JP7532707B2 - Prediction device and prediction method - Google Patents

Description

The present invention relates to a prediction device and a prediction method for predicting data.

Prediction technologies that use past numerical data to forecast future figures have been developed and are being applied in a variety of fields. For example, in the logistics field, predictive technologies are used to forecast the volume of goods being shipped and received at each warehouse, as well as to estimate the number of staff and storage space required within the warehouse. These volume forecasts are mainly performed manually by veteran workers with many years of experience (for example, on-site leaders), and there is a demand for automation and improved accuracy.

To improve prediction accuracy, it is desirable to make predictions by taking into account as comprehensive a range of variable factors (explanatory variables) that affect the prediction results as possible. However, when making predictions at logistics sites such as logistics warehouses, if it is too much work to obtain, manage, and use the data (explanatory variables) required for prediction, it places a burden on users of the prediction system. Therefore, it is necessary to carefully select explanatory variables and make highly accurate predictions using the minimum number of explanatory variables necessary.

For example, when a 3PL (Third Party Logistics) company that undertakes logistics operations for a shipper forecasts warehouse shipment volumes and the like, it may not be able to easily obtain the shipper's data due to contractual or geographical issues (such as the warehouse being geographically far from the store selling the goods). Even if the data is available, if there are many types of data and each piece of data is large in volume, the cost (effort) of managing that data and the effort required to input the data into the forecasting system will increase, making users of the forecasting system hesitant to use it.

Patent Document 1 also discloses a system for predicting a shipment forecast for a practical product. This system includes an information input means for inputting past shipment volume performance and weather conditions, a data series creation means for setting a data series for a dependent variable and a data series for an explanatory variable from information from the information input means, a multiple regression analysis means for generating a multiple regression model that fits the dependent variable data series and the explanatory variable data series and outputting a shipment volume performance theoretical value and a shipment volume forecast value using the multiple regression model, and a correction means for setting a rejection value of the shipment volume performance value as a corrected shipment volume performance value, calculating the ratio of the corrected shipment volume performance value and the shipment volume performance theoretical value as a correction value, and correcting the shipment volume forecast value using the correction value.

JP 2005-78277 A

In Patent Document 1, shipment volumes are predicted using a set of data on various variable factors such as weather and the number of business days, in addition to past shipment volumes, but the system does not allow users to determine the degree of contribution of the data sets used in the prediction to the predicted results.

The present invention aims to improve prediction accuracy.

A prediction device according to one aspect of the invention disclosed in the present application is a prediction device that predicts a predicted value of a prediction target based on actual values of a prediction target and time series data of each of a plurality of explanatory variables, and includes a processor that executes a program and a storage device that stores the program. The processor calculates prediction accuracy based on the actual values and the predicted values, calculates a first predicted value predicted using time series data of a first explanatory variable among the plurality of explanatory variables, a second predicted value predicted using time series data of a second explanatory variable different from the first explanatory variable among the plurality of explanatory variables, and a first contribution degree, which is the degree of influence of the first explanatory variable on the predicted value, and a second contribution degree, which is the degree of influence of the second explanatory variable on the predicted value, based on the prediction accuracy, and determines an explanatory variable to be excluded from the plurality of explanatory variables based on the first contribution degree and the second contribution degree, and outputs the excluded variable.

According to a representative embodiment of the present invention, prediction accuracy can be improved. Problems, configurations, and effects other than those described above will become clear from the explanation of the following examples.

FIG. 1 is a block diagram illustrating an example of a hardware configuration of a prediction device. FIG. 2 is a block diagram showing a first example of a functional configuration of a prediction device. FIG. 3 is a flowchart illustrating a first example of a prediction process procedure performed by the prediction device. FIG. 4 is an explanatory diagram illustrating an example of input data to the prediction device. FIG. 5 is an explanatory diagram illustrating an example of the first output data output from the prediction device. FIG. 6 is a flowchart illustrating an example of a detailed processing procedure of the prediction process by the prediction unit in step S303. FIG. 7 is an explanatory diagram illustrating an example of the second output data output by the information output unit. FIG. 8 is a block diagram illustrating a second example of a functional configuration of the prediction device. FIG. 9 is a flowchart illustrating a second example of a prediction process procedure performed by the prediction device. FIG. 10 is an explanatory diagram showing an example of a method for determining the number of shift days by the correlation calculation unit. FIG. 11 is a graph showing the effect of a user's selection of candidates for shifting. FIG. 12 is an explanatory diagram showing an example of presentation of explanatory variables to be excluded (or shifted).

Example 1 shows an example of application to prediction of product shipment volume at a logistics warehouse. In Example 1, an example is described in which the contribution level is output as the importance of the explanatory variables used in the prediction, and if the contribution level of any explanatory variable is lower than a predetermined value throughout the actual period and the prediction period, that explanatory variable is output as an exception target. Here, it is assumed that a user of the prediction device periodically predicts shipment volume using the prediction device at any frequency (for example, once a day or once a week). The configuration and processing flow of the prediction device are described below.

<Example of hardware configuration of prediction device>
FIG. 1 is a block diagram showing an example of a hardware configuration of a prediction device. The prediction device 100 includes a processor 101, a storage device 102, an input device 103, an output device 104, and a communication interface (communication IF) 105. The processor 101, the storage device 102, the input device 103, the output device 104, and the communication IF 105 are connected by a bus 106. The processor 101 controls the prediction device 100. The storage device 102 is a working area for the processor 101. The storage device 102 is a non-transient or temporary recording medium that stores various programs and data. Examples of the storage device 102 include a ROM (Read Only Memory), a RAM (Random Access Memory), a HDD (Hard Disk Drive), and a flash memory. The input device 103 is a device that inputs data. The input device 103 may be, for example, a keyboard, a mouse, a touch panel, a numeric keypad, or a scanner. The output device 104 is a device that outputs data. The output device 104 may be, for example, a display or a printer. The communication IF 105 is connected to a network 110 and transmits and receives data.

Connected to the network 110 is, for example, a warehouse management system (WMS) 120, from which shipping records of the product to be predicted can be obtained. However, the prediction device 100 does not necessarily need to be connected to the network 110, as long as the necessary data can be input. The prediction device 100 may also be implemented, for example, as an application on the WMS.

<Functional configuration example 1 of prediction device 100>
2 is a block diagram showing a first functional configuration example of the prediction device 100. The actual value 221 and the explanatory variable 222 are stored in the storage device 102. In addition, the predicted value 231, the prediction accuracy 232, the contribution degree 233, and the excluded target 234 can be displayed on a display, which is an example of the output device 104. Note that the prediction target is one type of data. In this case, for example, all products handled in a warehouse may be collectively treated as one prediction target, or only a specific product may be treated as a prediction target.

Here, the predicted value 231, the prediction accuracy 232, the contribution 233, and the excluded targets 234 do not necessarily need to be displayed on the output device 104. For example, if the predicted value 231, the prediction accuracy 232, the contribution 233, and the excluded targets 234 are transmitted as data to the WMS 120 via the network 110, the WMS 120 can use this data to plan work and personnel allocation within the warehouse. This can improve the efficiency of work within the warehouse.

The prediction device 100 has an actual value input unit 211, an explanatory variable input unit 212, a prediction unit 213, a predicted value output unit 214, and an information output unit 215. Specifically, these are functions that are realized by, for example, having the processor 101 execute a program stored in the storage device 102 shown in FIG. 1.

<Example of prediction processing procedure>
3 is a flowchart showing a first example of a prediction process procedure by the prediction device 100. When the prediction device 100 starts a prediction process, the prediction device 100 imports the actual value 221 through the actual value input unit 211 (step S301). Next, the prediction device 100 imports the explanatory variable 222 used for prediction through the explanatory variable input unit 212 (step S302). Details of the actual value 221 and the explanatory variable 222 will be described later with reference to FIG. 4.

Next, the prediction device 100 uses the prediction unit 213 to predict the shipment volume of the prediction target using the actual value 221 and the explanatory variables 222, and calculates the predicted value 231 (step S303). Here, the prediction unit 213 predicts the shipment volume, but this is not necessarily limited to the shipment volume. In addition, the prediction device 100 uses the prediction unit 213 to calculate the prediction accuracy 232 and the contribution 233 of each explanatory variable 222.

Next, the prediction device 100 causes the prediction value output unit 214 to display the prediction value 231 and prediction accuracy 232 calculated by the prediction unit 213 on a display, which is an example of the output device 104 (step S304).

Next, the prediction device 100 outputs the contribution 233 of each explanatory variable 222 calculated by the prediction unit 213 through the information output unit 215. Furthermore, the prediction device 100 refers to the contribution 233 of all explanatory variables 222 through the information output unit 215, and if the contribution 233 of each explanatory variable 222 is 0 or very close to 0 on any date in the prediction period, it regards that explanatory variable 222 as an exclusion target 234 and outputs that the explanatory variable 222 is an exclusion target (step S305).

<Example of input data>
4 is an explanatory diagram showing an example of input data to the prediction device 100. The input data 400 has a performance value (objective variable) 221 and explanatory variables 222 as time-series data. As an example, the variable name 401 of the performance value 221 is the daily shipping amount of a certain product in a warehouse, and the variable name 401 of the explanatory variable 222 is the business days and weather of a shipping destination store. In addition, the dates 402 are 1/1 to 2/4. The performance period is from 1/1 to 1/30, and the prediction period is from 1/31 to 2/4. The current date is 1/31.

Here, as an example, the prediction device 100 uses the actual values 221 of the actual period (from 1/1 to 1/30) to perform a prediction for the prediction period (from 1/31 to 2/4) on 1/31. Note that the explanatory variables 222 are not limited to business days and weather. In addition, any number of explanatory variables 222 may be used. Here, the explanatory variables 222 are set for both the actual period and the prediction period. Note that in FIG. 4, a date 402 in one-day units is used as an example of a period, but it is not limited to the date 402, and it may be a predetermined time in a day or multiple consecutive dates.

<An example of first output data>
FIG. 5 is an explanatory diagram showing an example of the first output data output from the prediction device 100. The first output data 500 is the predicted value 231 and the prediction accuracy 232 output from the predicted value output unit 214. In FIG. 5 and FIG. 7 described later, the actual value 221, the predicted value 231, and the prediction accuracy 232 from 1/1 to 1/28 are omitted for the sake of explanation. The prediction device 100 outputs the predicted value 231 for both the actual period and the prediction period (from 1/1 to 2/4). The prediction device 100 outputs the prediction accuracy 232 by the prediction unit 213 based on the deviation between the actual value 221 and the predicted value 231 for each date in the actual period. The prediction accuracy 232 may be evaluated by any known method.

In the example of FIG. 5, the prediction unit 213 calculates the prediction accuracy and the prediction error according to the following equations (1) and (2).
Prediction accuracy = 1 - prediction error (1)
Forecast error = | Forecast value - Actual value | / Actual value ... (2)
Here, the process (step S303) executed by the prediction unit 213 will be described with reference to FIG.

<Prediction process (step S303)>
6 is a flowchart showing a detailed example of the process procedure of the prediction process by the prediction unit 213 in step S303. The prediction device 100 calculates the predicted value 231 by the prediction unit 213 by setting the contribution degree 233 of each explanatory variable 222 for each date of the performance period so that the daily prediction accuracy 232 is maximized (step S601). The contribution degree 233 is the degree of influence on the predicted value 231 of the performance period.

Next, the prediction device 100 sets the contribution degree 233 determined in step S601 based on the explanatory variables 222 set for each date in the prediction period by the prediction unit 213, and calculates the prediction value 231. Next, the prediction device 100 outputs the prediction value 231, the contribution degree 233, and the prediction accuracy 232 for the actual period and the prediction period by the prediction unit 213.

The processes performed by the prediction unit 213 in steps S601 and S602 may be any prediction method that uses the explanatory variables 222 input by the explanatory variable input unit 212. For example, the prediction device 100 generates a prediction model by applying a combination of the actual values 221 and the explanatory variables 222 as a learning data set to autoregression analysis, multiple regression analysis, or deep learning. The prediction device 100 may then calculate a predicted value 231 of the shipment volume in the prediction period by providing the explanatory variables 222 for the prediction period to the prediction model.

The predicted value P is calculated by combining a prediction result P1 calculated from an arbitrary prediction method 1 (e.g., autoregression) with a prediction result P2 calculated using a prediction method 2 different from prediction method 1 (e.g., linear regression as shown in formula (4) below) using formula (3) below. Here, α and β are weighting coefficients (where α + β = 1).

P=α×P1+β×P2...(3)

For example, in step S601, the prediction unit 213 applies combinations of α and β (including β1 and β2 when applying the following formula (4)) in a brute-force manner, and calculates the contribution degree 233 so as to maximize the daily prediction accuracy 232 using the above formulas (1) and (2). Here, the contribution degree 233 of the prediction result P1 is α at which the prediction accuracy 232 is maximized, and the contribution degree 233 of the prediction result P2 is β at which the prediction accuracy 232 is maximized.

Furthermore, it is sufficient that one or both of the prediction results P1 and P2 are prediction results using the explanatory variable 222. For example, if only the prediction result P2 is a prediction result using the explanatory variable 222, and specifically, for example, P21 is a prediction value using the business days shown in FIG. 4, and P22 is a prediction value using the weather shown in FIG. 4, the prediction result P2 can be calculated by the following formula (4).

P2=β1×P21+β2×P22...(4)

Here, β1 and β2 are weighting coefficients, β1 + β2 = 1. In this case, the prediction unit 213 can calculate the contribution 233 of each business day and the contribution 233 of the weather as β × β1 and β × β2, respectively.

<An example of second output data>
Fig. 7 is an explanatory diagram showing an example of the second output data output by the information output unit 215. The second output data 700 is data showing the contribution degree 233 and the excluded target 234 output by the information output unit 215. In Fig. 7, as an example, the daily contribution degree 233 is displayed in percentage using a table 701 and a graph 702, and the "weather" of the explanatory variable 222, which is the excluded target 234, is displayed using a notification column 703. The prediction method 1 of the table 701 and the graph 702 is P1 in the above formula (3).

Looking at table 701, it can be seen that the contribution of weather 233 is 0% for both the forecast period and the actual period, and in this case, it is clear that the weather is unrelated to the forecasted shipment volume 231 (does not affect the forecast result). This applies, for example, to everyday items that are shipped regularly without being affected by the weather. Therefore, the information output unit 215 regards the weather as an exclusion target 234 and notifies the user by, for example, displaying the words "Weather is a candidate for exclusion" in the notification field 703.

Here, the method of notifying the excluded items 234 may be arbitrary. For example, in table 701 of FIG. 7, the information output unit 215 may display the weather row in a color different from the colors of the other rows.

The criteria by which the information output unit 215 identifies the exclusion target 234 are not necessarily limited to the contribution 233 being 0% for all dates in the forecast period and the actual period. For example, a method can be considered in which a threshold value (a value greater than 0) is set in advance, and any date for which the contribution 233 is less than the threshold value is considered to be an exclusion target. The information output unit 215 may also consider, for example, an explanatory variable 222 whose contribution 233 is greater than 0 only for a few dates (any period) in the first half of the actual period, and whose contribution 233 is 0 in the actual period and forecast period thereafter, to be an exclusion target 234.

According to Example 1, when the prediction device 100 periodically predicts shipment volume, the user can determine that the explanatory variable 222 specified as the excluded variable 234 is not contributing to the prediction result by looking at the explanatory variable 222. The prediction device 100 makes predictions from the next time onwards without using the excluded explanatory variable 222, so that the user can reduce the effort (man-hours) required to obtain and manage the data of the explanatory variable 222 when making predictions from the next time onwards, and the effort (man-hours) required to input data (explanatory variable 222) when using the prediction device 100. In addition, since the prediction device 100 can make predictions with a small number of explanatory variables 222, the time required for the prediction unit 213 to make predictions can be reduced.

Here, as a usage form of the prediction device 100, for example, when the prediction device 100 performs a first prediction, it may perform a prediction using a prediction method that is not very accurate but has a short processing time to obtain the excluded target 234, and then perform a prediction using a prediction method that is more accurate than the first prediction but has a long processing time, using new explanatory variables 222 that exclude the excluded target 234. This makes it possible to reduce the prediction time required for the first prediction.

Alternatively, for example, the exclusion targets 234 for each product identified using the prediction device 100 at one warehouse may be stored in a database that is accessible to another warehouse of the logistics company. Then, when making predictions using the prediction device 100 at another warehouse, the prediction device 100 may refer to the exclusion targets 234 for the same product or similar products (e.g., the same product from a different manufacturer) stored in the database.

Furthermore, when the explanatory variable input unit 212 performs the process of importing the explanatory variables 222 in step S302 shown in FIG. 3, the prediction device 100 may inform the user of the excluded targets 234 (or explanatory variables 222 that affect the prediction other than the excluded targets 234) via the output device 104. This makes it possible to make predictions using appropriate explanatory variables 222 even when it is not known which explanatory variables 222 will affect the prediction results in another warehouse.

The explanatory variable input unit 212 may also store the explanatory variable 222 designated as the exclusion target 234, and when the user inputs this exclusion target 234 to the prediction device 100 when the prediction device 100 makes the next prediction, notify the output device 104 that the explanatory variable 222 was the exclusion target 234 in the previous prediction. This can prevent the user from inputting an explanatory variable 222 that was the exclusion target in the previous prediction.

Next, Example 2 will be described. Except for the differences described below, each part of the prediction device 100 in Example 2 has the same function as each part in Example 1 with the same reference numerals shown in Figures 1 to 7, and therefore the description of those parts will be omitted. Also, in Example 2, in addition to Example 1, an example will be described that has a process for appropriately evaluating how many days after the occurrence of the variable factor the influence of the variable factor appears in the shipping volume when any of the explanatory variables 222 is a variable factor for the prediction result. Also, in Example 2, the prediction target is one type of data. In this case, for example, all products handled in a warehouse may be collectively treated as one prediction target, or only a specific product may be treated as a prediction target.

<Functional configuration example 2 of prediction device 100>
8 is a block diagram showing a second functional configuration example of the prediction device 100. In the second embodiment, in addition to the first embodiment, a shift target 801 is added to the input data, and a shift number of days 802 is added to the second output data 700. In addition, a correlation calculation unit 800 is added as a program module.

<Example of prediction processing procedure>
9 is a flowchart showing a second example of a prediction process procedure by the prediction device 100. Only steps that differ from the first example (FIG. 3) will be described. After step S301, the prediction device 100 inputs the explanatory variables 222 and the shift target 801 by the explanatory variable input unit 212 (step S901). Next, the prediction device 100 calculates the correlation coefficient between the actual value 221 and each explanatory variable 222 of the shift target 801 by the correlation calculation unit 800, and determines the number of shift days 802 for the explanatory variables 222 designated as the shift target (step S902).

Next, the prediction device 100 performs prediction using the actual value 221 and the explanatory variable 222 by the prediction unit 213, and calculates the predicted value 231, the prediction accuracy 232, and the contribution degree 233 (step S903). Here, for the explanatory variable 222, the explanatory variable 222 specified as the shift target 801 is used for prediction after shifting the number of days by the shift number of days 802 calculated by the correlation calculation unit 800.

Next, the prediction device 100 outputs the contribution 233 of each explanatory variable 222, the number of shift days 802, and the excluded target 234 by the information output unit 215 after step S304. Here, a method of specifying the explanatory variable 222 of the shift target 801 will be described with reference to FIG. 4. For example, the information output unit 215 specifies the explanatory variable 222 of the shift target 801 in a form that can be processed by the prediction device 100 in the specification field of the variable name 401 of the explanatory variable 222. For example, the information output unit 215 may add a symbol (such as @) to the end of the variable name 401 of the explanatory variable 222 of the shift target 801 (not shown).

Next, a method for outputting the number of shift days 802 by the information output unit 215 will be described with reference to FIG. 7. Any output method can be used, but for example, a column for the number of shift days 802 can be added to the right end of the table 701 in FIG. 7 to display the number of shift days 802 for each explanatory variable 222 (not shown).

<How to determine the number of shift days 802>
Fig. 10 is an explanatory diagram showing an example of a method for determining the number of shift days 802 by the correlation calculation unit 800. The contents of Fig. 10 can be displayed on a display. As an example, a case where a commercial (CM) is aired on television or the like as a factor for fluctuations in the shipment volume of a certain product will be described. In each of the graphs (a) to (c) of Fig. 10, the horizontal axis indicates the date 402, the left direction indicates the past, and the right direction indicates the future. In addition, the vertical axis in graph (a) indicates the shipment record, and the vertical axis in graphs (b) and (c) indicates the number of times a commercial was aired.

In Figure 10, (a) shows actual shipments, and (b) shows the time series (daily series) data for the days the commercial aired. Looking at these, we can see that the actual shipments rose sharply (pulse-like) on the +1 day after the commercial aired.

In FIG. 10, (c) is a diagram explaining how the explanatory variable 222 of the commercial airing 1001 is shifted in the correlation calculation unit 800. The correlation calculation unit 800 shifts the time series data of the commercial airing by one day at a time from the number of days 0, obtains a correlation with the shipping record, and, for example, finds the number of days of shift with the highest correlation coefficient.

In the example of FIG. 10(c), the correlation calculation unit 800 determines that the explanatory variable 222 of commercial airing becomes the highest correlation coefficient with the time series data of shipping results when the explanatory variable 222 of commercial airing is shifted by +1 day. In other words, the explanatory variable 222 of commercial airing is determined to have the highest correlation with shipping results one day (+1 day) after the commercial airing date (explanatory variable), and the correlation calculation unit 800 determines the number of days to shift 802 of the explanatory variable 222 of commercial airing to be +1 day. In this way, the correlation calculation unit 800 compares each of the explanatory variables 222 specified as the shift target 801 with the shipping results one by one to evaluate the correlation, calculates the number of days shift between the date of shipping results where the correlation is maximum and the date of the explanatory variable event, and determines the number of days to shift 802 for each explanatory variable 222.

For the sake of explanation, FIG. 10 shows a simple example of the number of shift days 802 that is easy for humans to understand. However, in an actual logistics warehouse, the actual shipping volume is constantly fluctuating daily due to various (multiple) variable factors (explanatory variables 222), so there are only a limited number of examples where the correlation (number of shift days 802) between each explanatory variable 222 and actual shipping volume as shown in FIG. 10 can be determined at a glance.

By performing the shift number of days evaluation method of the second embodiment, even if the shift number of days 802 cannot be determined at a glance, it is possible to appropriately take into account the number of days on which the influence of each explanatory variable 222 (variation factor) appears in the shipping performance and shipping forecast. Here, the shift evaluation period 1002 may be any number of days. In FIG. 10, an example is shown in which the date is shifted only in the positive direction (future) of the time direction of the date 402, but it may be shifted in the negative direction (past). Note that when determining the number of days of shift with the highest correlation coefficient, it is unlikely that the maximum value will occur, for example, about one month (+ about 30 days) from the event date of the explanatory variable 222, and it is reasonable to consider that the period in which a high correlation is obtained is somewhat limited. In addition, when the explanatory variable is a CM's viewing rate or number of views, it is also possible to set a threshold value for the correlation based on the actual performance. Therefore, if the correlation exceeds a threshold, it is determined that this is the maximum correlation, and the shift amount can be calculated by using the date closest to the event, or a maximum shift amount can be set for each explanatory variable, and the shift amount can be calculated by using the date within that range that results in the maximum correlation.

One variable that should be shifted in the negative direction is weather forecast data. For example, shipments of snow removal equipment and boots increase several days before snow is forecast, so the number of days to shift 802 can be calculated by negatively correlating actual snowy weather data with actual shipments. Here, the method of outputting the exclusion targets 234 in Example 2 will be explained using FIG. 7 again.

In the second embodiment, the prediction device 100 identifies the exclusion target 234 based on the same criteria as in the first embodiment (such as when the contribution level 233 is 0 or below a threshold). However, the notification content in the notification field 701 shown in FIG. 7 differs depending on whether the user has designated the exclusion target 234 as a shift target 801.

First, if the explanatory variable 222 has not been designated as a shift target 801, there is a possibility that the prediction result will be affected if the explanatory variable 222 is designated as a shift target 801. Therefore, the information output unit 215 displays, for example, "Explanatory variable z is a candidate for shift target or exclusion" in the notification field 703, and prompts the user to designate it as a shift target 801 or to exclude it from the next prediction onwards (z is an explanatory variable 222). Next, if the explanatory variable 222 has been designated as a shift target 801, there is a possibility that the prediction result will be affected. Therefore, the information output unit 215 displays, for example, "Explanatory variable z is a candidate for exclusion" in the notification field 703, and prompts the user to exclude it from the explanatory variables 222 from the next prediction onwards.

By the above method, the prediction device 100 can perform highly accurate predictions even in a situation where it is unknown how many days after the influence of the variable factor will appear in the predicted value. In addition, the information output unit 215 displays the explanatory variables 222 whose contribution degree 233 is lower than a predetermined value on the output device 104. Therefore, when making predictions next time or later, the user can perform highly accurate predictions with fewer explanatory variables 222 by excluding the explanatory variables 222 or changing the settings of the shift target. In addition, as in Example 1, when making predictions next time or later, the effort (man-hours) for obtaining and managing data of the explanatory variables 222 that have been excluded can be reduced, as well as the effort (man-hours) for inputting data (explanatory variables 222) when using the prediction device 100.

<Effect of selecting candidates for shifting>
Fig. 11 is a graph showing the effect of a user's selection of candidates for shifting. Graph 1100 in Fig. 11 can be displayed on a display. Graph 1100 is an example assuming a case where shipment forecasting is performed for seasonal products whose shipping performance gradually decreases from the first shipment of each year. In this case, for example, three explanatory variables 222 are used for the forecast: the first week W1, the second week W2, and the third week W3 of the shipment. The user can select the candidates for shifting from a selection screen (not shown) for selecting the candidates for shifting.

In this example, when the correlation calculation unit 800 calculates the number of shift days 802 for each explanatory variable 222 (W1 to W3) using step S902 in FIG. 9, depending on how the correlation is calculated, all explanatory variables 222 for W1 to W3 may be deemed to have the highest correlation with the shipping record for the first week (2/4 to 2/10), making it impossible to perform an appropriate prediction. Therefore, if the user does not specify each explanatory variable 222 for W1 to W3 as a shift target 801, the prediction device 100 can perform a prediction that appropriately takes into account the influence of W1 to W3.

Next, Example 3 will be described. In Examples 1 and 2, the case where there is one prediction target has been described. However, since a wide variety of products are handled in a logistics warehouse, there are cases where predictions need to be made separately for each product type. For example, in the case of cosmetics, the shipping trends differ between cosmetics destined for high-end stores (specialty stores, department stores, etc.) and cosmetics destined for mass retailers. In Example 3, an example where multiple types of data are the prediction target will be described. Note that, except for the differences described below, the same components as in Examples 1 and 2 are given the same reference numerals and their description will be omitted.

The prediction process of Example 3 will be explained using FIG. 9. Note that in the following explanation, an example will be described in which the process of shifting the date of explanatory variable 222 described in Example 2 is performed, but Example 3 can also be implemented in the case where explanatory variable 222 is not shifted as in Example 1.

When the prediction device 100 starts processing, the actual value input unit 211 imports actual values 221 of multiple prediction targets (step S301). Next, the prediction device 100 imports explanatory variables 222 and shift targets 801 via the explanatory variable input unit 212 (step S901). Here, the explanatory variable input unit 212 may specify the shift target 801 for each actual value 221 imported in step S301.

Next, the prediction device 100 uses the correlation calculation unit 800 to find the correlation between each actual value 221 and the explanatory variable 222 of the shift target 801, and determines the number of shift days 802 for each actual value 221 (step S902). Next, the prediction device 100 uses the prediction unit 213 to make a prediction using each explanatory variable 222 shifted by the number of shift days 802 for each actual value 221 determined in step S902 for each prediction target, and calculates the predicted value 231, prediction accuracy 232, and contribution 233 of each explanatory variable 222 to each prediction target (step S903).

Next, the prediction device 100 outputs the predicted value 231 and prediction accuracy 232 of each prediction target through the predicted value output unit 214 (step S304). Finally, the prediction device 100 outputs the contribution 233 of each explanatory variable 222 to each prediction target, the number of shift days 802, and the excluded targets 234 through the information output unit 215 (step S904).

Here, the prediction value output unit 214 may output the prediction value 231 and prediction accuracy 232 of each prediction target in a separate table for each prediction target, for example, in the manner shown in FIG. 5. In addition, the information output unit 215 may display the contribution 233 of each explanatory variable 222, the number of shift days 802, and the excluded targets 234 for each prediction target, for example, as second output data 700 as shown in FIG. 7.

According to the third embodiment, the prediction device 100 can make predictions for each product category, and the user can understand which explanatory variables 222 contribute to the prediction results for each product category. As a result, when using the prediction results for personnel planning in a logistics warehouse, for example, the user can appropriately set the personnel to be allocated for each product category. Furthermore, when making predictions from the next time onwards, the prediction device 100 can make predictions excluding unnecessary explanatory variables 222, thereby reducing the user's effort in obtaining, managing, and inputting the explanatory variables 222 into the prediction device 100.

In the first to third embodiments, an example is described in which the explanatory variables 222 of the exclusion target 234 (or the shift target 801) are presented to the user by outputting the contribution 233 as the importance of the explanatory variables 222. In the fourth embodiment, an example is described in which the prediction device 100 adds and outputs other information to improve prediction accuracy.

As mentioned above, logistics warehouses have a limited number of staff that can be secured, and so they want to avoid an event where the predicted value falls short of the actual value (this is called an "undersaturation"). When undersaturation occurs in staffing, there is a shortage of personnel equal to the predicted value minus the actual value. In addition, it is necessary to make the daily prediction error 321 as small as possible (to make the prediction accuracy 232 as high as possible). To achieve this, it is necessary to use an appropriate prediction method and to take into account the necessary explanatory variables 222 without deficiencies.

Therefore, in addition to the contribution 233 of each explanatory variable 222 described in Examples 1 to 3, the prediction device 100 displays on the display whether or not there is a downward deviation in the prediction results for the performance period, and also displays days on which the daily prediction error 321 for the performance period is particularly large (for example, days above a threshold). This makes it possible to present to the user the validity of the prediction method used for the prediction and the possibility of a shortage of explanatory variables 222.

Figure 12 is an explanatory diagram showing an example of the presentation of explanatory variables 222 for the exclusion target 234 (or the shift target 801). In Figure 12, (a) shows the predicted value 231 and actual value 221 of the shipment volume during the actual period (1/1 to 1/7). (b) shows the explanatory variable 222 used in the prediction (here, commercial airing 1201 (times/day)). Here, the prediction device 100 performs prediction using an arbitrary prediction method 1 and a prediction method 2 that uses commercial airing 1201 as the explanatory variable 222.

The table in (c) shows, from the top row, the date 402, the prediction error 1202, the contribution 233 of prediction method 1, and the contribution 233 of commercial airing 1201, which is the explanatory variable 222 of prediction method 2. Here, on a day when the prediction result falls short, "Falling short" is displayed in the row of the prediction error 1202. Furthermore, if the daily prediction error 1202 exceeds a threshold (prediction error threshold 1203. In the example of FIG. 12, it is evaluated using equation (2) and set to 0.5. However, the value of the prediction error threshold 1203 is arbitrary), the prediction device 100 changes the color of the table in (c) when displaying it.

(c) shows that the contribution 233 of commercial airing 1201 is low throughout the prediction period, and has almost no impact on the prediction results. On the other hand, it can be seen that the days when the downside occurred (1/2 to 1/4) or the days when the prediction error 1202 was equal to or greater than the prediction error threshold 1203 account for more than half of the actual period. In such a case, as shown in (c), if the contribution 233 of prediction method 1 is high and the contribution 233 of commercial airing 1201, which is the explanatory variable 222 of prediction method 2, is both lower than a specified value, it is possible that either or both of the prediction methods 1 and/or commercial airing 1201 used in the prediction are inappropriate.

Here, for example, the information output unit 215 displays in the notification field 703 the total number of days during the performance period when a downturn occurred or when the forecast error 1202 was equal to or greater than the forecast error threshold 1203. If this total accounts for a predetermined number of days (e.g., half) or more of the days in the performance period, a notification is displayed (not shown) encouraging the user to change the forecast method, add or change the explanatory variables 222 used in the forecast, or change the shift target 801 of each explanatory variable 222. This allows the user to become aware of the need to change the forecast method, as well as the need to change the explanatory variables 222 used in the forecast and the shift target 801.

The prediction device 100 may also re-predict the shipment volume for the day on which the forecast fell short, and set the contribution 233 so that the prediction error is minimized (the prediction accuracy is maximized). If the contribution 233 has fallen by a predetermined value or more before and after the re-prediction, the prediction device 100 may notify the user that a negative impact has occurred by highlighting the prediction method 1 or explanatory variable 222.

Note that while underperformance has been described in Example 4, the same applies to events in which the predicted value exceeds the actual value (called an "overperformance"). That is, in addition to the contribution 233 of each explanatory variable 222 described in Examples 1 to 3, the prediction device 100 displays on the display whether or not there is an overperformance in the prediction results for the actual period, and also displays days in the actual period where the daily prediction error 321 is particularly large (above a specific error). This makes it possible to present to the user the validity of the prediction method used for the prediction, and the possibility of a shortage of explanatory variables 222.

Furthermore, the embodiments to the fourth embodiment can be applied not only to logistics sites but also to other sites. For example, even in production sites, it is required to predict demand with as few explanatory variables 222 as possible due to reasons such as the difficulty in obtaining consumer data. If the embodiments to the fourth embodiment are applied to a production site, the prediction device 100 can perform highly accurate predictions by excluding unnecessary explanatory variables 222 when making predictions from the second time onwards, and it is possible to reduce the effort required for users to manage and obtain the explanatory variables 222.

The present invention is not limited to the above-described embodiments, but includes various modified examples and equivalent configurations within the spirit of the appended claims. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to having all of the configurations described. Furthermore, a portion of the configuration of one embodiment may be replaced with the configuration of another embodiment. Furthermore, the configuration of another embodiment may be added to the configuration of one embodiment. Furthermore, other configurations may be added, deleted, or replaced with part of the configuration of each embodiment.

Furthermore, each of the configurations, functions, processing units, processing means, etc. described above may be realized in part or in whole in hardware, for example by designing them as integrated circuits, or may be realized in software by having the processor 101 interpret and execute a program that realizes each function.

Information such as programs, tables, and files that realize each function can be stored in a storage device such as a memory, hard disk, or SSD (Solid State Drive), or in a recording medium such as an IC (Integrated Circuit) card, SD card, or DVD (Digital Versatile Disc).

In addition, the control lines and information lines shown are those considered necessary for explanation, and do not necessarily represent all control lines and information lines necessary for implementation. In reality, it is safe to assume that almost all components are interconnected.

100 Prediction device 101 Processor 102 Storage device 211 Actual value input unit 212 Explanatory variable input unit 213 Prediction unit 214 Prediction value output unit 215 Information output unit 221 Actual value 222 Explanatory variable 231 Prediction value 232 Prediction accuracy 233 Contribution degree 234 Excluded object 321 Prediction error 800 Correlation calculation unit 1202 Prediction error 1203 Prediction error threshold

Claims (10)

A prediction device that predicts a predicted value of a prediction target based on an actual value of the prediction target and time series data of each of a plurality of explanatory variables, comprising:
A processor for executing a program and a storage device for storing the program,
The processor,
Calculating prediction accuracy based on the actual value and the predicted value;
calculating a first contribution degree, which is a degree of influence that the first explanatory variable has on the predicted value, and a second contribution degree, which is a degree of influence that the second explanatory variable has on the predicted value, based on a first predicted value predicted using time-series data of a first explanatory variable among the plurality of explanatory variables, a second predicted value predicted using time-series data of a second explanatory variable among the plurality of explanatory variables different from the first explanatory variable, and the prediction accuracy;
determining an explanatory variable to be excluded from the plurality of explanatory variables based on the first contribution degree and the second contribution degree;
Outputting the excluded object;
re-predicting the first predicted value using time-series data of the first explanatory variable so as to maximize the prediction accuracy at a time point when an event occurs in which the first predicted value falls below the actual value;
re-predicting the second predicted value using time-series data of the second explanatory variable so as to maximize the prediction accuracy at a time point when an event occurs in which the second predicted value falls below the actual value;
calculating the first contribution and the second contribution based on the prediction accuracy, the re-predicted first predicted value, and the re-predicted second predicted value;
A prediction device comprising: A prediction device that predicts a predicted value of a prediction target based on an actual value of the prediction target and time series data of each of a plurality of explanatory variables, comprising:
A processor for executing a program and a storage device for storing the program,
The processor,
Calculating prediction accuracy based on the actual value and the predicted value;
calculating a first contribution degree, which is a degree of influence that the first explanatory variable has on the predicted value, and a second contribution degree, which is a degree of influence that the second explanatory variable has on the predicted value, based on a first predicted value predicted using time-series data of a first explanatory variable among the plurality of explanatory variables, a second predicted value predicted using time-series data of a second explanatory variable among the plurality of explanatory variables different from the first explanatory variable, and the prediction accuracy;
determining an explanatory variable to be excluded from the plurality of explanatory variables based on the first contribution degree and the second contribution degree;
predicting a predicted value of the prediction target using time series data of remaining explanatory variables excluding the excluded variable from among the plurality of explanatory variables;
re-predicting the first predicted value using time-series data of the first explanatory variable so as to maximize the prediction accuracy at a time point when an event occurs in which the first predicted value falls below the actual value;
re-predicting the second predicted value using time-series data of the second explanatory variable so as to maximize the prediction accuracy at a time point when an event occurs in which the second predicted value falls below the actual value;
calculating the first contribution and the second contribution based on the prediction accuracy, the re-predicted first predicted value, and the re-predicted second predicted value;
A prediction device comprising: The prediction device according to claim 2,
The processor,
Calculating the first predicted value and the second predicted value by a first prediction method;
predicting a predicted value of the prediction target using time-series data of the remaining explanatory variables by a second prediction method that has a longer processing time than the first prediction method but has higher prediction accuracy than the first prediction method;
A prediction device comprising: The prediction device according to claim 1 ,
The processor,
The time series data of the first explanatory variable is designated as a shift target;
determining a shift direction and a shift amount of the time series data of the first explanatory variable based on a correlation between the performance value and the time series data of the first explanatory variable;
calculating the first contribution degree and the second contribution degree after the shift based on a first predicted value after the shift, which is predicted using the time series data of the first explanatory variable after shifting the time series data of the first explanatory variable by the shift amount in the shift direction, and a second predicted value after the shift, which is predicted using the time series data of the second explanatory variable;
determining explanatory variables to be excluded based on the first contribution degree and the second contribution degree after the shift;
A prediction device comprising: The prediction device according to claim 4,
The processor,
determining the shift direction and the shift amount that maximize the correlation;
A prediction device comprising: The prediction device according to claim 1 ,
The processor,
Calculating the contribution of the first explanatory variable and the contribution of the second explanatory variable for each prediction target;
determining explanatory variables for the excluded objects for each of the prediction objects;
outputting the excluded object for each of the predicted objects;
A prediction device comprising: The prediction device according to claim 1 ,
The processor,
outputting a period during which the difference between the actual value and the predicted value falls outside a first allowable range;
A prediction device comprising: The prediction device according to claim 1 ,
The processor,
outputting a period during which the prediction accuracy falls outside a second acceptable range;
A prediction device comprising: A prediction method executed by a prediction device that predicts a predicted value of a prediction target based on an actual value of the prediction target and time-series data of each of a plurality of explanatory variables, comprising:
A processor for executing a program and a storage device for storing the program,
The prediction method includes:
The processor,
Calculating prediction accuracy based on the actual value and the predicted value;
calculating a first contribution degree, which is a degree of influence that the first explanatory variable has on the predicted value, and a second contribution degree, which is a degree of influence that the second explanatory variable has on the predicted value, based on a first predicted value predicted using time-series data of a first explanatory variable among the plurality of explanatory variables, a second predicted value predicted using time-series data of a second explanatory variable among the plurality of explanatory variables different from the first explanatory variable, and the prediction accuracy;
determining an explanatory variable to be excluded from the plurality of explanatory variables based on the first contribution degree and the second contribution degree;
Outputting the excluded object;
re-predicting the first predicted value using time-series data of the first explanatory variable so as to maximize the prediction accuracy at a time point when an event occurs in which the first predicted value falls below the actual value;
re-predicting the second predicted value using time-series data of the second explanatory variable so as to maximize the prediction accuracy at a time point when an event occurs in which the second predicted value falls below the actual value;
calculating the first contribution and the second contribution based on the prediction accuracy, the re-predicted first predicted value, and the re-predicted second predicted value;
A prediction method comprising: A prediction method executed by a prediction device that predicts a predicted value of a prediction target based on an actual value of the prediction target and time-series data of each of a plurality of explanatory variables, comprising:
A processor for executing a program and a storage device for storing the program,
The processor,
Calculating prediction accuracy based on the actual value and the predicted value;
calculating a first contribution degree, which is a degree of influence that the first explanatory variable has on the predicted value, and a second contribution degree, which is a degree of influence that the second explanatory variable has on the predicted value, based on a first predicted value predicted using time-series data of a first explanatory variable among the plurality of explanatory variables, a second predicted value predicted using time-series data of a second explanatory variable among the plurality of explanatory variables different from the first explanatory variable, and the prediction accuracy;
determining an explanatory variable to be excluded from the plurality of explanatory variables based on the first contribution degree and the second contribution degree;
predicting a predicted value of the prediction target using time series data of remaining explanatory variables excluding the excluded variable from among the plurality of explanatory variables;
re-predicting the first predicted value using time-series data of the first explanatory variable so as to maximize the prediction accuracy at a time point when an event occurs in which the first predicted value falls below the actual value;
re-predicting the second predicted value using time-series data of the second explanatory variable so as to maximize the prediction accuracy at a time point when an event occurs in which the second predicted value falls below the actual value;
calculating the first contribution and the second contribution based on the prediction accuracy, the re-predicted first predicted value, and the re-predicted second predicted value;
A prediction method comprising:

Images