JP2008305229A - Demand forecast method and device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To forecast demand related to a commodity or a market with sufficient accuracy in a medium and long term. <P>SOLUTION: Time series data showing temporal fluctuation of an actual purchase number related to the commodity are stored by a time series data storage part 14, the time series data are read from the time series data storage part 14 by a demand growth curve specification part 23, a demand growth curve related to the commodity is specified by performing regression analysis processing of the time series data, a demand forecast value of the commodity at an arbitrary time point is calculated based on the demand growth curve by a forecast value calculation part 24, and a function wherein an increase number per unit time of the purchase number of the commodity is proportional to a negative power of time length from a prescribed reference time point is used as the demand growth curve. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a demand prediction technique for predicting demand for products and services.

  Medium- to long-term quantitative forecasts of products and services (hereinafter referred to as “products”) provide a basic basis for estimating large-scale investment and profit trends. With a certain level of accuracy, both public organizations and companies can avoid waste and shortage of investment at the planning stage of policies and measures, and contribute to industrial and economic efficiency. High value.

  In general, quantitative prediction of demand for a product is performed by specifying a demand growth curve that shows a significant change in the product in the future. Conventional techniques for identifying demand growth curves are essentially as follows. First, the demand growth curve is a logistic curve, a Lotka = Volterra model, a bus model (see, for example, Non-Patent Document 1) or a modification thereof (for example, see Non-Patent Document 2, etc.), or a nonlinear function identification such as a neural network. Depending on the model, it can be understood as a family of functions with some parameters. Next, the values taken by the parameters of the function family are obtained from product sales records, survey data such as optional surveys, data obtained from questionnaire surveys, etc., and medium- to long-term prediction is performed using the determined function.

Bass, F. M (1969) "A New Product Growth Model for Consumer Durables," Management Science Theory, Vol.15, No.5, January 1969, pp.215-227 Translated by Morimura et al. (1997) Marketing Handbook Asakura Shoten (Eliashberg J and Lilien G.L., ed. (1993) Marketing, Handbooks in Operations Research and Management Science Vol.5) Zipf, G. K. (1949) Human Behavior and the Principle of Least-Effort, Addison-Wesley, Cambridge MA

However, in such a conventional technique, there has been a problem that the medium- to long-term prediction value greatly fluctuates with respect to the addition or deletion of the actual value, and sufficient accuracy cannot be obtained by the medium- to long-term prediction.
This seems to be due to the fact that the data and the function system do not match directly, and it is often the case that the prediction is short-term. In addition, in the analysis and estimation of demand, detection of data loss and detection of the impact of products on the market due to qualitative changes are fundamental problems. High utility value. However, the conventional technology does not have the accuracy and theoretical support that can distinguish this.

  Moreover, even if a function that matches data up to the present time can be constructed, there are many problems in adopting it as a prediction method only by this or constructing a device that implements the prediction method. There are at least two reasons for this, as described below.

  It is well known that the first reason is that the function cannot always be applied subsequently. In fact, it is always possible to construct a nonlinear function system with a large number of parameters by a neural network or the like, but in that case, when new test data is added, it is necessary to change the learning parameters. In fact, when appropriate non-linear transformation is performed, the growth data may well be applied to a straight line in a certain section. However, the curve may start after that section.

  The second reason is that, as long as the prediction describes the future, there is a risk associated with the divergence that the prediction and the reality do not always match. On the other hand, simply presenting a demand growth curve does not allow the forecast user to understand how the divergence between the forecast and reality should be perceived, and evaluates the risk associated with the divergence or assumes the risk. I can't.

  Therefore, it is necessary not only to construct a function of a robust demand growth curve that matches past data, but also to provide a theoretical explanation and support why such a demand growth curve is obtained. Predictive users should be able to examine the hypotheses and supporting facts presented in the theoretical explanation and determine whether they should evaluate the content or take the risk if the theoretical explanation or support is known. Become. In fact, even in the classic model, not only a function of a demand growth curve is presented, but also a theoretical hypothesis that it becomes such a function is presented. The fact that the demand growth curve function does not fit the data means that some of these hypotheses are denied. In fact, when a classic model is reexamined, there is a part where the structure of the model appears to be different from the user's behavior.

  For example, by taking the hypothesis that each user's usage behavior is directly influenced by the macro parameter of demand growth and the relationship with the final demand, demand growth is expressed as a macro parameter differential equation. And it is unlikely that they will be directly affected by final demand. In the classic model, this risk included in the hypothesis was clearly shown, and the discrepancy between the data and the model is the surface of this risk.

  An object of the present invention is to provide a demand prediction method and apparatus capable of predicting demand related to goods or a market with sufficient accuracy in the medium to long term.

  In order to achieve such an object, a demand prediction method according to the present invention includes a storage unit and a calculation processing unit, and forms the demand among the demands of products in the market based on the number of actual purchases related to the products. Demand forecasting method used in a demand forecasting device for forecasting demand after the reversal point where the demands of the predecessor and the late follower are reversed, and the storage unit shows a temporal change in the actual purchase number related to the product, and the reversal point A time-series data storage step for storing time-series data including a time-series data storage step and an arithmetic processing unit for reading time-series data from the time-series data storage unit and performing time-axis conversion processing and regression analysis processing on the time-series data The demand growth curve calculation step for identifying the demand growth curve of the product after the reverse time and the calculation processing unit to determine the demand growth curve after the reverse time based on the demand growth curve. A demand forecasting step for calculating and outputting a demand forecast value of the product at the point of time, and an increase in the number of product purchases per unit time as a demand growth curve is a negative of the time length from a predetermined reference time point. A function proportional to is used.

  At this time, a function in which the increase number is proportional to the reciprocal of the time length from the reference time point may be used as the demand growth curve.

  As a demand growth curve, when the vertical axis is the number of products purchased and the horizontal axis is the logarithmic or negative integral function time axis, it is a curve that increases monotonically with the passage of time in the section before the reversal point. In addition, a function that is a straight line that monotonously increases with the passage of time in the section after the reverse rotation time may be used.

In addition, the calculation processing unit compares the demand forecast value obtained in the forecast value calculation step with the actual purchase value of the time series data, and based on the comparison result, qualitative changes in the product or time series data loss A warning may be output.
Specifically, the calculation processing unit compares the demand forecast value obtained in the forecast value calculation step with the actual purchase value of the time series data in the section where the demand growth curve forms a straight line, and the demand forecast value is the actual purchase. When the displacement in the direction below the value is indicated, a warning indicating the qualitative change of the product may be output.

  Alternatively, the calculation processing unit compares the demand forecast value obtained in the forecast value calculation step with the actual purchase value of the time series data in the section where the demand growth curve forms a straight line, and the demand forecast value exceeds the actual purchase value. When indicating a displacement in the direction, a warning indicating a lack of time-series data may be output.

  In addition, as the demand growth curve calculation step, a time axis conversion step that converts the time axis of the time series data to a logarithmic or negative integral function time axis, and any of the converted time series data obtained in the time axis conversion step A regression line sequence calculation step for obtaining a regression line of the converted time series data after the start time of each of the plurality of start time sequence columns, and a predetermined convergence based on a predetermined convergence condition for the regression line sequence obtained in the regression line sequence calculation step A convergence determination step for determining whether or not the line has converged, a minimum start time determined to be converged in the convergence determination step as a reverse time point, and a function indicating a demand growth curve in a section after the reverse time point You may make it do.

  Further, as a convergence condition, an allowable value of an angle cosine value between any two regression lines and an allowable value of a distance between these regression lines in a predetermined time axis section may be used.

  In addition, the demand prediction device according to the present invention predicts the demand after the reversal point where the demands of the predecessor and the successor forming the demand are reversed among the demands of the commodity in the market based on the actual number of purchases related to the commodity. It is a prediction device, shows a time change of the number of actual purchases related to a product, and stores time-series data including time-series data including the time of reversal, and reads time-series data from the time-series data storage unit, Demand growth curve specifying unit that identifies the demand growth curve of the product after the reversal time by performing time axis conversion processing and regression analysis processing on the series data, and at any time after the reversal time based on the demand growth curve A forecast value calculation unit that calculates and outputs a demand forecast value for the product, and the increase in the number of product purchases per unit time as a demand growth curve Using function proportional to the negative power of the time length from the time.

  At this time, a function in which the increase number is proportional to the reciprocal of the time length from the reference time point may be used as the demand growth curve.

  As a demand growth curve, when the vertical axis is the number of products purchased and the horizontal axis is the logarithmic or negative integral function time axis, it is a curve function that increases monotonically with the passage of time in the section before the reversal point. In addition, a function that is a straight line that monotonously increases with the passage of time in the section after the reversal point may be used.

Also, the demand forecast value obtained by the forecast value calculation unit is compared with the actual purchase value of the time series data, and a warning indicating the qualitative change of the product or the lack of time series data is output based on the comparison result. A data evaluation unit may be further provided.
Specifically, in the section where the demand growth curve forms a straight line, the demand forecast value obtained by the forecast value calculation unit is compared with the actual purchase value of the time series data, and the demand forecast value is below the actual purchase value. In the case of indicating the displacement, a warning indicating a qualitative change of the product may be output.

  Or, in the section where the demand growth curve forms a straight line, the demand forecast value obtained by the forecast value calculation unit is compared with the actual purchase value of the time series data, and the displacement in the direction in which the demand forecast value exceeds the actual purchase value In the case of showing, a warning indicating a lack of time-series data may be output.

  In addition, the demand growth curve specifying unit includes a time axis conversion unit that converts the time axis of the time series data to a logarithmic or negative integral function time axis, and any of the converted time series data obtained by the time axis conversion unit. The regression line sequence calculation unit that obtains the regression line of the converted time series data after the start time of each of the plurality of start time columns, and the regression line sequence obtained by the regression line sequence calculation unit based on a predetermined convergence condition A function for indicating a demand growth curve in a section after the reversal time, wherein the convergence start unit is provided with a convergence determination unit that determines whether or not the line has converged. You may make it do.

  Further, as a convergence condition, an allowable value of an angle cosine value between any two regression lines and an allowable value of a distance between these regression lines in a predetermined time axis section may be used.

  According to the present invention, the time-series data storage unit stores the time-series data indicating the time change of the actual purchase number related to the product and includes the reversal point, and the arithmetic processing unit stores the time-series data from the time-series data storage unit. Data is read and the time-axis conversion process and regression analysis process are performed on the time series data to identify the demand growth curve of the product after the reversal point. Calculate the demand forecast value of the product at an arbitrary point in time, and use a function in which the increase in the number of purchases of the product per unit time is proportional to the negative of the length of time from the predetermined reference time as the demand growth curve As a result, demand forecasts based on the structural theory of demand growth can be realized and demand can be predicted with sufficient accuracy in the medium to long term. That.

  Therefore, in the planning and evaluation of development / capital investments, policies, measures, etc., such as companies, public organizations, government agencies, etc. Knowing the contents of theoretical hypotheses and support, and understanding the risks, you can use medium- to long-term forecasts. As a result, it is possible to avoid an excessive or shortage of investment and formulate an appropriate investment plan or funding plan to obtain a high industrial or economic effect.

Next, embodiments of the present invention will be described with reference to the drawings.
First, with reference to FIG. 1, the demand prediction apparatus concerning one Embodiment of this invention is demonstrated. FIG. 1 is a block diagram showing a configuration of a demand prediction apparatus according to an embodiment of the present invention.

  The demand forecasting apparatus 10 is composed of an information processing apparatus using a computer as a whole, and as main functional units, an operation input unit 11, a screen display unit 12, a main storage unit 13, a time series data storage unit 14, a processing data recording unit. Unit 15, output data storage unit 16, arithmetic processing unit 20, central control unit 21, default setting unit 22, demand growth curve specifying unit 23, predicted value calculation unit 24, and data evaluation unit 25. The demand growth curve specifying unit 23 is provided with a time axis converting unit 31, a regression line sequence calculating unit 32, a convergence determining unit 33, and a negative saddle value tuning unit 34 as main functional units.

  In the present embodiment, the time-series data storage unit 14 stores time-series data indicating the temporal change in the actual purchase number of products, and the demand growth curve specifying unit 23 stores the time-series data from the time-series data storage unit 14. The demand growth curve related to the product is identified by performing regression analysis processing on the time series data, and the predicted value calculation unit 24 calculates the demand predicted value of the product at an arbitrary time based on the demand growth curve, and the demand As the growth curve, a function is used in which the number of increases in the number of purchased products per unit time is proportional to the negative of the length of time from a predetermined reference time point.

[Structural theory of demand growth]
First, the structural theory of demand growth used in the present invention will be described.
In complex systems, characteristic quantities may be obtained with a power function that is scale invariant. Even in the demand growth phenomenon of products, this characteristic can be seen until a certain period with respect to the number of subscribers of various broadband access services such as ADSL and FTTH and social networking services. FIG. 2 is an explanatory diagram showing a demand growth curve related to the number of subscribers of broadband services. The property that the demand growth phenomenon can be expressed by a single positive function is until a certain period of the first half of the demand growth phenomenon. Therefore, when data fitting is performed while requiring inertia that is not limited to increments, the latter half of the period is adapted to negative growth, which is dual as a function class in the first half. Then, “−1” is well suited as the incremental exponent.

  The inventors have considered the demand growth of such broadband access services, constructed a structural theory of demand growth, and followed the concept of the first and second generation by the bus, As for the mechanism, we built a hypothesis that is significantly different from the conventional theory. It is assumed that the start-up behavior of late users occurs on average when the accumulated amount of information from familiar surroundings reaches a certain amount, and the number of people sharing the familiar surroundings with the size of the channel of the information It is a hypothesis that it depends on. Based on this hypothesis, the demand growth curve of late users is almost logarithmic, based on Zip's law (see, for example, Non-Patent Document 3) for the well-known distribution of population sizes in cities. It is led to become. In light of demand data, in fact, the demand growth curve for Japan as a whole and prefectures is a good match for logarithmic growth after the peak period of demand growth.

The present invention builds a demand forecasting or analysis method by detecting data deficiencies and qualitative changes of products based on the support by the structural theory of demand growth. At this time, there are further problems as follows. That is, how to extract the dominant part of the late user from the time series data of demand as widely as possible by mechanical procedures or algorithms, in other words, in the market growth, Calculation of the reversal point of transition to a specific period becomes a problem.
If this can be solved, the prediction curve can be calculated by determining the constant multiple and the constant sum of the logarithmic function or the negative integral function from the time series data after the reversal point. In fact, the theory is based on a statistical structure, and the data that actually appears is subject to variation. Therefore, in order to determine the parameters of the curve, it is desirable to be able to use not only new data that is considered to be sufficiently composed of late users, but also data immediately after the reversal point. It is also important to know the time of reversal in order to make predictions as early as possible.

  On the other hand, it appears that the reversal point coincides with the peak point and can be determined by the maximum point of the time series data for the increment. However, in reality, since there are deficiencies and very short-term fluctuations in the collection of time series data, the method using the time series of the increase causes a malfunction that causes a completely different time point to be the reverse time point. Therefore, it is necessary to devise a method for determining the reversal point and the parameter value based on the accumulated time series with a small error instead of the incremental time series in order to make an analysis and prediction method with a small error including these.

This problem can be solved by a structural theory of demand growth. In fact, according to the theory as described above, in the vicinity of the reversal point, the demand growth curve of logarithmic or negative integral function changes as the sum of the decrease in the preceding type of use initiator and the increase in the late type of use starter. The time of reversal that can be used to determine the parameters of this is nothing but the point of time when it can be considered that the prior-type user has disappeared.
Therefore, when the demand growth curve is displayed with a one-sided logarithm or a one-sided negative integral function only on the time axis, it is a downward convex curve that increases monotonically with the passage of time in the interval up to the time of reversal. It becomes a straight line that increases monotonously with the passage of time with a slope of. FIG. 3 is an explanatory diagram showing a demand growth curve after time axis conversion. For this reason, the regression line for the demand growth curve on the right half straight line at each time point from before the reversal point increases the slope when increasing each point, and matches the linear component of the demand growth curve after the reversal point. To do. Therefore, by constructing a regression line group from time-series data, it is possible to determine a demand growth curve composed of the time of reversal and the late-type user.

[Configuration of the embodiment]
Next, with reference to FIG. 1, the structure of the demand prediction apparatus concerning one embodiment of this invention is demonstrated in detail.
The operation input unit 11 includes an operation input device such as a keyboard and a mouse, and has a function of detecting various operations such as a demand prediction process start operation and a data input operation by an operator and outputting them to the arithmetic processing unit 20. .
The screen display unit 12 includes a screen display device such as an LCD or a PDP, and has a function of displaying a processing menu from the arithmetic processing unit 20 or an output result such as a demand prediction value or a demand prediction graph.
The main storage unit 13 includes a storage device such as a memory and a hard disk, and has a function of storing a default value (initial value) and a program for processing data such as parameters and designation requests used for processing in the arithmetic processing unit 20. Yes.

The time-series data storage unit 14 includes a storage device such as a memory or a hard disk, and the number of actual purchases (sales) for each region of a product such as a durable product, system, or service for a user that is a target of demand prediction. A function of storing time-series data indicating a temporal change in (number).
The processing data recording unit 15 includes a storage device such as a memory or a hard disk, and generates various processes such as converted time-series data, regression line function expressions, convergence determination values, and data evaluation values that are generated when the demand prediction process is executed. It has a function to record data.

The output data storage unit 16 includes a storage device such as a memory and a hard disk, and has a function of storing output results such as a demand prediction value and a demand prediction graph obtained by the demand prediction processing.
The data input / output unit 17 includes a dedicated communication interface circuit. The data input / output unit 17 is connected to an external device via a communication line to perform data communication, and reads time series data from the external device and stores it in the time series data storage unit 14. A function for reading the output result from the output data storage unit 16 and outputting the result to external data, and a function for reading a program from an external device and storing it in the main storage unit 13.

  The arithmetic processing unit 20 includes a microprocessor such as a CPU and its peripheral circuits, and reads the program from the main storage unit 13 and executes it to execute the demand prediction process in cooperation with the hardware and the program. It has a function to realize various functional units for the purpose. These functional units correspond to the central control unit 21, the default setting unit 22, the demand growth curve specifying unit 23, the predicted value calculation unit 24, and the data evaluation unit 25 described above.

The central control unit 21 has a function of controlling the flow of commands, messages, and data from each unit.
The default setting unit 22 has a function for setting default values (initial values) for processing data such as various parameters and designation requests of each unit in the demand forecasting process, and various types of display, input, and file storage for the default value setting request table. And a function of executing processing. Here, the initial time point indicating the start of demand growth, the setting of various convergence condition values (thresholds) used for convergence determination, the regression line family calculation time interval, the distance evaluation time interval for evaluating the distance of the regression line, the negative index, Default values such as presence / absence of negative value tuning and negative value tuning scenario are set.

  The demand growth curve specifying unit 23 has a function of reading a time series data from the time series data storage unit 14 and performing a regression analysis process on the time series data, thereby specifying a demand growth curve F (t) related to the product. Yes. At this time, as the demand growth curve, a function is used in which the increase in the number of purchased products per unit time is proportional to the negative length of time from a predetermined reference time (demand growth start time).

  The predicted value calculating unit 24 calculates a demand predicted value of a product at an arbitrary time based on the demand growth curve calculated by the demand growth curve specifying unit 23, and a natural scale or negative integral function according to an operator request. The demand growth curve and time-series data are plotted on both time axis scales, and a demand growth curve is created for each regression line read from the processing data recording unit 15 in response to an operator request. It has a function of calculating the deviation of the prediction curve and a function of displaying various data obtained by these functions on the screen display unit 12 or storing them in the output data storage unit 16 as a prediction result.

  Based on the threshold set by the default setting unit 22, the data evaluation unit 25 has a function of comparing the demand forecast value in the section where the demand growth curve forms a straight line and the actual purchase value of the time-series data. Based on the comparison result, the divergence status of the two is determined, and the function for outputting the divergence state based on the time zone and the divergence value, and various data to the screen display unit 12 based on the divergence factor considered according to the divergence state. It has a function of displaying on the screen or storing it in the output data storage unit 16 as a prediction result. At this time, if the demand forecast value indicates a displacement in a direction below the actual purchase value, a warning indicating a qualitative change of the product is output, and if the demand forecast value indicates a displacement in a direction above the actual purchase value, Output a warning indicating missing time series data.

  The demand growth curve specifying unit 23 includes a time axis conversion unit 31, a regression line sequence calculation unit 32, a convergence determination unit 33, and a negative saddle value tuning unit 34 as main processing function units. In addition, the convergence determination unit 33 includes an angle cosine value calculation unit 33A, a straight line distance calculation unit 33B, and a Cauchy type determination unit 33C as main function processing units.

The time axis conversion unit 31 has a function of performing time axis conversion of time series data based on the negative index set by the default setting unit 22 and the initial time point.
The regression line sequence calculation unit 32 is a function of the regression line for the data time point sequence in the regression line group calculation time interval set by the default setting unit 22 among the converted time series data obtained by the time axis conversion unit 31. It has a function of calculating a sequence of equations and their error evaluation values.

The convergence determination unit 33 has a function of determining whether or not the regression line sequence obtained by the regression line sequence calculation unit 32 has converged to a predetermined convergence line based on a predetermined convergence condition.
The angle cosine value calculation unit 33 </ b> A has a function of calculating an angle cosine value between any two regression lines based on the regression line function equation obtained by the regression line sequence calculation unit 32.

The distance calculation unit 33B between straight lines has a function of calculating the minimum distance between the regression lines based on the distance evaluation time interval set by the default setting unit 22.
Based on the various convergence condition values set by the default setting unit 22, the Cauchy type determination unit 33C calculates the angle cosine value obtained by the angle cosine value calculation unit 33A and the distance between straight lines calculation unit 33B and the minimum distance between straight lines. And a function for executing a Cauchy type determination, and a function for calculating a numerical sequence of convergence status, a convergence determination, an inversion time existing section, and a convergence straight line based on the determination result.

  The negative value tuning unit 34 is activated when the negative value tuning execution set by the default setting unit 22 is set to execute, and tunes the negative index using the error evaluation value of the convergence straight line. Has the function to perform. At this time, there are a scenario to be executed based on the gradient method with a differential value, and a scenario in which the exponent value is calculated for each set step value and takes the minimum value within the range, and is set by the default setting unit 22 Scenario.

[Operation of this embodiment]
Next, with reference to FIG. 4, operation | movement of the demand prediction apparatus concerning one embodiment of this invention is demonstrated. FIG. 4 is a flowchart showing a demand prediction process of the demand prediction apparatus according to the embodiment of the present invention. Here, it is assumed that time-series data indicating the actual purchase number (sales number) of a product to be predicted for demand is stored in advance from the external device to the time-series data storage unit 14 via the data input / output unit 17. To do.

The arithmetic processing unit 20 starts the demand prediction process of FIG. 4 when an operation indicating the start of the demand prediction process is detected by the operation input unit 11.
The arithmetic processing unit 20 first sets a default value (initial value) for processing data such as various parameters and designation requests of each unit in the demand prediction process by the default setting unit 22 (step 100).
Next, the arithmetic processing unit 20 reads the time series data from the time series data storage unit 14 by the time axis conversion unit 31 of the demand growth curve specifying unit 23, and sets the negative power index and the initial time point set by the default setting unit 22. Based on the above, time-axis conversion of time series data is performed (step 101).

In this case, when the actual purchase number at an arbitrary time point t (i) is N (i), the time series data D0 is expressed by the following equation (1). I is an integer of i = 1, 2,... N. Specific examples of the time series data D0 include quarterly time series data regarding the number of users who have contracted any one of a certain contract-based service category in a certain prefecture. If the number of users increases in a specific season, a time series for each season may be taken.

The demand growth curve is a curve that monotonously increases with the passage of time in the section before the reversal point, and uses a function that becomes a straight line that monotonously increases with the passage of time in the section after the reversal point. Here, it is assumed that the increase in the number of purchased products per unit time is a function that is proportional to the negative of the length of time from a predetermined reference time (demand growth start time). Specifically, the increase number may be a function proportional to the reciprocal of the time length from the reference time point. When the negative integral function is f (t), the converted time series data D1 obtained by the time axis conversion unit 31 is expressed by the following equation (2). For example, when the exponent a of 冪 is −1, the negative 冪 integral function f (t) is a logarithmic function of time t. Also, the case of a general negative powers 1 / (t ^a), may be utilized negative power of integration as shown in the following equation (3) instead of the logarithmic function.

Next, the arithmetic processing unit 20 uses the regression line set calculation unit 32 of the demand growth curve specifying unit 23 to set the regression set by the default setting unit 22 among the converted time series data obtained by the time axis conversion unit 31. For the data point sequence within the straight line group calculation time interval, a regression line function equation and its error evaluation value sequence are calculated (step 102).
FIG. 5 is an explanatory diagram illustrating a regression line sequence calculation process. Here, the regression line L (1) for each of the data time points t (1) to t (3) in the regression line family calculation time interval (convergence determination interval) set by the default setting unit 22 is shown. ~ L (3) is determined.

  At this time, for the time point t (i), the data D1 (i) after the time point t (i) is extracted from the converted time series data D1, and the functional equation of the regression line L (i) and the mean square error er2 are taken. (I) is calculated. For example, assume that the regression process minimizes the square error, and the straight line L (i) is parametrized by its intercept and slope. The mean square error er2 (i) is a value obtained by dividing the error value from the regression line data by the number of entries of the data D1 (i) used for the regression, that is, the concentration of the set D1 (i).

Next, the arithmetic processing unit 20 uses the convergence determination unit 33 of the demand growth curve specifying unit 23 to determine whether or not the regression line sequence obtained by the regression line sequence calculation unit 32 converges to a predetermined convergence line based on a predetermined convergence condition. Are determined (steps 103 to 107).
At this time, the convergence evaluation value is defined, for example, by the sum of the cosine value of the angle formed between the straight lines and the minimum distance between the straight lines in the distance evaluation time interval of interest set by the default setting unit 22. If the time-series data is in accordance with the structural theory of demand growth described above, the straight line gradually approaches a straight line, that is, a convergence straight line while increasing the slope, and becomes a convergence straight line at the time of reversal.

Actually, in the convergence determination unit 33, the angle between any two regression lines based on the functional equation of each regression line L (i) obtained by the regression line sequence calculation unit 32 by the angle cosine value calculation unit 33A. A cosine value cos (θ) is calculated (step 103), and the minimum distance between the regression lines is calculated by the distance calculation unit 33B based on the distance evaluation time interval set by the default setting unit 22 (step 104). .
FIG. 6 is an explanatory diagram showing a convergence evaluation value calculation process. Here, the angle cosine value cos (θ) of the angle θ formed by the regression lines La and Lb is calculated based on the functional expression of the regression lines La and Lb, and the regression lines La and Lb in the distance evaluation time interval S are calculated. The minimum distance dmin is calculated.

Subsequently, a convergence evaluation value is calculated from the sum of the angle cosine value and the minimum distance by the Cauchy type determination unit 33C (step 105). Based on various convergence condition values set by the default setting unit 22, each regression line is calculated. A Cauchy-type determination for the convergence evaluation value of L (i) is executed (step 106). As a Cauchy type determination method, a general known method may be used.
Here, when a regression line L (i) satisfying the convergence condition value is found within the regression line group calculation time interval (convergence determination interval) set by the default setting unit 22, the minimum time t (j) is reversed. As the time tx, a numerical sequence of the convergence state at this time, convergence determination, reverse rotation time existence section, and convergence straight line are calculated, and these determination results are stored in the output data storage unit 16 (step 107). This convergence straight line is used as the demand growth curve G (t) in the straight line function section Sb after the reversal time tx, that is, as a prediction curve used for demand prediction in the medium to long term.

Next, when the negative value tuning execution set by the default setting unit 22 is set to be executed, the arithmetic processing unit 20 uses the error evaluation value of the convergence straight line by the negative value tuning unit 34. Based on whether or not the mean square error of the convergence straight line is minimum, it is determined whether or not the negative power index needs to be tuned (step 108).
Here, the power index a of the negative power integral function is changed in the vicinity of −1 for each step value set by the default setting unit 22, and the power index a that minimizes the mean square error of the convergence straight line is selected. And a scenario based on a gradient method using a differential value obtained by differentiating the error evaluation value of the regression line with respect to the power index a, and a scenario set by the default setting unit 22 is used.

If it is determined that tuning of the negative power index is necessary (step 108: YES), after adjusting the power index a (step 109), the process returns to step 101, and the process using the adjusted power index a (Steps 101 to 108) are repeatedly executed.
Further, when it is determined that tuning of the negative index is unnecessary (step 108: NO), the arithmetic processing unit 20 is based on the demand growth curve calculated by the demand growth curve specifying unit 23 by the predicted value calculation unit 24. A demand forecast value of the product at an arbitrary time is calculated (step 110).

Here, for the linear function section Sb after the reversal time tx, the function formula of the convergence straight line stored in the output data storage unit 16 shows the demand growth curve G (t) after the time axis conversion. A function inversely converted to natural time (natural scale), that is, a demand growth curve F (t) is used for demand prediction.
In particular,

  Next, based on the threshold set by the default setting unit 22 by the data evaluation unit 25, the arithmetic processing unit 20 calculates the demand forecast value and the actual purchase value of the time series data in the section where the demand growth curve forms a straight line. Are determined based on the obtained comparison result, and the divergence state is output by the time zone and the divergence value (step 111).

  Further, the data evaluation unit 25 displays a warning on the divergence factor considered according to the divergence state on the screen display unit 12 based on the priority (step 112), and ends the series of demand prediction processes. At this time, if the demand forecast value indicates a displacement in a direction below the actual purchase value, a warning indicating a qualitative change of the product is output, and if the demand forecast value indicates a displacement in a direction above the actual purchase value, Output a warning indicating missing time series data.

If there is sufficient market size and sufficient persistence for continued use of the product, the regression line L (i) will usually have a high coefficient of determination value, or sufficient, leaving sufficient data size for D1 (i) It converges to a regression line with a small mean square error.
Despite the tendency to converge, if the time of reversal exceeds the range that can be assumed in advance, or the error or coefficient of determination does not decrease, data loss, product quality changes, and ultimately demand It means the failure of the growth theory hypothesis. Therefore, a cause that is assumed to be a time series having a large error from the regression line is warned and output.

On the other hand, when the regression line has a small error near the time of reversal and the error is large thereafter, the cause can be output with a high possibility. In other words, if there is a tendency to exceed the regression line, it is highly possible that the product is qualitatively changed.
In addition, when the middle board is below the regression line and the deviation tends to return, there is a possibility of data loss. In addition, if there is a large change in price, system, society, etc., this will be prioritized.

  FIG. 7 is an explanatory diagram showing a demand prediction result. This demand prediction curve shows the change in the number of subscribers to the broadband service, the curve 51A shows that predicted by the data up to 2003 by the conventional method (Non-Patent Document 1), and the curve 51B shows These are predicted with data up to the time point of 2004 by a conventional method. A curve 52A indicates the data predicted up to the year 2003 by the method of the present embodiment, and the curve 52B is an image predicted by the data up to the year 2004 by the method of the present embodiment. The demand forecast curve of the conventional method gradually deviates from the actual measurement value after the forecast time point, and shows a tendency that the demand forecast error increases. Even when it is aged, it shows almost the same change as the actual data, and it can be seen that it can be predicted with sufficient accuracy in the medium to long term.

[Effects of the present embodiment]
As described above, in the present embodiment, the time series data storage unit 14 indicates the time change of the actual purchase number related to the product and stores the time series data including the time of reversal. The time-series data is read from the time-series data storage unit 14, the time-axis conversion process and the regression analysis process are performed on the time-series data, thereby specifying the demand growth curve of the product after the reverse time point, and the predicted value calculation unit 24 Based on the demand growth curve, calculate the demand forecast value of the product at any time after the reversal point, and as the demand growth curve, the increase in the number of product purchases per unit time is the time from the predetermined reference time Since a function proportional to the negative of length is used, demand forecasts based on the structural theory of demand growth can be realized for demand related to goods or markets. It can be predicted with sufficient accuracy in the.

  Therefore, in the planning and evaluation of development / capital investments, policies, measures, etc., such as companies, public organizations, government agencies, etc. Knowing the contents of theoretical hypotheses and support, and understanding the risks, you can use medium- to long-term forecasts. As a result, it is possible to avoid an excessive or shortage of investment and formulate an appropriate investment plan or funding plan to obtain a high industrial or economic effect.

  In this embodiment, the demand growth curve may use a function in which the number of product purchases increased per unit time is proportional to the reciprocal of the time length from the reference time point, with sufficient accuracy in the medium to long term. Can be predicted.

In the present embodiment, as the demand growth curve, when the vertical axis indicates the number of purchases of the product and the horizontal axis indicates the logarithm or the negative integral function time axis, the time in the section before the predetermined reversal point You may use the function which becomes a curve which increases monotonously with progress, and becomes a straight line which monotonously increases with progress of time in the section after a reversal point.
As a result, if there is sufficient market size and sufficient persistence for continuous use of the product, it is determined whether the forecast point exceeds the reversal point. It can be calculated and displayed based on the theory of growth.

  In addition, the regression method based on the method of converting the time axis of the accumulated demand curve that makes use of the characteristics of the theoretical curve provides the inversion time estimation with low error and high stability, and the parameter value of the prediction curve. Therefore, it can be applied not only to nationwide demand growth with a large amount of data but also to analysis and prediction of demand growth that is small and prone to error, such as by prefecture. In particular, in a relatively small area, there is a unique attribute for each area, so that a blur is likely to occur with respect to the national demand curve, and the blur appears more prominently when the amount of data is small. According to the present embodiment, even in a prediction target area where such a shake is likely to occur, a demand growth curve that appropriately captures the shake can be specified, and extremely high prediction accuracy can be obtained.

  Also, based on the theory of demand growth, the hypothesis situation from which the risk of prediction is generated is clear, and the user of this method or device examines the hypothesis presented in the theoretical explanation and its support. By doing so, it will be possible to understand the risks contained in the predicted values and use them while judging whether or not to accept them. The demand growth theory on which this system is based is a good explanation of at least the demand growth of broadband access services in Japan and the growth situation of mobile phones in a range where the quality of services does not change significantly. Accompanying.

In the present embodiment, the data evaluation unit compares the demand forecast value obtained by the forecast value calculation unit with the actual purchase value of the time series data, and based on the comparison result, A warning indicating a lack of time-series data may be output. Specifically, in the section where the demand growth curve forms a straight line, the demand forecast value obtained by the forecast value calculation unit is compared with the actual purchase value of the time series data, and the demand forecast value is below the actual purchase value. In the section where the demand growth curve forms a straight line, the demand forecast value obtained by the forecast value calculation unit and the actual purchase value of the time series data are output. In comparison, when the demand forecast value indicates a displacement in a direction exceeding the actual purchase value, a warning indicating a lack of time-series data may be output.
This makes it possible to warn when there is a possibility of data loss or product qualitative changes based on demand growth theory.

It is a block diagram which shows the structure of the demand prediction apparatus concerning one embodiment of this invention. It is explanatory drawing which shows the demand growth curve regarding broadband service. It is explanatory drawing which shows the demand growth curve after time-axis conversion. It is a flowchart which shows the demand prediction process of the demand prediction apparatus concerning one embodiment of this invention. It is explanatory drawing which shows a regression line sequence calculation process. It is explanatory drawing which shows a convergence evaluation value calculation process. It is explanatory drawing which shows a demand prediction result.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 ... Demand prediction apparatus, 11 ... Operation input part, 12 ... Screen display part, 13 ... Main memory part, 14 ... Time series data storage part, 15 ... Process data recording part, 16 ... Output data storage part, 17 ... Data entry Output unit, 20 ... arithmetic processing unit, 21 ... central control unit, 22 ... default setting unit, 23 ... demand growth curve specifying unit, 24 ... predicted value calculation unit, 25 ... data evaluation unit, 31 ... time axis conversion unit, 32 A regression line sequence calculation unit, 33 a convergence determination unit, 33 A an angle cosine value calculation unit, 33 B a straight line distance calculation unit, 33 C a Cauchy type determination unit, and a negative saddle value tuning unit.

Claims (16)

Demand that has a storage unit and an arithmetic processing unit, and predicts the demand after the reversal point, where the demands of the predecessor and the successor forming the demand are reversed among the demands of the product in the market based on the actual number of purchases related to the product A demand prediction method used in a prediction device,
A time-series data storage step for storing time-series data indicating a temporal change in the number of actual purchases related to the product and including the reversal time by the storage unit;
By reading out the time series data from the time series data storage unit and performing time axis conversion processing and regression analysis processing on the time series data by the arithmetic processing unit, the demand growth curve of the product after the reversal point Demand growth curve calculating step for identifying
A demand prediction step of calculating and outputting a demand forecast value of the product at an arbitrary time after the reversal time based on the demand growth curve by the arithmetic processing unit;
The demand growth curve includes a function in which an increase in the number of purchases of the product per unit time is proportional to a negative length of a time length from a predetermined reference time point. The demand forecasting method according to claim 1,
The demand forecasting method, wherein the demand growth curve comprises a function in which the increase number is proportional to the reciprocal of the time length from the reference time point. The demand forecasting method according to claim 2,
The demand growth curve is a curve that monotonically increases with the passage of time in the section before the reversal point when the vertical axis indicates the number of purchases of the product and the horizontal axis indicates the time axis of the logarithmic or negative integral function. And a demand prediction method comprising a function that is a straight line that monotonously increases with the passage of time in a section after the reversal point. In the demand prediction method according to claim 3,
The arithmetic processing unit compares the demand forecast value obtained in the forecast value calculation step with the actual purchase value of the time series data, and based on the comparison result, the qualitative change of the product or the time series data A demand prediction method, further comprising a data evaluation step of outputting a warning indicating a deficiency of the data. In the demand prediction method according to claim 3,
The calculation processing unit compares the demand forecast value obtained in the forecast value calculation step with the actual purchase value of the time-series data in a section where the demand growth curve forms a straight line, and the demand forecast value is the actual record. The demand prediction method characterized by further comprising the data evaluation step which outputs the warning which shows the qualitative change of the said product, when showing the displacement to the direction which is less than a purchase value. In the demand prediction method according to claim 3,
The calculation processing unit compares the demand forecast value obtained in the forecast value calculation step with the actual purchase value of the time-series data in a section where the demand growth curve forms a straight line, and the demand forecast value is the actual record. A demand prediction method, further comprising a data evaluation step of outputting a warning indicating a loss of the time series data when a displacement in a direction exceeding a purchase value is indicated. In the demand prediction method according to claim 3,
The demand growth curve calculation step includes:
A time axis conversion step of converting the time axis of the time series data to a logarithmic or negative integral function time axis;
A regression line sequence calculating step for obtaining a regression line of converted time series data after an arbitrary start time among the converted time series data obtained in this time axis conversion step, for each of a plurality of start time columns,
A convergence determination step for determining whether or not the regression line sequence obtained in the regression line sequence calculation step has converged to a predetermined convergence line based on a predetermined convergence condition;
The demand forecasting method characterized in that the minimum start time determined as having converged in the convergence determination step is the reverse time point, and the convergence straight line is a function indicating the demand growth curve in the section after the reverse time point. In the demand prediction method according to claim 3,
The convergence condition includes a tolerance value of an angle cosine value between any two regression lines and a tolerance value of a distance between these regression lines in a predetermined time axis section. Based on the actual number of purchases related to the product, a demand prediction device that predicts the demand after the reversal point where the demand of the first and second generations forming the demand out of the demand of the product in the market is reversed,
A time-series data storage unit that indicates a temporal change in the number of actual purchases related to the product and stores time-series data including the reversal time;
A demand growth curve that identifies the demand growth curve of the product after the time of reversal by reading out the time series data from the time series data storage unit and performing time axis conversion processing and regression analysis processing on the time series data A specific part,
A predicted value calculation unit that calculates and outputs a demand predicted value of the product at an arbitrary time after the reversal time based on the demand growth curve,
The demand forecasting apparatus, wherein the demand growth curve comprises a function in which an increase in the number of purchases of the product per unit time is proportional to a negative length of time from a predetermined reference time point. In the demand prediction device according to claim 9,
The demand forecasting apparatus, wherein the demand growth curve comprises a function in which the increase number is proportional to the reciprocal of the time length from the reference time point. In the demand prediction device according to claim 10,
The demand growth curve is a curve that monotonically increases with the passage of time in the section before the reversal point when the vertical axis indicates the number of purchases of the product and the horizontal axis indicates the time axis of the logarithmic or negative integral function. And a demand prediction device comprising a function that is a straight line that monotonously increases with time in a section after the reversal point. The demand prediction device according to claim 11,
The demand forecast value obtained in the forecast value calculation unit is compared with the actual purchase value of the time series data, and a warning indicating a qualitative change of the product or a loss of the time series data based on the comparison result A demand prediction device further comprising a data evaluation unit for outputting. The demand prediction device according to claim 11,
In the section where the demand growth curve forms a straight line, the demand forecast value obtained by the forecast value calculation unit is compared with the actual purchase value of the time series data, and the demand forecast value is in a direction lower than the actual purchase value. A demand prediction device further comprising a data evaluation unit that outputs a warning indicating a qualitative change of the product when the change is indicated. The demand prediction device according to claim 11,
In a section where the demand growth curve forms a straight line, the demand forecast value obtained by the forecast value calculation unit is compared with the actual purchase value of the time series data, and the demand forecast value exceeds the actual purchase value. A demand prediction device further comprising a data evaluation unit that outputs a warning indicating a loss of the time-series data when the displacement is indicated. The demand prediction device according to claim 11,
The demand growth curve specifying unit is:
A time axis converter that converts the time axis of the time series data to a logarithmic or negative integral function time axis;
A regression line sequence calculating unit for obtaining a regression line of converted time series data after an arbitrary start time among the converted time series data obtained by the time axis conversion unit for each of a plurality of start time points;
A convergence determination unit that determines whether or not the regression line sequence obtained by the regression line sequence calculation unit converges to a predetermined convergence line based on a predetermined convergence condition;
The demand start device characterized in that the minimum start time determined to be converged by the convergence determination unit is the reversal time, and the convergence straight line is a function indicating the demand growth curve in the section after the reversal time. The demand prediction device according to claim 11,
The convergence condition includes a tolerance value of an angle cosine value between any two regression lines and a tolerance value of a distance between these regression lines in a predetermined time axis section.

Images