JPH09101947A - Time series forcasting method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve accuracy for forcasting the quantity of sale by reducing the predictive job load of a planner in planned job and making speedy and exact the extraction/correction work of a special value as the main factor of reduction in the accuracy of sales quantity prediction. SOLUTION: The respective variable factors of season fluctuation, tendency/ circulation fluctuation and irregular fluctuation are extracted from time series data concerning the sales quantity result of products, and the respective variable factors are simultaneously displayed on an output device as graphs. Then, the special values caused by any unexpected irregular fluctuation factors are fetched from an input device into the respective variable factors displayed as graphs and the time points, when those special values are generated, are specified. The various kinds of factor candidate information generated at the same time point as the time point of generation are outputted to the output device and based on the factor candidate information, the set generation factors and the corrected values of special values are fetched from the input device into the special values. When the generation factor of the special value is estimated within a predictive object period, the predictive sales quantity is calculated while considering the degree of influence of that generation factor in the past as well.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a demand forecasting method based on time-series data of sales volume of products, and particularly when it is suitable for estimating sales volume of highly seasonal products represented by air-conditioning products. A series prediction method.

[0002]

2. Description of the Related Art Sales volume forecasting is a task performed to obtain a base for making various plans such as sales plans.
The quality of a plan is often greatly affected by the accuracy of its prediction. Various mathematical statistics / forecasting techniques are applied to the sales volume prediction, and the data used at that time is often time-series data of actual sales volume. This is largely due to the fact that data can be easily obtained and that the future can be used as an extension of the past. In the prediction using time-series data, it is possible to improve the prediction accuracy by accurately grasping the four fluctuation factors of seasonal fluctuation, trend fluctuation, circulation fluctuation, and irregular fluctuation. To do so, first, for time series data that has been processed appropriately for irregular fluctuations (singular values) of time series data, seasonal fluctuations, trend fluctuations,
It is desirable to apply a prediction model that considers cyclic fluctuations. References (Kaya: “Knowledge of Prediction”, Morikita Publishing, 1977.7,
pp120), for singular values with clear causes, it is recommended to appropriately interpolate estimated values or make corrections before making prediction calculations.

[0003]

In the above prior art, there is no established technique for extracting / correcting the singular value in the sales record time series data, and the sales record time series data including the singular value is not available. Is often used for prediction calculations. Even if the singular value is extracted and corrected based on the experience and intuition of the planner, the singular value may be overlooked or an incorrect correction value may be set, which will affect the accuracy of sales volume prediction. There was a problem.

An object of the present invention is a time-series forecasting method that realizes highly accurate forecasting of sales volume by speeding up and optimizing the work of extracting / correcting singular values, which is a major cause of deterioration in the accuracy of sales volume forecasting. To provide.

[0005]

The above objects can be achieved by the following means (1) to (6). (1) An input device, an output device, a storage device including a storage region of information about a sales amount result of a product and a storage region of information about a factor result caused by a change in the sales amount of the product, and a sales amount in the storage device. In a time series forecasting method using a computer having a computing device for calculating the sales volume of a product in the forecast period input from the input device based on the information on the actual results, a time series original data regarding the sales volume results of the products From the data, seasonal fluctuations, trends / circulation fluctuations, and each fluctuation element of irregular fluctuations are extracted, and the time series transition of each fluctuation element and the time series transition of the original data are simultaneously displayed as a graph on an output device, The singular value in each time-series transition displayed in the graph is taken in from the input device, the time when the singular value occurs is specified, and various factor candidate information that occurs at the same time as the time when the singular value occurs is output. Series prediction method when and outputting the device.

(2) An input device, an output device, a storage device including a storage region for storing information on the actual sales volume of a product and a storage region for storing information on a factor actual result caused by fluctuations in the sales volume of the product, and in the storage device. In a time series forecasting method using a computer having a computing device for calculating the sales volume of a product in the forecast period input from the input device based on the information on the sales volume record of The generation factor set for the singular value in the original data and the correction value of the singular value are fetched from the input device,
The actual value at the time of occurrence of the singular value of the original data is replaced with the correction value to create corrected time series data, and the cause of occurrence of the singular value within the corrected time series data and the prediction target period. If it is expected that the sales amount of the product during the forecast period will be calculated based on the past degree of influence of the occurrence factor.

(3) A singular value set by the planner is accepted in a time series transition chart of original data and arbitrary transition charts of individual time series transition charts of seasonal fluctuations, trends / circulation fluctuations, and irregular fluctuations. (4) The cause of the singular value is set in an interactive manner with the planner based on various factor candidate information generated at the same time as the singular value occurs.

(5) When the set singular value is accepted in the time series transition chart of the original data and any one of the time series transition charts of the seasonal fluctuation, the trend / circulation fluctuation, and the irregular fluctuation , The display position of the set singular value is also highlighted in the remaining time series transition diagrams. (6) A value obtained by removing the irregular variable element of the actual value set as the singular value is used as the correction value of the singular value.

By providing the above means,
The effects as shown in (1) to (3) can be obtained.

(1) When the occurrence factor set for the singular value and the correction value of the singular value are fetched from the input device based on the factor candidate information, the degree of influence of the occurrence factor on the fluctuation of the sales amount is calculated. At the same time, the actual value of the singular value generation time of the original data is replaced with the above-mentioned correction value to create corrected time-series data composed of seasonal variation, tendency / circulation variation, and white noise. As a result, it is possible to eliminate a large cause of the decrease in the prediction accuracy, and it can be expected that the sales quantity prediction will be highly accurate.

(2) Since seasonal fluctuations, trends / circulation fluctuations, and irregular fluctuations extracted from time-series original data relating to the actual sales volume of products are displayed graphically on the same screen, Since the singular value can be easily recognized and the singular value can be set in any variable, the singular value can be set accurately and efficiently, and as a result, the load of the prediction work can be reduced.

(3) Since various factor candidate information generated at the same time as the occurrence time of the singular value, which was a major cause of the deterioration of the prediction accuracy, is comprehensively displayed, so that the causal relationship between the singular value and the occurrence factor thereof is displayed. Can be specified easily, and the allocation of the generation factor to the singular value can be optimized.

[0013]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described in detail below with reference to the drawings. FIG. 1 is a device configuration diagram of the present embodiment. The computer 1 includes an output device 2, an input device 3 and a storage device 4. The storage device 4 stores a sales performance information table in which time-series data of sales volume performance of forecast target products is registered.
Bull 200 (Fig. 2), event item information table 300 in which items of events likely to be caused by fluctuations in sales volume are registered.
(FIG. 3), it has an event result information table 400 (FIG. 4) in which the occurrence results of events are registered. The sales performance information table 200 includes a prediction target product name storage field 201 and a sales performance data field 2 indicating time-series data of product sales performance values.
And 02. The event item information table 120 includes an event item name storage field 301. Event record information table
The bull 400 includes an event occurrence time storage field 401 indicating the event occurrence time and an event detail content storage field 4 indicating the detailed content of the event.
And 02.

FIG. 5 is a flow chart showing the operation of this embodiment. The processing operation of the computer 1 and the information transmission operation with the output device 2, the input device 3, and the storage device 4 will be described with reference to FIG. Note that in this embodiment, a consumable material with high seasonality is set as a prediction target product.

<Step 501> Acquisition of Sales Result Data of Predicted Product Sales result information table 20 stored in storage device 4
Sales record data of the product to be predicted from 0 is fetched in 1 record and stored in the time series data table 600 of the format shown in FIG.

<Step 502> Extraction of variation factors from time-series data The variation of time-series data is multiplied by each variation factor of seasonal variation, trend / circulation variation, and irregular variation as shown in the following equation. Often composed by composition. Y (t) = T (t) * S (t) * I (t) where t: time point Y (t): observed value at time point T (t): 〃 combined value of trend fluctuation and circulation fluctuation S ( t): 〃 Seasonal fluctuation value I (t): 〃 Irregular fluctuation value Therefore, in this step, each of seasonal fluctuation, trend / circulation fluctuation, and irregular fluctuation is calculated from the sales performance data of the forecast target product captured in step 501. A variable element is extracted and a variable element information table 700 of the format shown in FIG. 7 is created. It should be noted that, in the present embodiment, a time-series data is used as a method for extracting each variable element.
Since a census station method or an EPA (the Economic Planning Agency) method, which is a well-known technique for data analysis, is applied, detailed description of the calculation method is omitted here.

<Step 503> Display of each variable element In order to assist the planner in setting a singular value, time series data is displayed.
Time-series data relating to the sales performance of the prediction target product stored in the table 600 and the variable element information table 70
Each variable element data stored in 0 is displayed on the same screen of the output device 2, and each display coordinate value data is stored in the display coordinate value table 800 shown in FIG. FIG. 9 shows an example of the display. This figure shows a time series transition diagram area 901 that displays the time transition of time series data, a seasonal variation transition map area 902 that displays the temporal transition of seasonal fluctuation, and a trend that displays the temporal transition of trend / circulation fluctuation. A cyclic fluctuation transition diagram area 903 and an irregular fluctuation transition diagram area 904 displaying a temporal transition of irregular fluctuations.

<Step 504> Acquisition of singular value Referring to each variable element information displayed in step 503, the singular value in the time series data set interactively by the planner with the computer 1 is acquired. The specific procedure is shown in FIG. First, as shown in FIG. 11, a singular value setting request message 1
After displaying 101 (step 1001), the display coordinate value of the singular value 1102 set by the planner is fetched (step 1002) and stored in the singular value display coordinate value table 1200 of the format shown in FIG. (Step 1
003).

<Step 505> Specification of Singular Value Occurrence Time The occurrence time of the singular value taken in step 504 is specified. The specific procedure is shown in FIG. First, the singular value display coordinate value table is loaded from the singular value display coordinate value table 1200 (step 1301), and the variable element display coordinate value table 8 is displayed.
The display coordinate value of the variable element showing the same value as the singular value display coordinate value is searched from 00 (step 1302), and the observation time point of the actual value corresponding to the display coordinate value is set as the generation time point of the singular value (step 1303). The result is stored in the singular value information table 1400 in the format shown in FIG. 14 together with the corresponding variable element name and display coordinate value (step 1304).

<Step 506> Display of set singular value points Points on the set singular value transition diagram are displayed in each of the transition diagrams 901 to 904 of the variable elements. The specific procedure is shown in FIG. First, the singular value occurrence time point is fetched from the singular value information table 1400 (step 15
01), search the display coordinate value table 800 for the observation time point of the actual value having the same value as the singular value generation time point (step 1
502), the display coordinate value of each variable element at the time of the observation is taken in (step 1503), and it is displayed on the transition chart of each variable element as a set singular value point (step 1504). FIG. 16 shows a display example thereof. In the figure, the points highlighted with black circles are the set singular values. In order to highlight the set singular value even with that method, it is also equipped with an identification function such as coloring and blinking.

<Step 507> Capture of Singular Value Generating Factors Factors that are most likely to be caused by the generation of the singular value set with reference to the candidate of the singular value generating factor are captured. The concrete procedure is shown in FIG. In the present embodiment, an example in which irregular fluctuation is used to set a singular value will be used below. First, the event name of the event item information table 300 is displayed near the set singular value as a candidate for the singular value occurrence factor as shown in FIG. 18 (step 1701), and the factor considered to be the singular value occurrence factor is determined by the planner. (Step 1702). Furthermore, when there is an event that occurs at the same time as the singular value occurrence time based on the event record information table 400, detailed information of the event is displayed as incidental information of the occurrence factor as shown in FIG. Step 1703). If the planner determines that the content is appropriate as a factor of the singular value generation, the factor is regarded as a factor of the singular value generation, and the singular value generation factor information table 2000 in the format shown in FIG.
(Step 1704).

<Step 508> Acquisition of Correction Value of Singular Value A correction value for the set singular value is acquired. A specific procedure will be described below with reference to FIG.

[Step 2101] Display of Correction Value Setting Method Input Request Message A correction value setting method input request message for selecting whether to set the correction value automatically or manually is displayed. FIG.
An example is shown in.

[Step 2102] Judgment of Correction Value Setting Method When the correction value is automatically set, the process proceeds to step 2103, and when it is manually set, the process proceeds to step 2104.

[Step 2103] Calculation of Correction Value A correction value is calculated by removing the degree of influence of irregular fluctuation, which is a major cause of the occurrence of singular values. Specifically, based on the trend / circular fluctuation value data and seasonal fluctuation value data of the fluctuation element information table 700, and the singular value occurrence time data of the singular value information table 1400. Correction value H at time u when singular value occurs
(u) is calculated by the following equation, and the process proceeds to step 2105.

H (u) = T (u) * S (u) where T (u): combined value of trend fluctuation and cyclic fluctuation at time u S (u): Seasonal fluctuation value [step 2104] correction value Import of the correction value of the singular value set manually. Specifically, as shown in FIG. 23, an interface that can move the singular value point 2301 to the correction value point 2302 is prepared, and the manually set correction value point of the singular value and the coordinate value of the singular value point are set. A correction value H (u) at the singular value generation time u is calculated from the difference amount by the following equation, and the process proceeds to step 2105.

H (u) = ((y (u) -d) / y (u)) * Y (u) where Y (u): observed value at time u y (u): Y (u) Display coordinate value d: Relative movement coordinate amount to correction value point [Step 2105] Storage of correction value The correction value calculated in Step 2103 or Step 2104 is stored in the correction value table 2400 in the format shown in FIG.

<Step 509> Creation of corrected time series data In the sales record information table 110, the record value at the time of occurrence of the singular value is replaced with the correction value in the correction value table 2300, and the format shown in FIG. The corrected time series data table 2500 is created.

<Step 510> Calculation of Prediction Value When a factor causing a singular value is predicted in the prediction target period, the prediction value in the prediction target period is calculated by considering the influence degree of the generation factor. .

X (w) = A (w) * B (w) * C (w) However, w: prediction target time A (w): seasonal fluctuation value at prediction target time w B (w): 〃 trend ・Circulation fluctuation value C (w): Irregular fluctuation value due to singular value generation factor Hereinafter, a specific procedure for calculating the predicted value will be described with reference to FIG. [Step 2601] Extraction of Fluctuation Elements of Corrected Time-series Data In the same manner as in step 502, the fluctuation elements of the corrected time-series data, that is, seasonal fluctuations, trends and circulation fluctuations are extracted. According to the variable element extraction technique (the census station method or the EPA method) used in this embodiment, the seasonal fluctuation value A (w) at the prediction target time point w can be obtained, so that the predicted value B (w of the trend / circulation fluctuation is obtained. ) And a random fluctuation value C (w) due to a singular value generation factor that cannot be regarded as white noise,
The predicted value X (w) can be calculated.

[Step 2602] Determination of regression order The result of interpolating several polynomials in the trend / cyclic fluctuation series is AIC (Akaike's Information Criter).
ion) to determine the regression order of the polynomial with the best fitting. For example, as shown in FIG. 27, in the polynomial fitting to the trend / circular fluctuation series 2701, if the AIC is the minimum in the linear expression 2702, it is set as the regression order for calculating the predicted tendency / cyclic fluctuation value. Although polynomial fitting is adopted as the interpolation technique for the time series data here, there are many alternatives established as well-known techniques such as the moving average method and the exponential smoothing method. An appropriate technique may be selected according to the characteristics of the time series and incorporated into this step and executed.

[Step 2603] Calculation of Prediction Trend / Circulation Fluctuation Value A prediction tendency / circulation fluctuation value B (w) at the prediction target time point w is calculated using the regression polynomial based on the regression order obtained in Step 2602. For example, the sales volume forecast period is shown in FIG.
When set as follows, the values P (t + 1) of the points 2801 and 2802 on the extension line of the linear expression 2702,
Let P (t + 2) be the prediction tendency / circulation fluctuation value of t + 1 and t + 2, respectively.

[Step 2604] Calculation of degree of influence of singular value generation factor The event name and event content of the singular value generation factor information table 2000 are displayed as shown in FIG. 29. For example, 2901 in FIG.
When the factor of is set, the correction value H (u) of the correction value table 2400 and the observation point of time Y (u) are obtained by using the generation time u of the factor as a key, and the singular value generation factor is calculated by the following equation. The degree of influence K (u) is calculated, and is used as the irregular fluctuation value C (w) due to the singular value generation factor at the prediction target time point w.

K (u) = H (u) / Y (u) ≈C (w) If no factor is set in this step, the correction value H (u) = observation time Y (u) is processed. To do.

[Step 2605] Calculation of Prediction Value at Prediction Target Time Prediction tendency / circulation fluctuation value B obtained in Step 2603
(w), the irregular variation value C (w) due to the singular value generation factor obtained in step 2604, and the seasonal variation value A (w) obtained in step 2601 are multiplied to obtain the prediction value X ( w) is calculated. The result is stored in the sales amount prediction result table 3000 in the format shown in FIG.

<Step 511> Calculation of Prediction Values of All Prediction Target Products and Judgment of Completion If calculation processing of prediction values for all prediction target products in the sales record information table 200 is completed, step 5
12. If not, return to step 501.

<Step 512> Display of Prediction Result The contents of the sales volume prediction result table 3000 are displayed on the output device 2. An example of the display is shown in FIG. In the same figure, the prediction result is displayed in the form of a form,
It also has a function to display in graph format.

As described above, according to the present embodiment, the accuracy of singular value extraction is improved by comprehensively displaying the candidate for the occurrence of singular value, which is a major cause of the decrease in the accuracy of sales volume prediction, and the singular value that has occurred in the past. It can be expected that the sales volume forecast will be made more accurate by using the impact of the factors causing the forecast.

Furthermore, according to the present embodiment, a series of forecasting operations for realizing highly accurate sales volume forecasting can be automated.
It is possible to reduce the load of the forecasting work of the planner in the planning work.

[0040]

According to the present invention, the work of extracting / correcting the singular value, which is a major cause of the decrease in the accuracy of sales volume prediction, is accelerated.
Since it can be accurately targeted, it is possible to improve the accuracy of sales volume forecast.

[Brief description of the drawings]

FIG. 1 is an apparatus configuration diagram of an embodiment.

FIG. 2 is a configuration diagram of a sales performance information table relating to a product to be predicted.

FIG. 3 is a configuration diagram of an event item information table.

FIG. 4 is a configuration diagram of an event record information table.

FIG. 5 is a flowchart showing a procedure of sales volume prediction.

FIG. 6 is a block diagram of a time series data table.

FIG. 7 is a configuration diagram of a variable element information table.

FIG. 8 is a configuration diagram of a display coordinate value table.

FIG. 9 is a display screen example of each variable element.

FIG. 10 is a flow chart showing a procedure for taking in a singular value.

FIG. 11 is an example of a screen for setting a singular value.

FIG. 12 is a configuration diagram of a singular value display coordinate value table.

FIG. 13 is a flowchart showing a specifying procedure at the time of occurrence of a singular value.

FIG. 14 is a configuration diagram of a singular value information table.

FIG. 15 is a flow chart showing a display procedure of set singular value points.
It is a chart.

FIG. 16 is an example of a display screen of set singular value points.

FIG. 17 is a flow chart showing a procedure for importing a singular value generation factor.
It is a chart.

FIG. 18 is an example of a screen for inputting a singular value generation factor.

FIG. 19 is an example of a display screen of detailed information regarding a cause of occurrence.

FIG. 20 is a configuration diagram of a singular value generation factor information table.

FIG. 21 is a flow chart showing a procedure for taking in a correction value of a singular value.
It is a chart.

FIG. 22 is an example of a display screen of a correction value setting method input request message.

FIG. 23 is an example of a screen for inputting a correction value of a singular value.

FIG. 24 is a configuration diagram of a correction value table.

FIG. 25 is a block diagram of a corrected time series data table.

FIG. 26 is a flowchart showing a procedure for calculating a predicted value.

FIG. 27 is a diagram showing an example of fitting a tendency / circular fluctuation series by a linear expression.

FIG. 28 is a diagram for explaining a method of calculating a predicted value by extension of a regression line.

FIG. 29 is an example of a screen for inputting prediction conditions.

FIG. 30 is a configuration diagram of a sales volume prediction result table.

FIG. 31 is a display screen example of a sales volume prediction result in a form.

FIG. 32 is a display screen example of a sales volume prediction result in a graph format.

[Explanation of symbols]

1 ... Calculator, 2 ... Output device, 3 ... Input device, 4 ... Storage device, 503 ... Display of each variable element, 506 ... Display of setting singular value point, 507 ... ..Incorporation of singular value generation factors, 508 ...
Incorporation of correction values for singular values, 510 ... Calculation of predicted values

Claims (6)

[Claims] 1. A storage device including an input device, an output device, a storage area for information on a sales volume result of a product, and a storage area for information on a factor sales performance due to fluctuations in the sales volume of the product,
Also, in the time series forecasting method using a computer having a computing device that calculates the sales volume of the product in the forecast period input from the input device based on the information about the sales volume record in the storage device, Extracting seasonal fluctuations, trend / circulation fluctuations, and irregular fluctuations from the original time-series data relating to the quantity results, and the time-series transition of each of the above-mentioned fluctuation elements and the time-series transition of the original data. Is simultaneously displayed on the output device as a graph, and the singular value in each time-series transition displayed on the graph is taken in from the input device, the occurrence time point of the singular value is specified, and various factor candidates that occur at the same time point as the occurrence time point A time-series prediction method, wherein information is output to the output device. 2. An input device, an output device, a storage device including a storage region of information about a sales volume result of a product and a storage region of information about a factor performance result due to a variation in the sales volume of the product,
Also, in the time series forecasting method using a computer having a computing device that calculates the sales volume of the product in the forecast period input from the input device based on the information about the sales volume record in the storage device, The factor of occurrence set for the singular value in the time-series original data relating to the actual quantity and the correction value of the singular value are fetched from the input device, and the actual value at the time of occurrence of the singular value of the original data is corrected. Corrected time-series data is created by substituting the values, and if a factor causing a singular value is expected within the corrected time-series data and the prediction target period, the past influence degree of that factor is calculated. A time-series forecasting method characterized by calculating the sales volume of products based on the forecast period. 3. The time series prediction method according to claim 1,
A time series characterized by accepting singular values set by the planner with a time series transition chart of original data and arbitrary transition charts of individual time series transition charts of seasonal fluctuations, trends / circulation fluctuations, and irregular fluctuations Prediction method. 4. The time series prediction method according to claim 1, wherein
A time-series prediction method characterized in that the cause of the singular value is set in an interactive form with the planner based on various factor candidate information generated at the same time as the time of occurrence of the singular value. 5. The time series prediction method according to claim 3,
When a set singular value is accepted in the time series transition chart of the original data and any one of the time series transition charts of seasonal fluctuations, trends / circulation fluctuations, and irregular fluctuations, other time series that remain A time series prediction method characterized by highlighting a display position of a set singular value also in a transition diagram. 6. The time series prediction method according to claim 2,
A time-series prediction method characterized in that a value obtained by removing an irregular variable element of an actual value set as a singular value is used as a correction value for the singular value.