Demand forecasting device, demand forecasting method, and program

Abstract

To properly use a plurality of forecasting models for demand forecasting, in accordance with applications that utilize a result of demand forecasting.SOLUTION: A demand forecasting apparatus 1 inputs to a prediction model, an objective variable, which is time-series data to be subjected to demand forecasting, and an explanatory variable, which is time-series data to be used for demand forecasting based on the objective variable, for demand forecasting. A data acquisition generation unit 11 of the demand forecasting apparatus 1 acquires an objective variable to generate explanatory variable candidate data which are candidates of an explanatory variable to be used for demand forecasting. A forecasting model diagnosis unit 12 selects a forecasting model suitable to an application which uses a result of demand forecasting, from among a plurality of forecasting models. A demand forecasting unit 13 selects an explanatory variable from the explanatory variable candidate data, inputs the objective variable and the selected explanatory variable to the forecasting model selected in the forecasting model diagnosis unit 12, for demand forecasting. A forecast result output unit 15 outputs visualized demand forecast result data indicating a result of the demand forecasting in the demand forecasting unit 13.SELECTED DRAWING: Figure 1

Description

[Technical field]

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

[0002] Demand forecasts for products and services using machine learning and statistical forecasting methods have become common, but there are still many tasks that must be done by humans, such as selecting a forecast model, setting parameters for the forecast model, and selecting data to use for demand forecasting. In addition, with demand forecasts using time-series data, a wide range of disturbances can affect the forecast, making it difficult to make perfect predictions. Normally, when creating a forecast model, a huge amount of data, memory, and processors are used to calculate and calculate optimal parameters in order to maximize the accuracy of the demand forecast. However, the demand for products and services can easily differ from predictions due to technological innovations and political events, so the results are difficult to see compared to the load of such tuning.

[0003] In response to this, the prediction model selection device described in Patent Document 1 sets an evaluation interval in time-series data, performs demand forecasting for the evaluation interval using multiple prediction models, and adopts the prediction model that produces the best results, thereby reducing the burden on people and equipment. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2006-85646 A Summary of the Invention [Problem to be solved by the invention]

[0005] Generally, the results of demand forecasting are used for other operations such as parts arrangement and inventory management, but the predicted values ​​that users place importance on differ depending on the operation. For example, if the operation is parts arrangement, there is a demand to know the maximum demand value because there is a limit to the amount of orders that can be received from a parts outsourcer and it may be necessary to arrange for more than a certain number of parts by distributing the load. For example, if the operation is inventory management, there is a demand to know the lower limit of demand for a specified period to minimize inventory, or the upper limit of demand for a specified period to minimize opportunity loss.

[0006] The technology described in Patent Document 1 does not take into consideration the type of business for which the calculated demand forecast results will be used, and it is not possible to use multiple forecast models depending on the business for which the demand forecast results will be used.

[0007] The present disclosure aims to enable multiple forecasting models to be used in demand forecasting depending on the business that uses the demand forecast results. [Means for solving the problem]

[0008] In order to achieve the above object, a demand forecasting device according to the present disclosure performs demand forecasting by inputting a dependent variable, which is time-series data to be the subject of demand forecasting, and explanatory variables, which are time-series data used for demand forecasting based on the dependent variable, into a forecasting model. The demand forecasting device includes a data acquisition and generation unit, a forecasting model diagnosis unit, a demand forecasting unit, and a forecast result output unit. The data acquisition and generation unit acquires the dependent variable, and generates explanatory variable candidate data, which are candidates for explanatory variables that can be used for demand forecasting. The forecasting model diagnosis unit Accepts input from users about tasks that use the results of demand forecasts, From multiple prediction models, The input A forecast model suitable for the business that uses the demand forecast results is selected. The demand forecasting unit selects explanatory variables from the explanatory variable candidate data, inputs the objective variable and the selected explanatory variables into the forecast model selected by the forecast model diagnosis unit, and performs demand forecasting. The forecast result output unit visualizes and outputs demand forecast result data that shows the demand forecast results of the demand forecasting unit. Effect of the Invention

[0009] According to the present disclosure, in demand forecasting, by selecting from multiple forecast models a forecast model suitable for the business that uses the demand forecast results, it becomes possible to use multiple forecast models depending on the business that uses the demand forecast results. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram showing a configuration example of a demand forecasting system according to an embodiment. [Diagram 2] FIG. 1 is a diagram showing an example of a response variable according to an embodiment; [Diagram 3] FIG. 1 shows an example of a format for in-house statistical data and external statistical data according to an embodiment. [Figure 4] FIG. 1 is a diagram showing an example of prediction model diagnostic data according to an embodiment; [Diagram 5] FIG. 13 is a diagram showing an example of shift month selection data according to an embodiment; [Figure 6] FIG. 13 is a diagram showing an example of a correlation matrix created from data for selecting a shift month according to an embodiment; [Figure 7] FIG. 1 is a diagram showing an example of explanatory variable selection data according to an embodiment; [Figure 8] FIG. 13 is a diagram showing an example of a correlation matrix created from explanatory variable month selection data according to the embodiment; [Figure 9] FIG. 1 is a diagram showing an example of a prediction data set according to an embodiment. [Figure 10] FIG. 13 is a diagram showing an example of demand forecast result data according to an embodiment; [Figure 11] FIG. 1 is a diagram showing an example of probability distribution calculation data according to an embodiment; [Figure 12] FIG. 13 is a diagram showing an example of a first pattern of a graph that visualizes demand forecast result data using probability distribution calculation data according to an embodiment. [Figure 13] FIG. 13 is a diagram showing an example of a second pattern of a graph that visualizes demand forecast result data using probability distribution calculation data according to the embodiment. [Figure 14] FIG. 13 is a diagram showing an example of a third pattern of a graph that visualizes demand forecast result data using probability distribution calculation data according to the embodiment. [Figure 15] 1 is a flowchart showing a data determination and generation process according to an embodiment. [Figure 16] 1 is a flowchart showing a prediction model diagnosis process according to an embodiment. [Figure 17] Flowchart showing demand forecasting processing according to an embodiment [Figure 18] 1 is a flowchart showing a probability distribution calculation process according to an embodiment. [Figure 19] A flowchart showing a prediction result output process according to an embodiment. [Figure 20] 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 PREFERRED EMBODIMENTS

[0011] A demand forecasting device, a demand forecasting method, and a program according to the present embodiment will be described in detail below with reference to the drawings. In the drawings, the same or corresponding parts are given the same reference numerals. In the following embodiment, an example of a demand forecasting device that predicts product demand with a demand forecast evaluation cycle of "month" will be described. Hereinafter, the time series data of the demand of the product that is the subject of the demand forecast will be referred to as the objective variable, and the time series data used for demand forecast based on the objective variable will be referred to as the explanatory variable. The demand forecasting device inputs the objective variable and explanatory variables for a certain interval, and predicts the demand for the product for a subsequent determined interval.

[0012] The configuration of a demand forecasting system 100 according to an embodiment will be described with reference to FIG. 1. As shown in FIG. 1, the demand forecasting system 100 includes a demand forecasting device 1 that forecasts demand for a product and a storage device 2 in which various databases are configured. Hereinafter, databases will be abbreviated as DB. The storage device 2 includes an in-house statistical DB 21 that stores in-house statistical data, an external statistical DB 22 that stores external statistical data, an explanatory variable candidate DB 23 that stores explanatory variable candidate data, a prediction model diagnosis DB 24 that stores prediction model diagnosis data, and an evaluation parameter DB 25 that stores evaluation parameters. The storage device 2 also includes a demand forecast result DB 26 that stores demand forecast result data, a shift month selection DB 27 that stores shift month selection data, an explanatory variable selection DB 28 that stores explanatory variable selection data, a prediction dataset DB 29 that stores a prediction dataset, and a probability distribution calculation DB 30 that stores probability distribution calculation data.

[0013] The in-house statistical data and the external statistical data are time series data that may affect the demand fluctuation of the product, which are collected within the company and outside the company, respectively. The company here refers to a business entity that provides the product that is the target of the demand forecast and is the entity that performs the demand forecast. The time series data that may affect the demand fluctuation of the product is, for example, time series data such as product inventory, back orders, and consumer trends. The in-house statistical data is, for example, obtained in advance from a system or database within the company and stored in the in-house statistical DB 21. The external statistical data is, for example, obtained in advance from an external aggregation company and stored in the external statistical DB 22. The explanatory variable candidate data is a candidate of an explanatory variable that can be used for demand forecasting, which is generated based on the in-house statistical data and the external statistical data. The prediction model diagnosis data is data that indicates a prediction model suitable for a combination of a business that uses the product demand forecast result and the data type of the objective variable. Details of the data type of the objective variable will be described later. The evaluation parameters are evaluation parameters used in various prediction models. The evaluation parameters are, for example, hyperparameters that set the behavior of a machine learning algorithm.

[0014] The demand forecast result data is time-series data showing the demand forecast result. The shift month selection data is explanatory variables shifted by the shift month by increasing the shift month by one month a fixed number of times in order to select the shift month that maximizes the correlation with the objective function. The explanatory variable selection data is explanatory variables shifted by the shift month that maximizes the correlation with the objective function in order to select explanatory variables to be input to the forecast model together with the objective variable. The forecast dataset is data in a format that allows a combination of the objective variable and the explanatory variable to be input to the forecast model. The probability distribution calculation data is data showing the variance of the demand forecast result data.

[0015] The demand forecasting device 1 includes a data acquisition and generation unit 11 that determines the data type of the objective variable and generates candidate explanatory variable data, a prediction model diagnosis unit 12 that selects a prediction model suitable for the business that uses the demand forecast results, a demand forecasting unit 13 that performs demand forecasting using the prediction model selected by the prediction model diagnosis unit 12, a probability distribution calculation unit 14 that calculates the variance of the demand forecast result data, and a prediction result output unit 15 that visualizes and outputs the demand forecast result data using the calculation results of the probability distribution calculation unit 14.

[0016] The data acquisition and generation unit 11 acquires a response variable. The response variable may be acquired by the data acquisition and generation unit 11 from an external device or system, or may be directly input to the data acquisition and generation unit 11 by a user. An example of the response variable is shown in Fig. 2. As shown in Fig. 2, the response variable is time-series data composed of a year and a value.

[0017] Returning to FIG. 1, the data acquisition and generation unit 11 first determines the data type of the acquired objective variable. In demand forecasting using time-series data, there is a tendency that "each forecasting model exhibits characteristic behavior depending on the characteristics of the objective variable data" and "the accuracy of each forecasting model differs depending on the method of demand forecasting." In view of these tendencies, in this embodiment, it is determined whether the objective variable "has periodicity" and "has noise" and four patterns of data types 1 to 4 are provided that combine the presence and absence of periodicity and noise.

[0018] Hereinafter, a detailed logic of a method for determining the data type of the objective variable by the data acquisition and generation unit 11 will be described. In this logic, the original series of the objective variable is decomposed using an additive model shown in Equation 1. t is the original series of the objective variable, TC t is the trend component, S t is the seasonal component, I t indicates the noise component. O t =TC t +S t +I t ...Formula 1

[0019] The data acquisition and generation unit 11 judges the periodicity of the objective variable. The periodicity is judged by taking the first-order difference of subtracting the next value from the previous value for the objective variable, which is time series data, and calculating the autocorrelation. At this time, in order to exclude calculation of a short period such as a period of 1 to 5 months, the minimum period is set to 6 months. If the maximum value of the autocorrelation shifted by the minimum period or more is equal to or greater than a threshold, the data acquisition and generation unit 11 determines that the objective variable has periodicity by setting the shifted month at which the autocorrelation has the maximum value as the period of the objective variable. If the maximum value of the autocorrelation shifted by the minimum period or more does not reach the threshold, the data acquisition and generation unit 11 judges that the objective variable does not have periodicity. The threshold for judging the presence or absence of periodicity is predetermined, but may be changeable by the user.

[0020] Next, the data acquisition and generation unit 11 performs trend calculation and noise calculation. When the objective variable has periodicity, for example, the data acquisition and generation unit 11 calculates the average value Y i and the overall average Y, and i -Y is calculated as the trend component. If the objective variable does not have periodicity, a moving average over a fixed period (for example, 12 months) is regarded as the trend component. If the objective variable has periodicity, the data acquisition and generation unit 11 calculates the seasonal component S t Since there exists, the original sequence of the objective variable O t From the seasonal component S t and trend component TC t The noise component I is subtracted from tIf the objective variable does not have periodicity, the seasonal component S t does not exist, so the original sequence of the objective variable O t From trend component TC t Only the noise component I is subtracted. t In addition, the original series O t From the seasonal component S t The method for subtracting this will use existing seasonal adjustment methods.

[0021] The data acquisition and generation unit 11 acquires noise components I t If the absolute value of exceeds 1, it is determined that there is noise in the time series data. The data acquisition and generation unit 11 acquires the noise component I t If the absolute value of does not exceed 1, it is determined that there is no noise in the time series data.

[0022] When the data acquisition and generation unit 11 determines that the time series data has periodicity and contains noise, it determines the data type to be 1, and when the data acquisition and generation unit 11 determines that the time series data has periodicity and contains no noise, it determines the data type to be 2. When the data acquisition and generation unit 11 determines that the time series data has no periodicity and contains noise, it determines the data type to be 3, and when the data acquisition and generation unit 11 determines that the time series data has no periodicity and contains no noise, it determines the data type to be 4.

[0023] Next, the data acquisition and generation unit 11 generates explanatory variable candidate data. The data acquisition and generation unit 11 first reads out the in-house statistical data and the external statistical data from the in-house statistical DB 21 and the external statistical DB 22 in the storage device 2. FIG. 3 shows an example of the format of the in-house statistical data and the external statistical data. The in-house statistical data and the external statistical data include a data information part A and a data detail part B. In the example of FIG. 3, the data information part A stores data such as the data name, country, city, classification, and unit. The data detail part B stores numerical values ​​in a time series.

[0024] Returning to FIG. 1, the data acquisition and generation unit 11 removes missing data from the in-house statistical data and the external statistical data read from the in-house statistical DB 21 and the external statistical DB 22, and extracts only data that can be used for demand forecasting. Rules for removing data are set in advance; for example, in China, data for the Chinese New Year (January to February) is often missing, so data from January to February where the country name is China is removed. In addition, data with a large portion of the past data missing, such as the usage status of 5G lines, is often unnecessary for demand forecasting, so it is removed. Hereinafter, the extracted in-house statistical data and external statistical data are referred to as extracted data.

[0025] The data acquisition and generation unit 11 normalizes the extracted data. For example, when a set of extracted data is X, the average is μ, and the standard deviation is σ, the normalization may be performed by calculating (X-μ) / σ, or by taking the maximum value of X as X. MAX , the minimum value is X MIN When (XX MIN ) / (X MAX -X MIN ) may be calculated. The data acquisition and generation unit 11 transmits the normalized extracted data to the storage device 2 as explanatory variable candidate data, and stores the normalized extracted data in the explanatory variable candidate DB 23. The normalized extracted data is a candidate of an explanatory variable that can be used for demand forecasting.

[0026] Returning to Fig. 1, the prediction model diagnosis unit 12 selects a prediction model to be used for demand forecasting. The prediction model diagnosis unit 12 first reads out prediction model diagnosis data from the prediction model diagnosis DB 24 in the storage device 2. An example of the prediction model diagnosis data is shown in Fig. 4. The prediction model diagnosis data includes a business content part C and a prediction model part D.

[0027] The task content section C stores data on tasks that utilize the demand forecast results and the location where the tasks are performed. In the example of FIG. 4, tasks that utilize the demand forecast results include, for example, "annual plan creation" and "personnel plan creation." Locations where the tasks are performed include, for example, "sales company" and "factory." It is sufficient that the task content section C stores at least data on tasks that utilize the demand forecast results. The task content section C may include, for example, data indicating conditions such as the business situation and product sales.

[0028] The prediction model part D stores the predicted value to be emphasized in the business, the data type of the objective variable, and prediction models 1 to 3 suitable for the predicted value to be emphasized. The predicted value to be emphasized in the business is set in advance. For example, when the business is to create an annual plan, the small variations in each month are not emphasized, and the annual total of the large mesh is emphasized. In addition, the prediction model suitable for the predicted value to be emphasized is selected and stored by performing, for example, simulation, testing, etc. in advance to calculate the predicted value to be emphasized for each data type. For example, if the predicted value to be emphasized is the maximum value of the demand, the prediction model with the largest maximum value of the demand for the determined period predicted by simulation, testing, etc. is selected from all the prediction models. In the figure, the prediction model is written as "~", but the prediction model can be, for example, a neural network, XG-Boost, support vector machine, etc. The prediction model suitable for the predicted value to be emphasized is not limited to three, and at least one may be sufficient.

[0029] When the business content section C contains data indicating a situation, the prediction model section D stores the predicted value that is important for use in business under that situation. For example, in inventory management when business conditions are poor, the predicted value of the lower limit of demand is emphasized in order to minimize stock on shelves. In inventory management when product sales are good, the predicted value of the upper limit of demand is emphasized in order to minimize opportunity losses.

[0030] Returning to FIG. 1, the prediction model diagnosis unit 12 outputs a task selection screen that displays a list of tasks included in the prediction model diagnosis data in a selectable manner. The user selects and inputs a task that uses a demand forecast result from the list of tasks displayed on the task selection screen via an interface. When task content portion C includes data indicating a situation, the task selection screen displays a list of tasks included in the prediction model diagnosis data and the situation in a selectable manner. The prediction model diagnosis unit 12 selects a prediction model to be used by referring to the prediction model diagnosis data, using the task input by the user and the data type of the objective variable as keys. The prediction model diagnosis unit 12 stores the prediction model to be used in a temporary storage area.

[0031] If the business content portion C contains data indicating a situation, the prediction model diagnosis unit 12 selects a prediction model to use by referring to the prediction model diagnosis data using the business and situation entered by the user and the data type of the objective variable as keys.

[0032] The demand forecasting unit 13 first normalizes the objective variables. For example, the demand forecasting unit 13 normalizes the objective variables using a method similar to the normalization of extracted data executed by the data acquisition and generation unit 11 when generating the explanatory variable candidate data. Next, the demand forecasting unit 13 selects explanatory variables. The reason for selecting explanatory variables is to prevent overlearning caused by learning too many explanatory variables.

[0033] Here, an example of a method for selecting explanatory variables using correlation coefficients as a scale will be described. The demand forecasting unit 13 reads explanatory variables one by one from the explanatory variable candidate data stored in the explanatory variable candidate DB 23. For example, when making a prediction k months ahead, the demand forecasting unit 13 needs to use data from k months or more back from the month to be predicted. For this reason, the demand forecasting unit 13 first shifts the explanatory variables k months ahead and stores them in the shift month selection data stored in the shift month selection DB 27 of the storage device 2. The demand forecasting unit 13 increases the shift month by one month and repeats this process a certain number of times. The reason for increasing the shift month by one month and repeating the process a certain number of times is that, as described above, when making a prediction k months ahead, data from k months ahead, that is, k+n months ahead, may be used (n is a positive integer).

[0034] An example of the shift month selection data is shown in FIG. 5. In the example of FIG. 5, the demand forecasting unit 13 repeats the process of shifting the explanatory variables by increasing the shift month by one month from k months ago 12 times. The shift month selection data includes an objective variable part E and an explanatory variable part F. The objective variable part E stores the normalized year and month of the objective variable and the objective variable value. The explanatory variable part F stores values ​​of the explanatory variables shifted k months ago to values ​​shifted k months + 12 months ago.

[0035] Returning to FIG. 1, the demand forecasting unit 13 repeats the process of shifting the explanatory variables by increasing the shift month from k months in advance a certain number of times (12 times in the example of FIG. 5), and then calculates the correlation coefficient between each of the objective variables and one of the explanatory variables to create a correlation matrix. The correlation coefficient calculated in the correlation matrix is, for example, Pearson's product-moment correlation coefficient. FIG. 6 shows an example of a correlation matrix created from data for selecting the shift month. In the correlation matrix, the correlation coefficient between the objective variable and the objective variable is always 1, so the shift month of the explanatory variable with the largest correlation coefficient excluding the correlation coefficient between the objective variable and the objective variable is the shift month with the largest correlation with the objective function.

[0036] The demand forecasting unit 13 stores all explanatory variables shifted by the shift month at which the correlation with the objective function is maximized in the explanatory variable selection data stored in the explanatory variable selection DB 28. FIG. 7 shows an example of the explanatory variable selection data. The explanatory variable selection data includes an objective variable part G and an explanatory variable part H. The objective variable part G stores the year and month of the objective variable and the objective variable value normalized in the same manner as the shift month selection data. The explanatory variable part H stores values ​​of each of the explanatory variables 1 to N shifted by the shift month at which the correlation with the objective function is maximized. While the shift month selection DB 27 is a database in which data of one explanatory variable shifted by the month for one objective variable is stored, the explanatory variable selection DB 28 stores data of multiple explanatory variables shifted by the shift month at which the correlation with the objective function is maximized for one objective variable.

[0037] Returning to FIG. 1, the demand forecasting unit 13 calculates the correlation coefficient between each of the objective variables and all the explanatory variables to create a correlation matrix. The correlation calculated in the correlation matrix is, for example, Pearson's product moment correlation coefficient, as in the creation of a correlation matrix between an objective variable and one explanatory variable. FIG. 8 shows an example of a correlation matrix created from explanatory variable month selection data. In the correlation matrix, the correlation coefficient between the objective variable and the objective variable is always 1, so the demand forecasting unit 13 selects a certain number of explanatory variables in descending order of correlation coefficient from the maximum value of the correlation coefficient with the objective variable, excluding the correlation coefficient between the objective variable and the objective variable, and stores them in a prediction dataset stored in the prediction dataset DB 29 of the storage device 2.

[0038] FIG. 9 shows an example of a prediction data set. The prediction data set includes an objective variable part I and an explanatory variable part J. The objective variable part I stores the year and month of the objective variable and the objective variable value normalized in the same manner as the shift month selection data and the explanatory variable selection data. The explanatory variable part J stores data of a certain number of explanatory variables selected in descending order of correlation coefficient from the maximum correlation coefficient, shifted by the shift month at which the correlation with the objective function is maximized. The format of the prediction data set shown in FIG. 9 is a format that can be input to a prediction model. Although the method of selecting explanatory variables using Pearson's product moment correlation coefficient has been described so far, the method of selecting explanatory variables is not limited to this. For example, a certain number of explanatory variables may be selected in descending order of distance from the objective variable using a method of measuring the distance between data, such as DTW (Dynamic Time Warping).

[0039] Returning to FIG. 1, the demand forecasting unit 13 reads out a prediction data set from the prediction data set DB29 of the storage device 2. The demand forecasting unit 13 reads out the evaluation parameters of the prediction model stored in the temporary storage area by the prediction model diagnosis unit 12 from the evaluation parameter DB25 of the storage device 2. The evaluation parameters of each prediction model are stored in advance in the evaluation parameter DB25. The demand forecasting unit 13 gives the prediction data set as an input to the prediction model stored in the temporary storage area by the prediction model diagnosis unit 12, and performs demand forecasting for a determined period using the evaluation parameters. In the case of a prediction model in which a seed value is set, as typified by machine learning, the demand forecasting unit 13 changes the seed value and performs prediction multiple times (for example, 1000 times). The demand forecasting unit 13 performs a process of returning normalized values ​​to their original values ​​for all demand forecast results. The demand forecasting unit 13 stores demand forecast result data indicating the demand forecast result in which the normalized values ​​have been returned to their original values ​​in the demand forecast result DB26 of the storage device 2.

[0040] Fig. 10 shows an example of demand forecast result data. The demand forecast result data includes a forecast method part L and a forecast result part M. The forecast method part L stores the forecast model used and the seed value that was set. The forecast result part M stores data in which the normalized values ​​of the demand forecast result are restored to the original values.

[0041] Returning to FIG. 1, when the demand forecasting unit 13 changes the seed value and performs prediction multiple times (for example, 1000 times), the demand forecast result has a certain degree of variation. The probability distribution calculation unit 14 quantifies the variation of the demand forecast result data. Here, the maximum value, the first quartile, the second quartile, the third quartile, and the minimum value are used as statistical values ​​for quantifying the variation. The statistical values ​​for quantifying the variation may be determined depending on the degree to which the variation of the demand forecast result data is to be visualized. Specifically, the probability distribution calculation unit 14 reads out the demand forecast result data of different seed values ​​for a month of interest of one forecasting model from the demand forecast result data stored in the demand forecast result DB 26 of the storage device 2, and calculates the maximum value, the first quartile, the second quartile, the third quartile, and the minimum value. The probability distribution calculation unit 14 repeats this process for all years and months of all prediction models, and stores the probability distribution calculation data indicating the calculation results, that is, the maximum value, the first quartile, the second quartile, the third quartile, and the minimum value, in the probability distribution calculation DB30.

[0042] An example of the probability distribution calculation data is shown in Fig. 11. The probability distribution calculation data has data on the maximum value, the first quartile, the second quartile, the third quartile, and the minimum value for each year and month calculated by the probability distribution calculation unit 14 for each prediction model.

[0043] Returning to Fig. 1, the prediction result output unit 15 reads out the probability distribution calculation data stored in the probability distribution calculation DB 30 of the storage device 2, and visualizes the demand prediction result data using the probability distribution calculation data. For example, the prediction result output unit 15 displays values ​​indicated by the probability distribution calculation data in the form of a graph. In this embodiment, from the viewpoint of providing information useful for business operations that utilize the demand prediction results, the following three example patterns are adopted as graphs to be displayed by the prediction result output unit 15.

[0044] 12 shows an example of a first pattern of a graph that visualizes demand forecast result data using the probability distribution calculation data according to the embodiment. In the graph of the first pattern, the second quartile = median indicated by the probability distribution calculation data for all forecast models (three types in the example of FIG. 12) is shown as a line graph.

[0045] Fig. 13 shows an example of a second pattern of a graph that visualizes demand forecast result data using the probability distribution calculation data according to the embodiment. In the graph of the second pattern, in addition to a line graph of the second quartile = median for all forecast models (three types in the example of Fig. 13), the area between the maximum and minimum values ​​indicated by the probability distribution calculation data, or the area between the first and third quartiles, is colored and graphed to display the upper and lower limits that the demand can take. Whether to display the graph of the maximum and minimum values ​​or the graph of the first and third quartiles depends on the level of risk that the user wants to grasp, and therefore the user can change the display.

[0046] FIG. 14 shows an example of a third pattern of a graph that visualizes demand forecast result data using the probability distribution calculation data according to the embodiment. The graph of the third pattern displays a composite distribution for all prediction models (three types in the example of FIG. 14). The graph of the third pattern can be viewed as one distribution by extracting data for a certain year and month in addition to the graph of the second pattern. The viewable distribution is, for example, a distribution in which the average and standard deviation of multiple (e.g., 1000) prediction data are calculated and the possible values ​​are expressed as a probability distribution, assuming that the variation of each prediction model forms a normal distribution. Note that the graph that visualizes the demand forecast result data using the probability distribution calculation data may be displayed on a screen by the prediction result output unit 15 and presented to the user, or may be transmitted to a terminal used by the user and displayed.

[0047] Here, the flow of the process executed by the demand forecasting device 1 will be described with reference to Figs. 15 to 19. The data determination and generation process shown in Fig. 15 starts when the data acquisition and generation unit 11 of the demand forecasting device 1 acquires a response variable. The data acquisition and generation unit 11 determines whether or not the input response variable has periodicity (step S11). Specifically, if the maximum value of the autocorrelation shifted by the minimum period or more is equal to or greater than a threshold value, the data acquisition and generation unit 11 determines that the response variable has periodicity by taking the shifted month at which the autocorrelation became maximum as the period of the response variable. If the maximum value of the autocorrelation shifted by the minimum period or more does not reach the threshold value, the data acquisition and generation unit 11 determines that the response variable does not have periodicity.

[0048] When it is determined that the objective variable has periodicity (step S11; YES), the data acquisition and generation unit 11 determines whether or not the objective variable has noise (step S12). Specifically, the data acquisition and generation unit 11 determines, for example, the average value of the objective variable in the i month of each year. Y i and the overall average Y And, Y i - Y is calculated as the trend component. When the time series data has periodicity, the data acquisition and generation unit 11 calculates the noise component by subtracting the seasonal component and the trend component from the original series of the objective variable. When the absolute value of the noise component for the most recent three years exceeds 1, the data acquisition and generation unit 11 determines that the time series data contains noise. When the absolute value of the noise component for the most recent three years does not exceed 1, the data acquisition and generation unit 11 determines that the time series data does not contain noise.

[0049] If it is determined that the objective variable has noise (step S12; YES), the data acquisition and generation unit 11 determines that the data type of the objective variable is data type 1 (step S13). If it is determined that the objective variable does not have noise (step S12; NO), the data acquisition and generation unit 11 determines that the data type of the objective variable is data type 2 (step S14).

[0050] On the other hand, when it is determined that the objective variable has no periodicity (step S11; NO), the data acquisition and generation unit 11 determines whether the objective variable has noise or not (step S15). Specifically, the data acquisition and generation unit 11 regards the moving average of a determined period (for example, 12 months) as the trend component. When the time series data has no periodicity, the data acquisition and generation unit 11 calculates the noise component by subtracting only the trend component from the original series of the objective variable. When the absolute value of the noise component for the most recent three years exceeds 1, the data acquisition and generation unit 11 determines that the time series data has noise. When the absolute value of the noise component for the most recent three years does not exceed 1, the data acquisition and generation unit 11 determines that the time series data has no noise.

[0051] If it is determined that the objective variable has noise (step S15; YES), the data acquisition and generation unit 11 determines that the data type of the objective variable is data type 3 (step S16). If it is determined that the objective variable does not have noise (step S15; NO), the data acquisition and generation unit 11 determines that the data type of the objective variable is data type 4 (step S17).

[0052] Next, the data acquisition and generation unit 11 reads out the in-house statistical data and the external statistical data from the in-house statistical DB 21 and the external statistical DB 22 in the storage device 2 (step S18). As shown in Fig. 3, the in-house statistical data and the external statistical data include a data information part A and a data details part B. In the example of Fig. 3, the data information part A stores data such as the data name, country, city, classification, and unit. The data details part B stores numerical values ​​in a time series.

[0053] 15, the data acquisition and generation unit 11 removes missing data from the in-house statistical data and the external statistical data read from the in-house statistical DB 21 and the external statistical DB 22, and extracts only data that can be used for demand forecasting (step S19). The rules for removing data are preset.

[0054] The data acquisition and generation unit 11 normalizes the extracted data (step S20). For example, when a set of extracted data is X, the average is μ, and the standard deviation is σ, the normalization may be performed by calculating (X-μ) / σ, or by calculating the maximum value of X as X MAX , the minimum value is X MIN When (XX MIN ) / (X MAX -X MIN ) may be calculated. The data acquisition and generation unit 11 stores the normalized extracted data as explanatory variable candidate data in the explanatory variable candidate DB 23 of the storage device 2 (step S21), and ends the process. The normalized extracted data is a candidate of an explanatory variable that can be used for demand forecasting. Steps S18 to S21 are an example of a data generation step.

[0055] The prediction model diagnosis process shown in FIG. 16 starts when the data acquisition and generation unit 11 transmits explanatory variable candidate data to the storage device 2. The prediction model diagnosis unit 12 reads out prediction model diagnosis data from the prediction model diagnosis DB 24 of the storage device 2 (step S31). As shown in FIG. 4, the prediction model diagnosis data includes a task content part C and a prediction model part D. In the example of FIG. 4, the task content part C stores tasks that use demand forecast results, such as "annual plan creation" and "personnel plan creation," and data on locations where tasks are performed, such as "sales company" and "factory." The prediction model part D stores a predicted value that is important for use in the task, a data type of the objective variable, and prediction models 1 to 3 that are suitable for the predicted value that is important.

[0056] Returning to FIG. 16, the prediction model diagnosis unit 12 outputs a task selection screen that displays a list of tasks included in the prediction model diagnosis data in a selectable manner (step S32). The prediction model diagnosis unit 12 judges whether or not a task that uses the demand forecast result has been selected and input by the user (step S33). When the user has not input a task that uses the demand forecast result (step S33; NO), the prediction model diagnosis unit 12 repeats step S33 and waits for input. When the user has input a task that uses the demand forecast result (step S33; YES), the prediction model diagnosis unit 12 selects a prediction model to be used by referring to the prediction model diagnosis data using the task that uses the input demand forecast result and the data type of the objective variable as keys (step S34). The prediction model diagnosis unit 12 stores the prediction model to be used in a temporary storage area (step S35) and ends the process. Steps S31 to S35 are examples of prediction model diagnosis steps.

[0057] 17 starts when the prediction model diagnosis unit 12 stores the prediction model to be used in a temporary storage area. The demand prediction unit 13 normalizes the objective variables (step S41). The demand prediction unit 13 normalizes the objective variables using, for example, a method similar to the normalization performed by the data acquisition and generation unit 11 in step S20 of the data determination and generation process.

[0058] Next, the demand forecasting unit 13 reads explanatory variables one by one from the explanatory variable candidate data stored in the explanatory variable candidate DB 23 (step S42). When making a prediction k months ahead, the demand forecasting unit 13 needs to use data from k months or more back from the month to be predicted. For this reason, the demand forecasting unit 13 first shifts the read explanatory variables k months ahead (step S43) and stores them in the shift month selection data stored in the shift month selection DB 27 of the storage device 2 (step S44). The demand forecasting unit 13 judges whether or not steps S43 and S44 have been repeated a certain number of times (step S45). If the certain number of times has not been repeated (step S45; NO), the demand forecasting unit 13 increases the shift month by one month at a time (step S46) and repeats steps S43 and S44. The reason for repeating the process a certain number of times is that when making a prediction k months ahead, data from k months ahead, that is, k+n months ahead, may be used (n is a positive integer).

[0059] An example of the shift month selection data is shown in Fig. 5. In the example of Fig. 5, the demand forecasting unit 13 repeats the process of shifting the explanatory variables by increasing the shift month by one month from k months ago 12 times. The shift month selection data includes an objective variable part E and an explanatory variable part F. The objective variable part E stores the year and month of the objective variable and the value of the objective variable. The explanatory variable part F stores values ​​of the explanatory variables shifted k months ago to values ​​shifted k months + 12 months ago.

[0060] Returning to FIG. 17, when the process has been repeated a certain number of times (step S45; YES), the demand forecasting unit 13 calculates the correlation coefficient between each of the objective variables and one explanatory variable, and creates a correlation matrix between the objective variable and one explanatory variable (step S47). The correlation coefficient calculated in the correlation matrix is, for example, Pearson's product-moment correlation coefficient. The demand forecasting unit 13 selects the shift month that maximizes the correlation with the objective function (step S48). FIG. 6 shows an example of a correlation matrix created from data for selecting the shift month. In the correlation matrix, the correlation coefficient between the objective variable and the objective variable is always 1, so the shift month of the explanatory variable that has the largest correlation coefficient excluding the correlation coefficient between the objective variable and the objective variable is the shift month that maximizes the correlation with the objective function.

[0061] Returning to Fig. 17, the demand forecasting unit 13 stores all explanatory variables shifted by the shift month at which the correlation with the objective function is maximized in explanatory variable selection data stored in the explanatory variable selection DB 28 (step S49). An example of the explanatory variable selection data is shown in Fig. 7. The explanatory variable selection data includes an objective variable part G and an explanatory variable part H. The objective variable part G stores the year and month of the objective variable and the objective variable value, similar to the shift month selection data. The explanatory variable part H stores values ​​of each of the explanatory variables 1 to N shifted by the shift month at which the correlation with the objective function is maximized.

[0062] Returning to FIG. 17, the demand forecasting unit 13 judges whether or not the processes of steps S43 to S49 have been performed for all explanatory variables included in the explanatory variable candidate data (step S50). If the processes have not been performed for all explanatory variables (step S50; NO), the demand forecasting unit 13 returns to step S43 and repeats the processes of steps S43 to S49. If the processes have been performed for all explanatory variables (step S50; YES), the demand forecasting unit 13 calculates the correlation coefficients between the objective variable and all explanatory variables, and creates a correlation matrix between the objective variable and all explanatory variables (step S51). As in step S47, the correlation calculated in the correlation matrix is, for example, Pearson's product moment correlation coefficient. FIG. 8 shows an example of a correlation matrix created from explanatory variable month selection data.

[0063] 17, since the correlation coefficient between the objective variables and the objective variables in the correlation matrix is ​​surely 1, the demand forecasting unit 13 selects a certain number of explanatory variables in descending order of correlation coefficient from the maximum value of the correlation coefficient excluding the correlation coefficient between the objective variables and the objective variables (step 52), and stores them in the prediction dataset stored in the prediction dataset DB 29 of the storage device 2 (step 53). The reason for selecting explanatory variables is to prevent overlearning caused by learning too many explanatory variables.

[0064] An example of a prediction data set is shown in Fig. 9. The prediction data set includes a response variable part I and an explanatory variable part J. The response variable part I stores the year and month of the response variable and the value of the response variable, as well as the shift month selection data and explanatory variable selection data. The explanatory variable part J stores data of a certain number of explanatory variables selected in descending order of correlation coefficient from the maximum correlation coefficient, shifted by the shift month at which the correlation with the objective function is maximized.

[0065] Returning to FIG. 17, the demand forecasting unit 13 reads out a forecast data set from the forecast data set DB29 in the storage device 2. The demand forecasting unit 13 reads out the evaluation parameters of the forecast model stored in the temporary storage area by the forecast model diagnosis unit 12 from the evaluation parameter DB25 in the storage device 2 (step S54). The evaluation parameters of each forecast model are stored in advance in the evaluation parameter DB25. The demand forecasting unit 13 gives the forecast data set as an input to the forecast model stored in the temporary storage area by the forecast model diagnosis unit 12, and executes demand forecast for a determined period using the evaluation parameters (step S55). In the case of a forecasting model in which a seed value is set, as typified by machine learning, the demand forecasting unit 13 changes the seed value and performs prediction multiple times (for example, 1000 times). The demand forecasting unit 13 performs a process of returning normalized values ​​to their original values ​​for all demand forecast results. The demand forecasting unit 13 stores demand forecast result data indicating the demand forecast result in which the normalized values ​​have been returned to their original values ​​in the demand forecast result DB26 in the storage device 2 (step S56), and ends the process. Steps S41 to S56 are an example of a demand forecasting step.

[0066] Fig. 10 shows an example of demand forecast result data. The demand forecast result data includes a forecast method part L and a forecast result part M. The forecast method part L stores the forecast model used and the seed value that was set. The forecast result part M stores data in which the normalized values ​​of the demand forecast result are restored to the original values.

[0067] The probability distribution calculation process shown in Fig. 18 starts when the demand forecast unit 13 stores the demand forecast result data in the demand forecast result DB 26 of the storage device 2. The probability distribution calculation unit 14 reads out the demand forecast result data of different seed values ​​of a month of interest for one forecast model from the demand forecast result data stored in the demand forecast result DB 26 (step S61), and calculates a statistical value that quantifies the variation (step S62). For example, the maximum value, the first quartile, the second quartile, the third quartile, and the minimum value are used as the statistical value that quantifies the variation. The statistical value that quantifies the variation may be determined depending on the degree to which the variation in the demand forecast result data is to be visualized.

[0068] The probability distribution calculation unit 14 determines whether or not the processes of steps S61 and S62 have been performed for all years and months of all prediction models (step S63). If the processes have not been performed for all years and months of all prediction models (step S63; NO), the process returns to step S61 and repeats the processes of steps S61 and S62. If the processes have been performed for all explanatory variables (step S63; YES), the probability distribution calculation unit 14 stores probability distribution calculation data indicating the calculated statistical values ​​in the probability distribution calculation DB 30 (step S64), and ends the process.

[0069] An example of the probability distribution calculation data is shown in Fig. 11. The probability distribution calculation data has data on the maximum value, the first quartile, the second quartile, the third quartile, and the minimum value for each year and month calculated by the probability distribution calculation unit 14 for each prediction model.

[0070] The prediction result output process shown in Fig. 19 starts when the probability distribution calculation unit 14 stores the probability distribution calculation data in the probability distribution calculation DB 30. The prediction result output unit 15 reads out the probability distribution calculation data stored in the probability distribution calculation DB 30 of the storage device 2 (step S71). The prediction result output unit 15 visualizes the demand prediction result data using the read out probability distribution calculation data (step S72) and ends the process. For example, the prediction result output unit 15 displays the values ​​indicated by the probability distribution calculation data in a graph. Steps S71 and S72 are examples of prediction result output steps.

[0071] 12 shows an example of a first pattern of a graph that visualizes demand forecast result data using the probability distribution calculation data according to the embodiment. In the graph of the first pattern, the second quartile = median indicated by the probability distribution calculation data for all forecast models (three types in the example of FIG. 12) is shown as a line graph.

[0072] 13 shows an example of a second pattern of a graph that visualizes the demand forecast result data using the demand forecast result data by using the probability distribution calculation data according to the embodiment. In the graph of the second pattern, in addition to a line graph of the second quartile = median for all prediction models (three types in the example of FIG. 13), the area between the maximum and minimum values ​​indicated by the probability distribution calculation data, or the area between the first and third quartiles, is colored and graphed to display the upper and lower limits that the demand can take. The user may be able to change whether to display the graph of the maximum and minimum values ​​or the graph of the first and third quartiles.

[0073] FIG. 14 shows an example of a third pattern of a graph that visualizes demand forecast result data using probability distribution calculation data according to an embodiment. The third pattern graph displays a composite distribution for all prediction models (three types in the example of FIG. 14). In addition to the second pattern graph, the third pattern graph can extract data for a certain year and month and view it as a single distribution. The viewable distribution is, for example, a distribution in which the average and standard deviation of multiple (e.g., 1000) prediction data are calculated and the possible values ​​are expressed as a probability distribution, assuming that the variation of each prediction model forms a normal distribution.

[0074] According to the demand forecasting device 1 of the embodiment, by selecting a forecasting model suitable for a task that uses the demand forecast results from among multiple forecasting models, it becomes possible to use multiple forecasting models depending on the task that uses the demand forecast results. In addition, by providing a user with forecast results using a forecasting model suitable for the task, it is expected to reduce the burden on people and equipment in the task that uses the demand forecast results, and to reduce management risks such as opportunity loss and increased inventory.

[0075] The hardware configuration of the demand forecasting device 1 will be described with reference to Fig. 20. As shown in Fig. 20, the demand forecasting device 1 includes a temporary storage unit 101, a storage unit 102, a calculation unit 103, an input unit 104, a transmission / reception unit 105, and a display unit 106. The temporary storage unit 101, the storage unit 102, the input unit 104, the transmission / reception unit 105, and the display unit 106 are all connected to the calculation unit 103 via a BUS.

[0076] The calculation unit 103 is, for example, a CPU (Central Processing Unit). The calculation unit 103 executes the processes of the data acquisition and generation unit 11, the prediction model diagnosis unit 12, the demand prediction unit 13, the probability distribution calculation unit 14, and the prediction result output unit 15 in accordance with a control program stored in the storage unit 102.

[0077] The temporary storage unit 101 is, for example, a random-access memory (RAM). The temporary storage unit 101 loads a control program stored in the storage unit 102 and is used as a working area for the calculation unit 103.

[0078] The storage unit 102 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 102 pre-stores a program for causing the calculation unit 103 to perform processing of the demand forecasting device 1, and also supplies data stored by this program to the calculation unit 103 according to instructions from the calculation unit 103, and stores the data supplied from the calculation unit 103. Various DBs configured in the storage device 2 may be configured in the storage unit 102.

[0079] The input unit 104 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 a user is supplied to the calculation unit 103 via the input unit 104. In a configuration in which a user directly inputs a target variable to the data acquisition and generation unit 11, the input unit 104 functions as the data acquisition and generation unit 11.

[0080] The transmitting / receiving unit 105 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. In a configuration in which the data acquisition / generation unit 11 acquires a target variable from an external device or system, the transmitting / receiving unit 105 functions as the data acquisition / generation unit 11. In a configuration in which the prediction result output unit 15 transmits a graph in which the demand prediction result data is visualized using the probability distribution calculation data to a terminal used by a user, the transmitting / receiving unit 105 functions as the prediction result output unit 15.

[0081] The display unit 106 is a display device such as an LCD (Liquid Crystal Display), an organic EL (Electro Luminescence) display, etc. In a configuration in which the demand forecasting device 1 displays a graph that visualizes the demand forecast result data using the probability distribution calculation data, the display unit 106 functions as the forecast result output unit 15.

[0082] The processing of the data acquisition and generation unit 11, the prediction model diagnosis unit 12, the demand prediction unit 13, the probability distribution calculation unit 14 and the prediction result output unit 15 of the demand prediction device 1 shown in Figure 1 is executed by a control program using the temporary memory unit 101, the calculation unit 103, the memory unit 102, the input unit 104, the transmission / reception unit 105, the display unit 106, etc. as resources.

[0083] Additionally, the above hardware configuration and flowchart are merely examples and can be changed or modified as desired.

[0084] The core parts of the demand prediction device 1, such as the calculation unit 103, the temporary storage unit 101, the storage unit 102, the input unit 104, the transmission / reception unit 105, and the display unit 106, can be realized by using a normal computer system, not a dedicated system. For example, the demand prediction device 1 that executes the above-mentioned processing may be configured by storing and distributing 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 installing the computer program on a computer. Alternatively, the demand prediction 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 the computer program by a normal computer system.

[0085] Furthermore, when the functions of the demand forecasting device 1 are realized by sharing the functions of 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.

[0086] It is also possible to superimpose a computer program on a transport wave and provide it via a communication network. For example, the computer program may be posted on a bulletin board (BBS, Bulletin Board System) on the communication network and provided via the communication network. Then, the computer program may be started and executed under the control of the OS in the same way as other application programs, thereby enabling the above-mentioned processing to be performed.

[0087] In the above embodiment, the demand prediction device 1 predicts the demand for a product, but the demand prediction device 1 is not limited to this and may predict any material or service whose demand fluctuates. For example, the demand prediction device 1 may predict the demand for a service, agricultural products, etc. In this case, the objective variable is time series data of the demand for the service, agricultural products, etc. that are the target of the demand prediction.

[0088] In the above embodiment, an example in which the evaluation cycle of the demand forecast is "monthly" has been described, but this is not limiting. For example, the evaluation cycle of the demand forecast may be "weekly," "half-monthly," "quarterly," "half a year," "yearly," etc.

[0089] In the above embodiment, the storage device 2 is configured with an in-house statistics DB21, an external statistics DB22, an explanatory variable candidate DB23, a prediction model diagnosis DB24, an evaluation parameter DB25, a demand forecast result DB26, a shift month selection DB27, an explanatory variable selection DB28, a prediction data set DB29, and a probability distribution calculation DB30, but some or all of these DBs may be configured in the storage device of the demand forecasting device 1.

[0090] In the above embodiment, the prediction result output unit 15 visualizes and outputs the demand prediction result data using the calculation result of the probability distribution calculation unit 14, but this is not limited to the above. The prediction result output unit 15 may visualize and output the demand prediction result data without using the calculation result of the probability distribution calculation unit 14. In this case, the demand prediction device 1 does not need to include the probability distribution calculation unit 14.

[0091] In the above embodiment, the data acquisition and generation unit 11 generates explanatory variable candidate data based on in-house statistical data and external statistical data, but the present invention is not limited to this, and explanatory variable candidate data may be generated based on either one of them. Also, the entity that performs demand forecasting is not limited to the business entity that provides the product that is the target of the demand forecasting. [Explanation of symbols]

[0092] 1 Demand forecasting device, 2 Memory device, 11 Data acquisition and generation unit, 12 Forecasting model diagnosis unit, 13 Demand forecasting unit, 14 Probability distribution calculation unit, 15 Forecast result output unit, 21 In-house statistical DB, 22 External statistical DB, 23 Explanatory variable candidate DB, 24 Forecasting model diagnosis DB, 25 Evaluation parameter DB, 26 Demand forecast result DB, 27 Shift month selection DB, 28 Explanatory variable selection DB, 29 Forecast dataset DB, 30 Probability distribution calculation DB, 100 Demand forecasting system, 101 Temporary storage unit, 102 Storage unit, 103 Calculation unit, 104 Input unit, 105 Transmission / reception unit, 106 Display unit.

Claims

1. A demand forecasting device that performs demand forecasting by inputting a response variable, which is time-series data to be used for demand forecasting based on the response variable, into a forecasting model, the device comprising: a data acquisition and generation unit that acquires the objective variables and generates explanatory variable candidate data that is a candidate for the explanatory variables that can be used for demand forecasting; a prediction model diagnosis unit that receives an input of a business that utilizes the demand forecast result from a user, and selects, from a plurality of the prediction models, a prediction model that is suitable for the business that utilizes the input demand forecast result; a demand forecasting unit that selects the explanatory variables from the explanatory variable candidate data, inputs the objective variable and the selected explanatory variables into the prediction model selected by the prediction model diagnosis unit, and performs demand forecasting; a forecast result output unit that visualizes and outputs demand forecast result data indicating a demand forecast result of the demand forecast unit; A demand forecasting device comprising:

2. A probability distribution calculation unit that calculates the variance of the demand forecast result data, The prediction result output unit visualizes and outputs the demand prediction result data using the calculation result of the probability distribution calculation unit. The demand forecasting device according to claim 1 .

3. The data acquisition and generation unit removes missing data from the in-house statistical data and the external statistical data, and normalizes the data to generate the explanatory variable candidate data. The demand forecasting device according to claim 1 or 2.

4. the data acquisition and generation unit determines a data type based on the presence or absence of periodicity and the presence or absence of noise in the objective variable; the prediction model diagnosis unit selects, from the plurality of prediction models, the prediction model suitable for a combination of data types of the objective variables determined to be a business utilizing the input demand forecast result; The demand forecasting device according to claim 1 .

5. The demand forecasting unit is normalizing the dependent variable, selecting a certain number of the explanatory variables from the explanatory variable candidate data in order of the greatest correlation with the dependent variable, inputting the normalized dependent variable and the certain number of the selected explanatory variables into the prediction model selected by the prediction model diagnosis unit, and performing demand forecasting; The demand forecasting device according to claim 1 .

6. A demand forecasting device that performs demand forecasting by inputting a response variable, which is time-series data to be used for demand forecasting based on the response variable, and explanatory variables, which are time-series data used for demand forecasting based on the response variable, into a forecasting model, comprising: a prediction model diagnosis unit that receives an input of a business that utilizes the demand forecast result from a user, and selects, from a plurality of the prediction models, a prediction model that is suitable for the business that utilizes the input demand forecast result; a demand forecasting unit that inputs the objective variable and the explanatory variable into the prediction model selected by the prediction model diagnosis unit and performs demand forecasting; A demand forecasting device comprising:

7. A demand forecasting method executed by a demand forecasting device that performs demand forecasting by inputting a response variable, which is time-series data to be subjected to demand forecasting, and an explanatory variable, which is time-series data used for demand forecasting based on the response variable, into a forecasting model, comprising: a data generation step of generating explanatory variable candidate data which is a candidate of the explanatory variable available for demand forecasting; a prediction model diagnosis step of receiving an input of a business using the demand forecast result from a user, and selecting, from a plurality of the prediction models, a prediction model suitable for the business using the input demand forecast result; a demand forecasting step of selecting the explanatory variables from the explanatory variable candidate data, inputting the objective variable and the selected explanatory variables into the prediction model selected in the prediction model diagnosis step, and performing a demand forecast; a forecast result output step of visualizing and outputting demand forecast result data indicating a demand forecast result of the demand forecast step; The demand forecasting method includes:

8. Computer, a data acquisition and generation unit that acquires a response variable that is time-series data to be subjected to demand forecasting, and generates explanatory variable candidate data that is a candidate for an explanatory variable that is time-series data to be used in the demand forecasting based on the response variable; a prediction model diagnosis unit that receives an input of a business operation that utilizes the demand forecast result from a user, and selects, from a plurality of prediction models, a prediction model that is suitable for the business operation that utilizes the input demand forecast result; a demand forecasting unit that selects the explanatory variables from the explanatory variable candidate data, inputs the objective variable and the selected explanatory variables into the prediction model selected by the prediction model diagnosis unit, and performs demand forecasting; and a forecast result output unit that visualizes and outputs demand forecast result data indicating the demand forecast result of the demand forecast unit; A program that functions as a

Images