JP2013131259A - Integrated demand prediction device, integrated demand prediction method, and integrated demand prediction program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an integrated demand prediction device capable of highly accurately predicting a demand for a plurality of different prediction periods, and to provide an integrated demand prediction method and an integrated demand prediction program.SOLUTION: The integrated demand prediction device construct a plurality of demand prediction models on the basis of a demand result quantity stored in demand result quantity storage means for storing past demand result quantities; calculates a prediction demand quantity for every constructed demand prediction model; calculates a prediction error for each of the plurality of demand prediction models in a common period, on the basis of the calculated prediction demand quantity and the demand result quantities stored in the demand result quantity storage means; calculates, for every prediction period, a blend ratio for each of the plurality of demand prediction models for making the calculated prediction error with respect to the prediction error within a predetermined prediction error range; and calculates the blend ratio for every calculated demand prediction model and the prediction demand quantity for every calculated demand prediction model so as to calculate the prediction demand quantity in the predetermined prediction period.

Description

  The present invention relates to an integrated demand prediction apparatus, an integrated demand prediction method, and an integrated demand prediction program.

  Conventionally, prediction is performed by a plurality of different methods as a method and apparatus for predicting the future demand amount in power, products, financial products, etc. whose demand amount changes over time. A prediction device that includes a portion and a portion that calculates an average value of the prediction data as prediction data has been proposed (for example, see Patent Document 1). In the conventional technique, prediction is performed using a plurality of different prediction methods, and based on the prediction error and other indicators, an optimization method is used so that the prediction error is reduced. A “weighted ratio” that is a combination ratio is determined, and prediction is performed based on the ratio.

Japanese Patent Laying-Open No. 2005-332122

  In order to perform highly accurate demand forecasting, it is important to construct a forecasting model that combines optimal forecasting methods based on sales results that change over time. Assuming demand forecasts at retail stores, the purchase business performs cumulative long-term forecasts, and the ordering business performs short-term forecasts until the delivery date. Depending on the period of prediction, an appropriate prediction method is different. Therefore, it is required that the demand forecasting apparatus considering a plurality of viewpoints such as “short-term viewpoint” and “long-term viewpoint” can be commonly used.

  Furthermore, not only differences in the viewpoint of retail operations, but also companies that have different positions, such as manufacturers and retail stores, or wholesalers and retail stores, use the common demand forecast to predict results (forecast demand). We think that it will be possible to reduce confusion between companies (excess inventory, missing items, etc.).

  However, it does not consider the long-term and short-term viewpoints, and uses the same prediction model, so the prediction period is high and accurate according to the prediction period, such as the period for which the prediction is performed is near or far from the present. There is a problem that cannot be realized. Even if multiple forecast models are used according to each forecast period, long-term and short-term forecasts are not evaluated for forecast errors considering both, so a series of operations such as purchasing In operations such as planning and ordering, the forecast values had to be used properly according to the forecast period.

  The present invention has been made in consideration of such circumstances, and an object of the present invention is to provide an integrated demand prediction apparatus and an integrated demand prediction that can perform more accurate demand prediction for a plurality of different prediction periods. It is to provide a method and an integrated demand forecasting program.

  In order to solve the above-described problem, the present invention provides a plurality of demand predictions based on a demand actual amount storage unit that stores past actual demand amounts and a demand actual amount stored in the demand actual amount storage unit. Demand prediction model construction means for constructing a model, predicted demand amount calculation means for calculating a predicted demand amount for each demand prediction model constructed by the demand prediction model construction means, and calculated by the predicted demand amount calculation means Prediction error calculation means for calculating a prediction error for each of the plurality of demand prediction models in a common period based on the predicted demand quantity and the actual demand quantity stored in the demand actual quantity storage means, and the prediction error calculation The blending ratio for each of the plurality of demand prediction models is calculated for each prediction period, so that the prediction error calculated by the means is within a predetermined prediction error range with respect to the prediction error. Integrated demand forecast blend ratio calculating means, blend ratio for each demand forecast model calculated by the integrated demand forecast blend ratio calculating means, and forecast demand for each demand forecast model calculated by the forecast demand quantity calculating means. An integrated demand forecasting apparatus comprising: a demand forecast result output means for calculating a forecast demand amount in a forecast period determined in advance.

  Further, in order to solve the above-described problem, the present invention predicts a demand prediction model construction step of constructing a plurality of demand prediction models based on the past demand actual amount, and for each of the constructed demand prediction models. A predicted demand amount calculating step for calculating a demand amount; a prediction error calculating step for calculating a prediction error for each of the plurality of demand prediction models in a common period based on the calculated predicted demand amount and the actual demand amount And an integrated demand forecast blend ratio for calculating a blend ratio for each of the plurality of demand forecast models for each forecast period, wherein the calculated forecast error is within a predetermined forecast error range with respect to the forecast error. By calculating the calculation step, the blend ratio for each calculated demand prediction model, and the predicted demand amount for each calculated demand prediction model, An integrated demand forecasting method, which comprises a demand forecast result output step of calculating a predicted demand in the period.

  In order to solve the above-described problems, the present invention provides a computer with a demand prediction model construction step of constructing a plurality of demand prediction models based on the past demand actual amount, and for each of the constructed demand prediction models. A prediction demand amount calculating step for calculating a predicted demand amount, a prediction error for calculating a prediction error for each of the plurality of demand prediction models in a common period based on the calculated predicted demand amount and the actual demand amount An error calculating step, and an integrated demand for calculating a blend ratio for each of the plurality of demand prediction models for each prediction period, wherein the calculated prediction error is within a predetermined prediction error range with respect to the prediction error. By integrating the predicted blend ratio calculating step, the calculated blend ratio for each demand forecast model, and the predicted demand amount for each calculated demand forecast model, An integrated forecast program for executing the demand prediction result output step of calculating a predicted demand in the prediction period is determined because.

  According to the present invention, a plurality of demand prediction models are constructed based on the past demand actual amount, the predicted demand amount is calculated for each of the constructed prediction models, and the calculated predicted demand amount and the actual demand result are calculated. Based on the quantity, calculate a prediction error for each of the plurality of demand prediction models in the common period, calculate a blend ratio for each of the plurality of demand prediction models that minimizes the prediction error for each of the plurality of demand prediction models, Based on the blend ratio, a predicted demand amount in a predetermined prediction period is calculated using a plurality of prediction methods. Therefore, there is an advantage that demand prediction with higher accuracy can be performed for a plurality of different prediction periods.

  As an example, an integrated blend ratio for each of multiple demand prediction models that minimizes short-term prediction errors that are short-term prediction errors and long-term prediction errors that are long-term prediction errors is calculated for each prediction period. . Therefore, there is an advantage that demand prediction with higher accuracy can be performed for a plurality of different prediction periods.

  As an example, the integrated prediction error is calculated by adding and adding a predetermined weight value to the short-term prediction error and the long-term prediction error. Therefore, there is an advantage that demand prediction with higher accuracy can be performed for a plurality of different prediction periods.

  Also, as an example, when the predicted demand is outside a predetermined threshold range, the prediction model that calculates either the maximum or minimum predicted demand is removed, and a plurality of demand prediction models are constructed again. Therefore, there is an advantage that it can be more adapted to the operator's sense of prediction.

It is a block diagram which shows the structure of the integrated demand prediction apparatus by embodiment of this invention. It is a conceptual diagram which shows the data structure of the goods sales performance table 1 of the integrated demand prediction apparatus by embodiment of this invention. It is a conceptual diagram which shows the data structure of the prediction demand table 3 classified by goods for every prediction technique of the integrated demand prediction apparatus by embodiment of this invention. It is a conceptual diagram which shows the data structure of the optimization weight table 5 of the integrated demand prediction apparatus by embodiment of this invention. It is a conceptual diagram which shows the data structure of the blend ratio table 7 of the integrated demand prediction apparatus by embodiment of this invention. It is a conceptual diagram which shows the data structure of the prediction period table 8 of the integrated demand prediction apparatus by embodiment of this invention. It is a conceptual diagram which shows the data structure of the demand prediction result table 10 of the integrated demand prediction apparatus by embodiment of this invention. It is a flowchart for demonstrating operation | movement of the integrated demand prediction apparatus by embodiment of this invention. It is a conceptual diagram explaining the construction method of the prediction method by this embodiment. It is a conceptual diagram which shows the example of calculation of the blend ratio of each prediction method by this embodiment. It is a conceptual diagram for demonstrating the subject in a prior art. It is a key map for explaining demand forecast by this embodiment.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

  FIG. 1 is a block diagram showing a configuration of an integrated demand prediction apparatus according to an embodiment of the present invention. As shown in FIG. 2, the product sales record table (demand performance amount storage means) 1 includes, for each product JAN code for identifying the product, the date of sale, the number of days elapsed since the release, and the sales volume. The product sales record (history) is held. The prediction value calculation unit (demand prediction model construction means, prediction demand amount calculation means) 2 includes demand prediction model construction means for constructing a plurality of demand prediction models, and from the sales volume (actual value) of the product sales performance table 1 Build multiple demand forecast models (prediction methods) prepared for each product JAN code, and calculate the forecast sales volume (forecast demand) for each product JAN code (k) for each forecast period for each constructed forecast model. And stored in the forecast demand table 3 for each product by forecast method.

  In the present embodiment, the prediction period is described in units of one day. As a prediction method, AR (Auto Regressive model), MA (Moving Average model), ARIMA (Auto Regressive Integrated Moving Average model), simple moving average, exponential smoothing method, multiple regression model, etc. of time series analysis are used. When using data other than sales quantity such as a multiple regression model, it is necessary to prepare data other than sales-related elements such as the number of customers on the day, weather, and temperature as a table.

  As shown in FIG. 3, the forecast demand table 3 for each product according to the forecast method holds a forecast method (prediction model), a forecast date, and a forecast sales quantity (predicted value) for each product JAN code. The prediction value error calculation unit (prediction error calculation means) 4 predicts from the sales volume (actual value) of the forecast demand amount table 3 for each product by prediction method and the forecast sales volume (predicted demand amount) of the product sales result table 1. Calculate the error. The period in which the prediction error is calculated is a common period in which the predicted value is calculated by a plurality of constructed prediction methods. This is because the number of data required for model construction differs depending on the prediction method.

  As shown in FIG. 4, the optimization weight table 5 includes “weights” used to determine which prediction error is prioritized among a plurality of prediction errors to be minimized in the calculation of the integrated demand prediction blend ratio. Hold. The plurality of prediction errors include, for example, a short-term prediction error and a long-term prediction error, which will be described later. The former is an average of squared errors of Equation (2) described later, and the latter is Equation (3) described later. ) Is represented by the square of the error mean. These prediction errors are prioritized with weights when priority is given to either short-term or long-term accuracy. Usually, the same weight may be used. The integrated demand prediction blend ratio calculation unit (integrated demand prediction blend ratio calculation means) 6 uses a prediction error for each prediction date calculated by the prediction value error calculation unit 4 to calculate a short-term viewpoint and a long-term viewpoint. The prediction error is processed from both sides (short-term prediction error calculation means, long-term prediction error calculation means).

  Further, the integrated demand prediction blend ratio calculation unit (demand prediction result output means) 6 integrates these processed prediction errors and the weights of the optimization weight table 5, and integrates the prediction errors from two viewpoints. The combination ratio (blend ratio) of each prediction method is calculated and stored in the blend ratio table 7 so as to minimize the integrated prediction error. The blend ratio is formulated in the form of a “secondary planning problem” of an optimization problem described later. As shown in FIG. 5, the blend ratio table 7 holds a prediction method (prediction model), a prediction date, and a blend ratio for each product JAN code.

  As shown in FIG. 6, the prediction target period table 8 holds a prediction target period for each product JAN code. The demand forecast result output unit 9 uses the forecast target period of the forecast target period table 8 and the constructed forecast method stored in the forecast demand table 3 for each product according to the forecast method to obtain the forecast value for the forecast target period. By calculating and integrating the predicted value and the blend ratio of the blend ratio table 7, a predicted sales quantity (predicted demand amount) in each final prediction period is calculated and stored in the demand prediction result table 10. As shown in FIG. 7, the demand prediction result table 10 holds a final forecast sales quantity that is a forecast period and a forecast value for each product JAN code.

Next, the operation of this embodiment will be described.
FIG. 8 is a flowchart for explaining the operation of the integrated demand prediction apparatus according to the present embodiment. The predicted value calculation unit 2 constructs a plurality of demand prediction models (prediction methods) for each product JAN code from the sales volume (actual value) in the product sales performance table 1, and for each constructed prediction model, (i The forecasted sales volume (predicted demand) for each product JAN code (date) is calculated and stored in the forecasted demand table 3 for each product by forecast method (step S1).

  Next, the predicted value error calculation unit 4 calculates a prediction error from the sales volume (actual value) in the forecast demand amount table 3 for each product by prediction method and the predicted sales volume (predicted demand amount) in the product sales result table 1. And it supplies to the integrated demand prediction blend ratio calculation part 6 (step S2). The period in which the prediction error is calculated is a common period in which the predicted value is calculated by a plurality of constructed prediction methods.

  Next, the integrated demand prediction blend ratio calculation unit 6 processes the prediction error from both the short-term viewpoint and the long-term viewpoint using the prediction error for each prediction date calculated by the prediction value error calculation unit 4. (Step S3). That is, from a short-term perspective, the forecast error is squared for each forecast period in the forecast demand amount table 3 by product for each forecast method, and totaled for the forecast target period. The prediction error is processed by taking the sum of the prediction target periods for the same prediction error and squaring it. Next, these processed prediction errors and the weights (0 to 1) of the optimization weight table 5 are integrated and integrated by adding the prediction errors from the two viewpoints, so that these integrated prediction errors are minimized. The blend ratio of each prediction method is calculated and stored in the blend ratio table 7 (step S4).

  Here, a method for calculating the blend ratio will be described. As described above, the blend ratio is formulated in the form of the “secondary programming problem” of the optimization problem. The quadratic programming problem is a problem that minimizes and maximizes the quadratic expression under the constraints of the linear expression, and portfolio optimization, optimal placement problem, and the like are formulated in this form. The minimum integrated prediction error minimum p (x) can be expressed by the following equation (1).

_wl is a weight for each prediction error. l indicates a prediction error, where l = 1 corresponds to the first prediction error (short term), and l = 2 corresponds to the second prediction error (long term), and corresponds to each prediction error in FIG. x _ij is a rate at which the prediction method j is incorporated after i days. In this embodiment, the prediction error to be minimized is expressed by the following equations (2) and (3).

Equation (2) is the first prediction error and represents the daily prediction error, and Equation (3) is the second prediction error and represents the cumulative error for a certain period. In addition, the predicted demand amount of the commodity k after i days at the time point t by the prediction method j is y (^) _jtik , and the actual demand amount is y _tik . That is, a prediction method after n days may be selected so as to satisfy two constraints. The symbol y (^) is represented as a symbol with the symbol ^ on the symbol y in the formula (2) or (3).

  FIG. 9 is a conceptual diagram illustrating a calculation example of the blend ratio of each prediction method according to the present embodiment. In the figure, demand prediction with higher accuracy (small prediction error) is performed by appropriately combining the blend ratios of the prediction methods 1 to N for each prediction period. When the prediction period is one day later, the prediction method 1 has a blend ratio of 0.1, the prediction method 2 has a blend ratio of 0.01,..., And the prediction method N has a blend ratio of 0.4 (total = 1). 2 days later, the blend ratio of prediction method 1 is 0.03, the blend ratio of prediction method 2 is 0.24,..., The blend ratio of prediction method N is 0.01 (total = 1), and the prediction period is 30. After the day, the blend ratio of the prediction method 1 is 0.1, the blend ratio of the prediction method 2 is 0.25,..., And the blend ratio of the prediction method N is 0.03 (total = 1).

  Next, the demand prediction result output unit 9 uses the prediction target period of the prediction target period table 8 and the constructed prediction method stored in the prediction demand table 3 for each product by prediction method for the prediction target period. A predicted value is calculated (step S5), and the predicted value and the blend ratio in the blend ratio table 8 are integrated, thereby calculating a predicted value in the prediction target period and storing it in the demand prediction result table 10 (step S6). .

  FIGS. 10A and 10B are conceptual diagrams for explaining the problems in the prior art. In retail stores, demand prediction from a plurality of positions (ordering operations, purchasing operations) is required. In other words, in the ordering operation shown in FIG. 11A, it is necessary to predict the order quantity every day (verification date) and one day later, whereas in the purchase plan shown in FIG. It is necessary to predict a long-term (one month) cumulative demand such as one day later, two days later,... 30 days later.

  FIG. 11 is a conceptual diagram for explaining demand prediction according to the present embodiment. In the present embodiment, by the above-described processing, by optimizing so as to minimize two of “cumulative prediction error” and “daily prediction error”, a prediction value one day after the verification date, two days after Predicted values in the prediction period are calculated as predicted values, predicted values after 15 days, predicted values after 16 days,... Predicted values after 30 days.

  FIG. 12 is a conceptual diagram illustrating a method for constructing a prediction method according to a modification of the present embodiment. In the modified example of the present embodiment, when a demand threshold (SS1: upper limit / lower limit) is determined from past experience, the prediction method that greatly exceeds the prediction range is removed, and the above-described steps S1 to S6 are repeated. It is characterized by constructing a prediction method. In the figure, the threshold (upper limit / lower limit) of the demand amount is determined from past experience (SS1). If the demand amount predicted by the integrated demand prediction model is outside the threshold value, the prediction model predicted to be the largest (or smallest) is removed (SS2), and the prediction model is constructed again. If the predicted value does not fall within the threshold range even if the prediction model is repeatedly constructed, the immediately preceding predicted value sales number (predicted demand amount) or actual value is used (SS3).

  According to the above-described embodiment, by using an optimization method, an “integrated demand prediction model” that can be commonly used for a plurality of different forecast periods is constructed, and thus an evaluation axis that is emphasized from a plurality of different viewpoints. Certain prediction errors can be suppressed together.

  Further, by using the common prediction value, it is possible to make a production plan in cooperation with a retail store or a wholesaler. Such information sharing makes it possible to reduce the risk of divergence of predicted demand values.

  In addition, since the blend ratio of the prediction method is determined according to the prediction period, a more accurate model corresponding to the prediction period can be constructed.

  In addition, when the upper and lower thresholds are set and the predicted value is out of the range, the operator's sense of prediction can be taken in more by reconstructing the integrated demand prediction model.

  In the above-described embodiment, the prediction model is constructed in consideration of different viewpoints, such as reducing the prediction error from a long-term viewpoint and reducing the prediction error from a short-term viewpoint. When the range of the integrated prediction error can be determined based on the past results, the range of the integrated prediction error such as “I want to keep the integrated prediction error within XX and below Δ △” By adding to the constraint equation shown in (3), it can be incorporated into the prediction model.

  Moreover, in this embodiment mentioned above, although the prediction period was demonstrated as the unit of 1 day, it is good also considering several days as one prediction period as needed. For example, the prediction period may be a week unit.

  In addition, a program for realizing each function by the above-described prediction value calculation unit 2, prediction value error calculation unit 4, integrated demand prediction blend ratio calculation unit 6, and demand prediction result output 9 is recorded on a computer-readable recording medium. Then, the program recorded on the recording medium may be read into the computer system and executed to perform the encoding process and the decoding process. Here, the “computer system” may include an OS and hardware such as peripheral devices. Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used. The “computer-readable recording medium” means a flexible disk, a magneto-optical disk, a ROM, a writable nonvolatile memory such as a flash memory, a portable medium such as a CD-ROM, a hard disk built in a computer system, etc. This is a storage device.

  Further, the “computer-readable recording medium” means a volatile memory (for example, DRAM (Dynamic DRAM) in a computer system that becomes a server or a client when a program is transmitted through a network such as the Internet or a communication line such as a telephone line. Random Access Memory)), etc., which hold programs for a certain period of time. The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, what is called a difference file (difference program) may be sufficient.

DESCRIPTION OF SYMBOLS 1 Merchandise sales result table 2 Predicted value calculation part 3 Predicted demand table according to goods according to prediction method 4 Predicted value error calculation part 5 Optimization weight table 6 Integrated demand forecast blend ratio 7 Blend ratio table 8 Forecast period table 9 Demand forecast result output Part 10 Demand forecast result table

Claims (3)

Demand actual quantity storage means for storing past demand actual quantity;
A demand prediction model construction means for constructing a plurality of demand prediction models based on the demand actual quantity stored in the demand actual quantity storage means;
For each demand forecast model constructed by the demand forecast model construction means, a forecast demand quantity calculation means for calculating a forecast demand quantity;
Based on the predicted demand amount calculated by the predicted demand amount calculation means and the actual demand amount stored in the actual demand amount storage means, a prediction error is calculated for each of the plurality of demand prediction models in a common period. A prediction error calculation means;
Integrated demand for calculating a blend ratio for each of the plurality of demand prediction models for each prediction period, wherein the prediction error calculated by the prediction error calculation means is within a predetermined prediction error range with respect to the prediction error. A predicted blend ratio calculating means;
A forecast period that is determined in advance by integrating the blend ratio for each demand forecast model calculated by the integrated demand forecast blend ratio calculation means and the forecast demand for each demand forecast model calculated by the forecast demand quantity calculation means An integrated demand forecasting apparatus, comprising: a demand forecast result output means for calculating a forecast demand amount in A demand prediction model construction step of constructing a plurality of demand prediction models based on the past demand actual amount;
For each constructed demand prediction model, a predicted demand amount calculating step for calculating a predicted demand amount;
A prediction error calculating step of calculating a prediction error for each of the plurality of demand prediction models in a common period based on the calculated predicted demand amount and the actual demand amount;
An integrated demand prediction blend ratio calculation step of calculating a blend ratio for each of the plurality of demand prediction models for each prediction period so that the calculated prediction error is within a predetermined prediction error range with respect to the prediction error. When,
A demand prediction result output step of calculating a predicted demand amount in a predetermined prediction period by integrating the calculated blend ratio for each demand prediction model and the predicted demand amount for each of the calculated demand prediction models. Integrated demand forecasting method characterized by including. On the computer,
A demand prediction model construction step of constructing a plurality of demand prediction models based on the past demand actual amount;
For each constructed demand prediction model, a predicted demand amount calculating step for calculating a predicted demand amount;
A prediction error calculating step of calculating a prediction error for each of the plurality of demand prediction models in a common period based on the calculated predicted demand amount and the actual demand amount;
An integrated demand prediction blend ratio calculation step of calculating a blend ratio for each of the plurality of demand prediction models for each prediction period so that the calculated prediction error is within a predetermined prediction error range with respect to the prediction error. When,
A demand prediction result output step of calculating a predicted demand amount in a predetermined prediction period by integrating the calculated blend ratio for each demand prediction model and the predicted demand amount for each of the calculated demand prediction models. Integrated demand forecasting program for execution.

Images