11, statistical information extraction unit 12, feature generation unit 13, demand forecasting model generation unit 14, and demand forecasting model application unit 15 of the demand forecasting device 1 shown in Figure 1 is executed by a control program using the temporary memory unit 401, calculation unit 403, memory unit 402, input unit 404, transmission / reception unit 405, display unit 406, etc. as resources.

[0061] Furthermore, the above hardware configuration and flowchart are merely examples and can be changed and modified as desired.

[0062] The core processing components of the demand forecasting device 1, such as the calculation unit 403, temporary storage unit 401, storage unit 402, input unit 404, transmission / reception unit 405, and display unit 406, can be realized using an ordinary computer system rather than a dedicated system. For example, a computer program for executing the above operations may be stored and distributed on a computer-readable recording medium such as a flexible disk, a CD-ROM (Compact Disc - Read Only Memory), or a DVD-ROM (Digital Versatile Disc - Read Only Memory), and the demand forecasting device 1 that executes the above processing may be configured by installing the computer program on a computer. Alternatively, the demand forecasting device 1 may be configured by storing the computer program in a storage device of a server device on a communication network, such as the Internet, and downloading it into an ordinary computer system.

[0063] Furthermore, when the functions of the demand forecasting device 1 are realized by sharing the functions between an OS (Operating System) and an application program, or by cooperation between the OS and the application program, only the application program portion may be stored in a recording medium or storage device.

[0064] It is also possible to superimpose a computer program on a carrier wave and provide it via a communication network. For example, the computer program may be posted on a bulletin board system (BBS) on the communication network and provided via the communication network. The computer program may then be started and executed under the control of an OS in the same way as other application programs, thereby enabling the above-mentioned processing to be performed.

[0065] In the above embodiment, the demand forecasting device 1 includes the demand forecasting model generation unit 14 and the demand forecasting model application unit 15, but this is not limited thereto. For example, the demand forecasting model generation device that generates a demand forecasting model and the demand forecasting model application device that applies the demand forecasting model may be separate devices. In this case, the demand forecasting model generation device includes the storage unit 10, the customer information generation unit 11, the statistical information extraction unit 12, the feature generation unit 13, and the demand forecasting model generation unit 14. The demand forecasting model application device includes the customer information generation unit 11, the statistical information extraction unit 12, the feature generation unit 13, and the demand forecasting model application unit 15. The feature generation unit 13 of the demand forecasting model application device acquires time series data of the objective variable to obtain the time resolution of the time series data of the objective variable. Therefore, if the time resolution of the time series data of the objective variable is stored in advance, the objective variable acquisition unit 132 may not be included. In this case, the storage unit 10 may not include the objective variable storage unit 105.

[0066] In the above embodiment, the demand prediction device 1 includes the storage unit 10, but the storage unit 10 may be included in an external device or system.

[0067] In the above embodiment, the demand forecasting system 100 includes a demand forecasting device 1, an SC network generation system 2, and a sales management system 3, but is not limited to this. The SC network generation system 2 and the sales management system 3 may be the same system, or the demand forecasting device 1 may acquire the SC network, statistical information, and time-series data of the objective variable from another device or system.

[0068] Although the preferred embodiments have been described in detail above, the present invention is not limited to the above-described embodiments, and various modifications and substitutions can be made to the above-described embodiments without departing from the scope of the claims.

[0069] Various aspects of the present disclosure are summarized below as appendices.

[0070] (Appendix 1) A demand forecasting device that forecasts demand for a product, a customer information generation unit that acquires and aggregates company information on end-customer companies from a supply chain network of the products, including sales channels for 0th-order products to Nth-order products, and generates customer information indicating the aggregation results; a statistical information extraction unit that extracts statistical information about the end customer based on the customer information; a feature generation unit that generates feature amounts related to the end customer based on statistical information related to the end customer; a demand forecasting model generating unit that generates a demand forecasting model by learning demand fluctuations using the feature quantities as explanatory variables and the demand for the product as a target variable; a demand forecasting model application unit that applies the demand forecasting model to forecast demand for the product; Equipped with Demand forecasting device. (Appendix 2) The customer information generation unit a supply chain network acquisition unit that acquires nodes and edges of the supply chain network; an end customer node acquisition unit that extracts the end customer node from the supply chain network; an aggregation unit that acquires and aggregates the company information held by the end customer node and generates the customer information; Equipped with Demand forecasting device according to appendix 1. (Appendix 3) The statistical information extraction unit a business management information acquisition unit that acquires business management information relating to the business management of the end customer's business based on the customer information; an industry economic information acquisition unit that acquires industry economic information regarding economic trends in the industry of the end customer based on the customer information; Equipped with the statistical information on the end customers is time-series data of the business management information acquired by the business management information acquisition unit and the industry economic information acquired by the industry economic information acquisition unit; 3. The demand forecasting device according to claim 1 or 2. (Appendix 4) The feature generation unit an objective variable acquisition unit that acquires time-series data of the performance of the objective variable; a time resolution unifying unit that unifies time series data of statistical information related to the end customer to the time resolution of the objective variable; a time shift unit that time-shifts the time series data of the statistical information related to the end customer, which has been unified to the time resolution of the objective variable, to match the time series of the objective variable; a statistical information prediction unit that predicts the company management information and the industry economic information by autoregression to supplement the lacking time series data of the statistical information about the end customers that has been time-shifted in accordance with the time series of the objective variable, and generates feature quantities about the end customers; Equipped with 4. The demand forecasting device according to claim 3. (Appendix 5) The demand forecasting model generation unit an objective variable acquisition unit that acquires time-series data of the performance of the objective variable; a learning feature acquisition unit that acquires time-series data of the feature generated by the feature generation unit; a feature quantity selection unit that selects the explanatory variables to be used for learning demand fluctuations from the feature quantities generated by the feature quantity generation unit; a learning unit that learns demand fluctuations using time-series data of the explanatory variables and time-series data of actual results of the objective variables and generates the demand forecasting model; Equipped with 5. A demand forecasting device according to any one of appendixes 1 to 4. (Appendix 6) The demand forecasting model application unit a prediction feature acquisition unit that acquires the feature used for demand forecasting; a demand forecasting unit that inputs the feature quantities acquired by the prediction feature quantity acquisition unit as the explanatory variables into the demand forecasting model and obtains an output of the objective variables; a prediction result output unit that outputs time-series data indicating the result of the demand forecast for the product, which is the output of the objective variable; Equipped with 6. A demand forecasting device according to any one of appendixes 1 to 5. (Appendix 7) a demand forecasting device according to any one of Supplementary Notes 1 to 6; a supply chain network generation system that generates the supply chain network; and a sales management system that manages sales of the products; the supply chain network generation system generates the supply chain network and the statistical information; the sales management system generates time-series data of the performance of the objective variable based on the sales performance of the product. Demand forecasting system. (Appendix 8) A method for generating a demand forecasting model for predicting demand for a product, executed by a demand forecasting model generation device, comprising: Acquire and aggregate company information on end-customer companies from the supply chain network of the products, including sales channels for the zeroth-order products to the Nth-order products, and generate customer information indicating the aggregation results; extracting statistical information about the end customer based on the customer information; generating features related to the end customer based on statistical information related to the end customer; The feature quantities are used as explanatory variables and the demand for the product is used as a target variable, and a demand forecasting model is generated by learning demand fluctuations. Demand forecast model generation method. (Appendix 9) Computer, a customer information generation unit that acquires and aggregates company information on end-customer companies from a product supply chain network including sales channels for 0th-order products to Nth-order products, and generates customer information indicating the aggregation results; a statistical information extraction unit that extracts statistical information regarding the end customer based on the customer information; a feature generation unit that generates features related to the end customer based on statistical information related to the end customer; a demand forecasting model generating unit that generates a demand forecasting model by learning demand fluctuations using the feature quantities as explanatory variables and the demand for the product as a target variable; and a demand forecasting model application unit that applies the demand forecasting model to forecast demand for the product; A program that functions as a [Explanation of symbols]

[0071] 1 Demand forecasting device, 2 SC network generation system, 3 Sales management system, 10 Memory unit, 11 Customer information generation unit, 12 Statistical information extraction unit, 13 Feature generation unit, 14 Demand forecasting model generation unit, 15 Demand forecasting model application unit, 100 Demand forecasting system, 101 SC network memory unit, 102 Customer information memory unit, 103 Statistical information memory unit, 104 Intermediate information memory unit, 105 Objective variable memory unit, 106 Feature memory unit, 107 Demand forecasting model memory unit, 111 SC network acquisition unit, 112 End customer node acquisition unit, 113 Aggregation unit, 121 Customer information acquisition unit, 122 Corporate management information acquisition unit, 123 Industry economic information acquisition unit, 124 Intermediate information storage unit, 131 Intermediate information acquisition unit, 132 Objective variable acquisition unit, 133 Time resolution unification unit, 134 Time shift unit, 135 Statistical information prediction unit, 141 Objective variable acquisition unit, 142 learning feature acquisition unit, 143 feature selection unit, 144 learning unit, 151 prediction feature acquisition unit, 152 demand prediction model acquisition unit, 153 demand prediction unit, 154 prediction result output unit, 401 temporary storage unit, 402 storage unit, 403 calculation unit, 404 input unit, 405 transmission / reception unit, 406 display unit.

Claims

1. A demand forecasting device that forecasts demand for a product, a customer information generation unit that acquires and aggregates company information relating to end-customer companies from a supply chain network of the products, including sales channels for the zeroth-order products to the Nth-order products, and generates customer information indicating the aggregation results; a statistical information extraction unit that extracts statistical information about the end customer based on the customer information; a feature generation unit that generates feature amounts related to the end customer based on statistical information related to the end customer; a demand forecasting model generating unit that generates a demand forecasting model by learning demand fluctuations using the feature quantities as explanatory variables and the demand for the product as a target variable; a demand forecasting model application unit that applies the demand forecasting model to forecast demand for the product; Equipped with Demand forecasting device.

2. The customer information generation unit a supply chain network acquisition unit that acquires nodes and edges of the supply chain network; an end customer node acquisition unit that extracts the end customer node from the supply chain network; an aggregation unit that acquires and aggregates the company information held by the end customer node and generates the customer information; Equipped with The demand forecasting device according to claim 1 .

3. The statistical information extraction unit a business management information acquisition unit that acquires business management information relating to the business management of the end customer's business based on the customer information; an industry economic information acquisition unit that acquires industry economic information regarding economic trends in the industry of the end customer based on the customer information; Equipped with the statistical information on the end customers is time-series data of the business management information acquired by the business management information acquisition unit and the industry economic information acquired by the industry economic information acquisition unit; The demand forecasting device according to claim 1 or 2.

4. The feature generation unit an objective variable acquisition unit that acquires time-series data of the performance of the objective variable; a time resolution unifying unit that unifies time series data of statistical information related to the end customer to the time resolution of the objective variable; a time shift unit that time-shifts the time series data of the statistical information related to the end customer, which has been unified to the time resolution of the objective variable, in accordance with the time series of the objective variable; a statistical information prediction unit that predicts the company management information and the industry economic information by autoregression to supplement the missing time series data of the statistical information about the end customers that has been time-shifted in accordance with the time series of the objective variable, and generates feature quantities about the end customers; Equipped with The demand forecasting device according to claim 3 .

5. The demand forecasting model generation unit an objective variable acquisition unit that acquires time-series data of the performance of the objective variable; a learning feature acquisition unit that acquires time-series data of the feature generated by the feature generation unit; a feature quantity selection unit that selects the explanatory variables to be used for learning demand fluctuations from the feature quantities generated by the feature quantity generation unit; a learning unit that learns demand fluctuations using time-series data of the explanatory variables and time-series data of actual results of the objective variables and generates the demand forecasting model; Equipped with The demand forecasting device according to claim 1 or 2.

6. The demand forecasting model application unit a prediction feature acquisition unit that acquires the feature used for demand forecasting; a demand forecasting unit that inputs the feature quantities acquired by the prediction feature quantity acquisition unit as the explanatory variables into the demand forecasting model and obtains an output of the objective variables; a prediction result output unit that outputs time-series data indicating the result of the demand forecast for the product, which is the output of the objective variable; Equipped with The demand forecasting device according to claim 1 or 2.

7. a supply chain network generation system that generates the supply chain network; and a sales management system that manages sales of the products, the supply chain network generation system generates the supply chain network and the statistical information; the sales management system generates time-series data of the performance of the objective variable based on the sales performance of the product. Demand forecasting system.

8. A method for generating a demand forecasting model for predicting demand for a product, executed by a demand forecasting model generation device, comprising: Acquire and aggregate company information relating to end customer companies from a supply chain network of the products, including sales channels for the zeroth-order product to the Nth-order product, and generate customer information indicating the aggregation results; extracting statistical information about the end customer based on the customer information; generating features related to the end customer based on statistical information related to the end customer; The feature quantities are used as explanatory variables and the demand for the product is used as a target variable, and a demand forecasting model is generated by learning demand fluctuations. Demand forecast model generation method.

9. Computer, a customer information generation unit that acquires and aggregates company information related to end customer companies from a product supply chain network including sales channels for 0th-order products to Nth-order products, and generates customer information indicating the aggregation results; a statistical information extraction unit that extracts statistical information regarding the end customer based on the customer information; a feature generation unit that generates features related to the end customer based on statistical information related to the end customer; a demand forecasting model generating unit that generates a demand forecasting model by learning demand fluctuations using the feature quantities as explanatory variables and the demand for the product as a target variable; and a demand forecasting model application unit that applies the demand forecasting model to forecast demand for the product; A program that functions as a

Images