JP2007316745A - Server and program for predicting number of visitors - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To obtain market population of a store and its change without using a map information system, and to predict the number of visitors, based on the information. <P>SOLUTION: The number of the visitors in the past stored in time series is acquired (S203) to calculate an average number of the visitors during a prescribed period before a target date, based on the acquired number of the visitors in the past (S207, S209, S212, S214, S216). <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a store visitor number prediction server and a store visitor number prediction program for predicting the number of store visitors.

Conventionally, the population of the store trade area has been used as a factor for predicting the number of customers visiting the store. The commercial area population has been acquired from a map information system that combines population data issued by government offices and map information with roads, buildings, household population, and the like. For example, the store's trade area is determined to be 500 m from the store, and the population within the trade area is acquired from the map information system (see, for example, Patent Document 1).
JP 2002-324160 A

  However, it takes time and money to build a map information system to obtain a trade area population. Also, because the trade area population changes with time, it lacks accuracy.

  The present invention has been made in view of the above circumstances, and its purpose is to grasp the store's commercial area population and changes in the commercial area population without using a map information system, and predict the number of customers based on that information. The purpose of the present invention is to provide a store visitor number prediction server and a store visitor number prediction program for calculating the average number of visitors.

  The present invention is a store visitor number prediction server that acquires the number of actual store customers stored in a time series in a day unit, and calculates the average store visitor number for a predetermined period before the prediction target date from the acquired actual store customer number.

  According to the present invention, it is possible to provide a store visitor number prediction server and a store visitor number prediction program for grasping a store's commercial area population and changes in the commercial area population without using a map information system and predicting the number of store visitors based on the information.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a diagram schematically showing a network system. As shown in FIG. 1, the network system includes a store server 1 that is a store visitor number prediction server for predicting the number of store visitors, and a plurality of POS terminal devices 2 connected to the store server 1 via the network 3. Is done. The network 3 includes a LAN (local area network), a WAN (wide area network), and the Internet. In a store where only one or a few POS terminal devices 2 are provided and the store server 1 is not installed, the store server 1 is not provided, and one of the POS terminal devices 2 has the function of the store server 1. You may do it.

  FIG. 2 is a block diagram showing a main configuration of the store server 1. As illustrated in FIG. 2, the store server 1 includes a CPU 11, a ROM 12, a RAM 13, a communication control unit 14, a recording unit 15, a display unit 16, an input unit 17, and a clock unit 18. CPU11 controls the store server 1 whole. The ROM 12 holds a program for controlling the store server 1. The RAM 13 has a work area that is necessary when the CPU 11 executes a program held in the ROM 12. The communication control unit 14 controls data transmission / reception with the POS terminal device 2 and the like performed via the network. The recording unit 15 records (saves) various data. The recording unit 15 records, for example, the number of customers, date, day of the week, precipitation amount during business hours described later, and customer number data indicating a specific day obtained from POS data acquired when a POS data acquisition process described later is executed. The The display unit 16 performs display on a display or the like. The input unit 17 performs input processing from a mouse, a keyboard, or the like. The clock unit 18 measures time and generates date / time information.

  FIG. 3 is a block diagram illustrating a configuration of a main part of the POS terminal device 2. As illustrated in FIG. 3, the POS terminal device 2 includes a CPU 21, a ROM 22, a RAM 23, a communication control unit 24, a recording unit 25, a display unit 26, an input unit 27, a barcode connection unit 28, and a clock unit 29. . The CPU 21 comprehensively controls the POS terminal device 2 as a whole. The ROM 22 holds a program for controlling the POS terminal. The RAM 23 has a work area that is necessary when the CPU 21 executes a program held in the ROM 22. The communication control unit 24 controls data transmission / reception with the store server 1 performed via the network. The recording unit 25 records (saves) various data. The recording unit 25 records, for example, POS data indicating sales information (sales date / time, day of week, number of customers, purchased products, etc.) for each customer that has been settled by the POS terminal device 2. The display unit 26 performs display on a display or the like. The input unit 27 performs input processing from the keyboard. The barcode connection unit 28 is connected to a barcode reader and inputs barcode data read by the barcode reader. The clock unit 29 measures time and generates date / time information.

  Next, processing when the POS data acquisition program stored in the ROM 12 of the store server 1 is executed will be described. FIG. 4 is a flowchart showing the POS data acquisition process of the store server 1. This POS data acquisition process is a process of periodically acquiring POS data from the POS terminal device 2 connected to the network 3.

  The store server 1 acquires POS data from the POS terminal device 2 via the network 3 periodically, for example, once a day (S101). The timing for acquiring the POS data is a predetermined timing before the start of business, but any timing may be used. As a result, the store server 1 acquires POS data for one day from the time when the previous POS data was acquired until the current time. The store server 1 counts the number of people who have passed through the POS terminal device 2 from the acquired POS data, and calculates the number of customers visiting the store per day (S102).

  Next, the store server 1 records the number of customers calculated in step S102 as the customer number data together with the date and day of the week included in the POS data, the precipitation amount during business hours obtained from the input unit 17, and the specific date described later. The data is stored in the unit 15 (S103).

  Here, the precipitation is obtained through a screen displayed on the display unit 16. The precipitation input from this screen may be entered as a numerical value of actual results, or by inputting data “0”, “1”, “2”, “3” after discretization described later. Alternatively, “no rain”, “light rain”, “weak rain”, or “strong rain”, which replaces the data after discretization with words, may be selected. It will be described later how the precipitation is discretized or replaced with words. Further, the precipitation may be acquired from the communication control unit 14 from a server that provides the daily precipitation on the Internet of the network 3.

  A special day is a day with a different trend in the number of customers, such as New Year holidays, Golden Week, Obon, Christmas, New Year's Eve, and three or more consecutive holidays. Weeks that include special days need to be considered separately from regular weeks because the trend of the whole week also changes. For the data that is a special day, for example, “1” is set for New Year holidays, “2” is set for Golden Week, and “0” is set for data that is not a special day. In addition, it may be possible to set all consecutive holidays such as New Year holidays and Golden Week to “1”, abnormal weather such as a typhoon to “2”, and data that is not a specific day to “0”.

  Here, the specific date can be set through a screen displayed on the display unit 16. On this screen, select the date you want to use as a special day from the calendar or select from the list box. After selecting the specific date in this way, the type of the specific date is further selected from the list box. Then, register a special day with the setting button. The registered contents are recorded in the recording unit 15 of the store server 1.

  FIG. 5 is a table T showing customer number data recorded in the recording unit 15. As shown in FIG. 5, customer data is recorded in a tabular format. In this table T, in correspondence with the date acquired from the POS data, an area for recording the data of the day of the week and the number of customers (number of customers), an area for recording the precipitation data acquired from the input unit 17, in the recording unit 15 There are an area for recording data on the specific date acquired from the specific date set in the calendar, and an area for recording the predicted value of the number of customers visiting the store. It is desirable to retain the customer number data recorded in this table T for at least one year before the current date.

  Next, a method of using multiple regression analysis for predicting the number of customers visiting the store will be described. Multiple regression analysis is a generally well-known statistical method for creating a prediction model (multiple regression equation) from the relationship between an objective variable and explanatory variables, and analyzing the influence between multiple explanatory variables and objective variables. The details of the multiple regression analysis are described in “Regression Analysis Understandable by Illustration” by Yoshiyuki Sakurai and Sadami Sakurai; Note that it is not always necessary to use multiple regression analysis to create a prediction formula.

Various factors can be considered to explain the change in the number of customers visiting the store. In the present embodiment, day of the week, trade area population, and weather are used.
Next, details of the objective variable and each explanatory variable will be described.

  First, the number of customers on the forecast date is used as the objective variable. Next, explanation variables will be described. Visitors tend to visit periodically on certain days of the week. In general, the number of customers visiting on holidays is higher than on weekdays. On the other hand, at the store in front of the station, there are many shoppers on the way home from work. On the contrary, the number of customers on weekdays may be larger than that on holidays. In addition to stores that sell on a specific day of the week, a specific visitor pattern can be created for each day of the week. Therefore, the day of the week is adopted as the explanatory variable. This makes it possible to express a tendency for each day of the week. Day of the week is a dummy variable. Here, dummy variables are provided for a certain variable by the number of types that the variable can take, and the newly provided variables are represented by “0” and “1”.

  The trade area population is useful information for determining the approximate number of customers visiting the store. As a trade area population, it is possible to use a trade area population obtained from a map information system that combines road information, buildings, household population, etc. with map information, but the disadvantage is that it takes cost and time to build a map information system is there. It is also possible to use the population data of the area including the store issued by the government office where the store is located, but this population data does not always match the store trade area, and these population data are updated monthly. Therefore, it has a drawback that it cannot capture the latest upward / downward trend within one month of forecasting.

  Therefore, the past visitor number stored in the table T of the recording unit 15 of the store server 1 as customer number data is used as an explanatory variable instead of the commercial area population. Since the number of customers in the store has a period of seven days (one week), the average from the previous day to the seventh day before the target date of the objective variable is used. In this way, by using the average of the number of store visitors from the day before the forecast date to 7 days before, useful information for determining the approximate number of store customers replacing the population in the commercial area can be obtained. It is also possible to capture the latest trend of rising and falling customer numbers.

  Here, an arithmetic average is used as the average value, but a weighted average may be used. For example, the number of customers on the same day of the week seven days ago is considered to be related to the predicted day more than other days of the week, so the average value is obtained with a greater weight than other days. In this case, the weight is determined based on the user's experience.

The weather is also an important factor in fluctuations in visitor numbers. In particular, the impact of rain is large, and if it rains during business hours, the number of customers will generally decrease. The more rain there is, the more the number of visitors will fluctuate. However, it does not directly affect the number of customers in proportion to precipitation. For example, doubling the precipitation from 2 mm to 4 mm does not halve the number of customers visiting the store. Therefore, precipitation is discretized by dividing it within an appropriate range (that is, dividing it unevenly). As an example, the precipitation is divided into four stages: no rain, light rain, light rain, and heavy rain. Each range is set to 0 mm, 1 to 5 mm, 6 to 15 mm, 16 mm or more, and numerical values such as “0”, “1”, “2”, and “3” are assigned in order from the smallest. The discretized data is used as an explanatory variable. By discretizing precipitation, it is possible to make a prediction that properly considers the effect of precipitation on customer visits. A method other than the above four categories may be used, and the discrete value “1” may be defined as precipitation that does not require an umbrella. As described above, the user of the store server 1 may select a word corresponding to the precipitation range using the input unit 17 and input the precipitation,
From the above, the target variable uses the number of customers on the forecast day, and the explanatory variable uses the day of the week, the average number of customers on the seven days before the forecast date, and the amount of precipitation. Other factors that may affect the number of customers visiting the store include regional events, advertisements, and trends in competing other stores, and may be expanded to a multiple regression equation that takes these factors into account.

  Next, a method for predicting the number of store customers executed by the store server 1 will be described. FIG. 6 is a flowchart showing store visitor number prediction processing for predicting the store visitor number. This process is started when an instruction for starting the process is given by the input unit.

  The store server 1 acquires the predicted date input by operating the input unit 17 by the user from the screen for executing the prediction process displayed on the display unit 16 (S201), and acquires the predicted precipitation amount on the predicted date (S201). S202). Here, the forecast precipitation is calculated by the store server 1 when the user inputs the digitized data “0,“ 1 ”,“ 2 ”,“ 3 ”on the screen for executing the prediction process. Predicted precipitation may be obtained, or by selecting and inputting “no rain”, “light rain”, “weak rain”, “strong rain”, which replaces the digitized data with words The store server 1 may acquire the predicted precipitation amount, and the predicted precipitation amount may be acquired from the communication control unit 14 from a server that provides the predicted precipitation amount existing on the network 3.

  Next, the store server 1 acquires customer number data for the past 365 days from the predicted date acquired from the table T of the recording unit 15 (S203, acquisition means). The customer number data acquired at this time is not limited to the customer number data for 365 days.

  Then, the store server 1 determines whether or not the predicted date is a week with a specific date (S204). That is, it is determined whether or not there is a special day in one week from Monday to Sunday including the predicted date. FIG. 7 is a diagram for explaining a case where there is no specific day in one week including the predicted date. As shown in FIG. 7, for example, when the predicted date is October 19, 2005, October 19 (Wednesday) is not a special day, and furthermore, October 17 which is a week including October 19 (Monday) October 18th (Tuesday), October 20th (Thursday), October 21st (Friday), October 22nd (Saturday), October 23rd (Sunday) does not include any special days It is judged how. This determination is made with reference to the customer number data acquired in step S203.

  When the store server 1 determines that the predicted date is not a week with a specific date, that is, when the predicted date is not a specific date and there is no specific date in one week including the predicted date (NO in S204), A multiple regression equation is obtained from the day of the week without a specific day (S205). That is, the store server 1 calculates a multiple regression equation using the customer number data obtained by excluding the specific day and the data of the day of the week not including the specific day within one week from the customer number data acquired in step S203.

  Here, how to obtain a multiple regression equation from a week without a specific day (S205) will be described using the sub-flowchart of FIG. The store server 1 acquires data for a day that is not a specific day from the customer number data acquired in step S203 and data for a day that does not include a specific day within a week for a predetermined number of days from the closest to the predicted execution date (S301). ). In other words, data with a unique day item “0” and data of a week day that does not include a day with a unique day item not “0” within one week are acquired from the table T of the recording unit 15. In addition, it is desirable that the data for the predetermined number of days acquired is 50 days or more (8 weeks or more) in consideration of the accuracy of the created multiple regression equation. Next, the store server 1 calculates the average number of customers for the previous seven days for each number of customers data for a predetermined number of days (S302, calculation means). Here, if there is data removed by the specific day in the previous seven days, the calculation is performed using the data of the previous day. Next, the store server 1 performs multiple regression analysis by using the day of the visitor data for a predetermined number of days, the amount of precipitation, and the average number of customers for the previous seven days as explanatory variables, and the actual value of the number of customers visiting the store as an objective variable to obtain a multiple regression equation (S303). Here, the day of the week is used as a dummy variable, and the precipitation is converted to the discretized data described above.

  When the multiple regression equation is obtained in this way (S205), the store server 1 determines whether the prediction execution date for prediction and the prediction date of the prediction target are not continuous (S206). If the predicted execution date and the predicted date are continuous, that is, if the predicted date is the day after the predicted execution date (NO in S206), the average number of customers for the seven days before the predicted date is obtained and the multiple regression created in step S303 is performed. The forecast value is calculated by substituting the forecast day of the week, the forecasted precipitation amount obtained in step 202, and the average number of customers for seven days before the forecast date into the formula (S207, one of the forecasting means). The prediction result is stored in the table T of the recording unit 15 as a predicted value item of the number of customers data and simultaneously displayed via the display unit 16 (S208). FIG. 9 is a diagram showing an example of the prediction result displayed on the display unit 16 in this way, and shows the prediction result on December 11, 2005 (Sunday). As shown in FIG. 9, the prediction result may be displayed as a line graph of past prediction values together with the prediction values up to that day, or only the prediction values of the prediction date may be displayed.

  On the other hand, if it is determined that the predicted execution date and the predicted date are not consecutive (YES in S206), the predicted number of customers is calculated until the predicted date and the predicted number of customers on the predicted date is calculated (209, 1 of prediction means). One). In other words, since the average number of customers for the previous seven days required to calculate the forecast value for the forecast date cannot be calculated, first the number of customers between the forecast run date and the forecast date is forecasted, and then the forecast run date obtained The number of customers is predicted after calculating the average of the seven days before the forecast date, using the forecast value of the date between the forecast date and the provisional customer number actual value. FIG. 10A is a diagram for explaining a case where the predicted execution date and the predicted date are not continuous. As shown in FIG. 10, since the predicted execution date is October 16, while the predicted date is October 19, the predicted execution date and the predicted date are not continuous.

  Here, the process of step S209 will be described using the sub-flowchart of FIG. First, the store server 1 determines whether or not the next day of the predicted execution date is a predicted date (S401). If the next day of the predicted execution date is not the predicted date (NO in S401), after obtaining the average number of customers for the previous seven days, the number of predicted customers is obtained from the multiple regression equation created in S303, and that value is the provisional customer actual value for that day. (S402). Here, regarding precipitation, it may be “0”, that is, it may not rain, or a new predicted precipitation may be acquired. After the process of step S402, the process returns to the process of step S401 again, and it is determined whether the day following the day obtained by the process of step S402 is the predicted date.

  If the next day of the prediction execution date is prediction (YES in S401), the store server 1 determines the average number of customers for the seven days before the prediction date, including the provisional customer number actual value obtained in step S402, and The number of customers is predicted using the multiple regression equation (prediction equation) created in step S303 (S403). In this way, the number of customers is predicted in order from the next day of the prediction execution date, and the obtained number of customers predicted value is set as a provisional number of customers actual value. Then, the number of customers on the forecast date is predicted by obtaining the average number of customers for the seven days before the forecast date using the provisional customer count actual value.

  In addition, as shown in FIG. 10 (b), even in the case where a specific day is included between the predicted execution date and the predicted date, the number of tentative customers is calculated using the multiple regression equation generated by the data not including the specific date generated in step S303. Find the predicted value.

  The store server 1 stores the prediction result thus obtained in the table T of the recording unit 15 as a predicted value item of the number of customers data, and at the same time displays it through the display unit 16 (S208).

  On the other hand, if the store server 1 determines that the predicted date is a week with a specific date, that is, if the predicted date is a specific date or there is a specific date in one week including the predicted date (YES in S204) The store server 1 acquires from the table T of the recording unit 15 the customer number data in the same week as the singular day of the same type as the singular day of the forecast date or the same kind of singular day of the forecast date, A multiple regression analysis is performed to obtain a multiple regression equation (S210). In addition, FIG. 12 is a figure for demonstrating the case where an estimated date exists in a week containing a specific day or a specific day. FIG. 12A shows a case where the prediction date is a unique day, and FIG. 12B shows a case where the week of the prediction date includes a unique day.

  Here, how to obtain a multiple regression equation from the number of customers on the day of the week including a specific day (S210) will be described using the sub-flowchart of FIG.

  The store server 1 acquires from the customer number data acquired in step S203 all customer number data that includes the same type of unique date as the predicted date or the same type of specific date as the predicted date within the week (S501). That is, singular day data that has the same type of data as the area type of the forecast day, and the day of the week that includes the singular day that has the same type of data as the area type of the forecast day. Get the data. Next, the store server 1 determines whether or not the acquired number of customer data is larger than a predetermined number (S502). If the acquired number of data is less than or equal to the predetermined number (NO in S502), a warning that the number of data is small is issued via the display unit 16 (S503). Here, for example, the predetermined number is 20. If the number of data is small, processing may be terminated after issuing a warning. Further, the number of data may continue at least without giving a warning. Next, the store server 1 calculates the average number of customers for the previous seven days for the customer number data acquired in step S501 (S504). Then, the store server 1 performs a multiple regression analysis using the customer number data acquired in step S501 as day of the week, precipitation, the average number of customers in the previous seven days as explanatory variables, and the actual value of the number of customers acquired in step S501 as an objective variable, A multiple regression equation is obtained (S505). Here, the day of the week is used as a dummy variable, and the precipitation is converted to the discretized data described above.

  When the multiple regression equation is obtained in this way (S210), the store server 1 determines whether the predicted date and the predicted execution date are not consecutive (S211). Here, if the predicted date is continuous with the predicted execution date, that is, if the predicted date is the day after the predicted execution date (NO in S211), the average number of customers for the seven days before the predicted date is obtained and created in step S505. The forecast value is obtained by substituting the forecast day of the week, the discretized forecast precipitation, and the average number of customers for the previous seven days into the multiple regression equation for singular days (S212, one of the forecasting means). The prediction result obtained in this way is stored in the table T of the recording unit 15 as a prediction value item of the customer number data, and at the same time, for example, the display shown in FIG. 9 is performed via the display unit 16 (S208).

  Therefore, for the forecast date that is a specific date or the forecast date that has a specific date within a week, multiple regression using the customer data of the day that has the same type of specific date and the same type of specific date in the week By finding the formula, it is possible to grasp the tendency of the number of customers given by an unusual day on a weekly basis.

  Further, when the predicted execution date and the predicted date are not continuous as shown in FIG. 10A (YES in S211), the store server 1 determines that the week different from the predicted date is between the predicted execution date and the predicted date. Is included (S213). FIG. 14 is a diagram illustrating the relationship between the predicted execution date and the predicted date. FIG. 14A shows a case where the day between the predicted execution date and the predicted date is the same week as the predicted date, and FIG. 14B shows the case between the predicted execution date and the predicted date. The case where there is a week different from the forecast date is shown.

  As shown in FIG. 14A, when there is only the same week as the predicted date between the predicted execution date and the predicted date (NO in S213), the previous seven days necessary for obtaining the predicted value of the predicted date The average number of customers cannot be calculated. First, the number of customers between the forecast execution date and the forecast date is forecasted, and the forecast value of the date between the forecast run date and the forecast date is calculated as the provisional customer actual value. Then, after calculating the average for the seven days before the prediction date, the number of customers is predicted to obtain the predicted number of customers visiting the store (S214, one of the prediction means). Here, the process of step S214 is the same as the process of step S209, and the same process as the sub-flowchart of FIG. 11 is performed, but for the specific day created in step S505 instead of the multiple regression equation created in step S303. The number of customers on the forecast day is predicted using the multiple regression equation. The prediction result obtained in this way is stored in the table T of the recording unit 15 as a prediction value item of the customer number data, and at the same time, for example, the display shown in FIG. 9 is performed via the display unit 16 (S208).

  In the process of S214, for the customer number prediction (S401, S402) in the period between the prediction execution date and the prediction date, the multiple regression equation for the singular day created in step S505 uses the prediction execution date and the prediction date. This is because the days between are in the same week as the singular day and are influenced by the singular day.

  As shown in FIG. 14B, when there is a week different from the predicted date between the predicted execution date and the predicted date (YES in S213), the store server 1 tentatively stores the data of the week not including the predicted date. In order to obtain the predicted value, a multiple regression equation is obtained by using the customer number data obtained by removing the unique day and the weekday data including the unique day from the customer number data acquired in step S203 (S215). Here, the process of step S215 is the same as the process of step S205, and the same process as the sub-flowchart of FIG. 8 is performed. Even if a specific day is included in a week different from the predicted date, a multiple regression equation is obtained using the customer number data excluding the specific day and the day of the week including the specific day.

  When the multiple regression equation is obtained in this way (S215), a process of calculating the predicted number of customers on the predicted date by calculating the predicted number of customers on the predicted date is performed (S216, one of predicting means).

  Here, the process for obtaining the predicted number of customers in step S216 will be described using the sub-flowchart shown in FIG. First, the store server 1 determines whether or not the next day of the predicted execution date is a predicted date (S601). If the next day of the predicted execution date is not the predicted date (NO in S601), it is determined whether or not the next day is the same week as the predicted date (S602). If the next day of the predicted execution date is not the same week as the predicted date (NO in S602), after calculating the average number of customers for the previous seven days, the multiple regression equation created in step S303 (multiple regression equation when there is no specific day) ) To obtain the predicted number of customers, and set the value as the provisional number of customers for the day (S603). Here, regarding precipitation, it may be “0”, that is, it may not rain, or a new predicted precipitation may be acquired. When the process of step S603 ends, the process returns to the process of step S601, and it is determined again whether or not the next day is the predicted date. If the next day of the predicted execution date is the same week as the predicted date (YES in S602), the multiple regression equation (multiple regression equation for a specific day) created in step S505 after obtaining the average number of customers for the previous seven days is obtained. The number of customers is predicted by using (S604). Here, regarding precipitation, it may be “0”, that is, it may not rain, or a new predicted precipitation may be acquired. When the process of step S604 ends, the process returns to step S601 to determine again whether the next day is the predicted date.

  On the other hand, if the next day of the predicted execution date is the predicted date (YES in S601), the average number of customers for the previous seven days including the temporary customer number actual value obtained in the processing of steps S603 and S604 is obtained and then created in step S505. The number of customers is predicted using the multiple regression equation for a specific day (S605). For precipitation, for example, data obtained by discretizing the data acquired in step S202 is used.

  In this way, the number of customers is predicted in order from the next day of the prediction execution date, and the obtained number of customers predicted value is set as a provisional number of customers actual value. Then, the number of customers on the forecast date is predicted using the provisional customer number actual value to obtain the average number of customers for the seven days before the forecast date. The prediction result obtained in this way is stored in the table T of the recording unit 15 as a prediction value item of customer number data, and at the same time, for example, the display shown in FIG. 9 is performed via the display unit (S208).

  Also, if there is a week that includes a specific day between the predicted execution date and the predicted date, if the week that includes the specific day is different from the week that includes the predicted date, the duplicate created with data that does not include the specific day Use the regression formula to calculate the estimated number of customers.

  Here, let us consider a case where the predicted number of customers and the actual value differ greatly. In other words, if the actual number of customers suddenly fluctuates, if the actual number of customers is used for the next forecast, the predicted value after that may be greatly distorted. Better. Therefore, when the average of the previous seven days is calculated from the value of the actual number of customers, if the difference between the actual number of customers and the predicted value is greater than or equal to the threshold specified by the user, the number of customers data for that week is not used for the calculation. Like that. Or, when calculating the average value, the predicted value at that time is used instead of the actual number of customers. Hereinafter, a case will be described in which, when the difference between the actual number of customers and the predicted value is greater than or equal to a threshold specified by the user, the number of customers data for that week is not used in the calculation.

A process performed when the store server 1 records the POS data in the recording unit 15 to compare the predicted value with the actual number of customers will be described with reference to the flowchart of FIG.
Since the processes in steps S701 and S702 are the same as those in steps S101 and S102, description thereof will be omitted. The store server 1 compares the actual number of customers tabulated in step S702 with the predicted value of the same date as the number of customers tabulated in step S702 recorded in the recording unit 15, and thereby the error between the predicted value and the actual number of customers. It is checked whether the absolute value of is not less than a predetermined value (S703). Here, 1000 is used as the predetermined value. That is, data when there is a difference of 1000 or more between the number of customers and the predicted number of customers is not used.

  Next, the store server 1 combines the actual number of customers with the date and day of the week acquired from the POS data, the precipitation amount during business hours acquired from the input unit 17, and the comparison result of step S703, as shown in FIG. The number of customers is stored in the table T ′ in the table format 15 (S704). FIG. 17 is a diagram showing a table T ′ in which error areas are provided corresponding to dates in the table T described in FIG. The comparison result in step S704 is recorded in the error area of the table T ′ as “1” when the error is 1000 or more and “0” when the error is less than 1000, for example.

  Thus, by adding the result of comparing the actual number of customers and the predicted value made in the past to the number of customers data, for example, when the store server 1 acquires the number of customers data from the recording unit 15 in the process of step ST203, an error occurs. Area number “0”, that is, only the number-of-customer data in which the error between the predicted value and the actual number of customers is less than 1000 can be acquired. In addition, when the store server 1 acquires customer data in units of weeks, if there is one error item in the data in the week, it is possible to acquire customer data excluding all data for that week. It becomes possible. Alternatively, the store server 1 may acquire a predicted value instead of actual data when the error area is “1”. Also in this case, when the store server 1 acquires customer number data on a weekly basis, if there is one error area in the data in the week, the prediction value data is acquired for all the weeks. You may do it.

  The reason why the forecast value differs greatly from the actual number of customers is that the precipitation used for the forecast and the actual precipitation differ greatly, or that local events other than those set as special days were held, it was extremely hot The influence such as. If the effect on the number of visitors is a factor for the day, it is necessary to prevent that factor from affecting subsequent forecasts. For this reason, it is possible to improve the prediction accuracy by not using the actual number of customers that are considered to be affected by this, that is, the actual number of customers whose error from the predicted value is 1000 or more.

  In addition, if the impact is not limited to the day, and the factor that greatly differs from the predicted value and the actual number of customers is not limited to that day, it will be necessary to incorporate these impacts into subsequent forecasts. Forecast using the actual number of customers.

Whether the factor of the error between the predicted value and the actual number of customers is limited to that day or whether it continues thereafter can be determined by whether the value of “1” continues for a predetermined number of days or more due to an error factor in the number of customers data. Here, for example, “7” is used as the predetermined number of days. Next, a process for acquiring customer number data when the error between the predicted customer number and the actual value is large will be described with reference to the flowchart shown in FIG. This process is performed instead of the process of step S203.

  First, the store server 1 reads past data for one week from the predicted execution date (S801). Then, it is determined whether or not there is an error, that is, whether or not a day in which the error area of the read data for one week is “1” is included (S802). When it is determined that there is no error, that is, when there is no day in which the error area of the data for one week is “1” (NO in S802), the read one week of data is added to the acquired data ( S803). When the week thus read is added to the acquired data, it is determined whether or not the number of acquired data is a predetermined number (for example, 365) or more (S804). If it is determined that the number is not greater than the predetermined number (NO in S804), the process returns to step S801.

  If it is determined that there is an error, that is, if there is one or more days in which the data error area for one week is “1” (YES in S802), the error is “1”. It is determined whether a predetermined number or more of the areas are continuous (S805). If there are more than a predetermined number of days with an error area of “1” (YES in S805), the process returns to step S803, and this week is added to the acquired data (S803). Here, for example, 7 is used as the predetermined number. That is, if it continues for 7 consecutive days or more, it is considered that the influence continues, not the error.

  If the predetermined number or more does not continue (NO in S805), it is determined whether the area connected to the preceding and succeeding weeks, that is, the error area of either or both of the first day and the last day, which is the end of the week, is “1” ( S806). If there is no day whose error item is “1” at the end of the week, the read week is not acquired, and the process returns to the process of step S801 to read data for the next week (S801).

  If there is a day where the error area is “1” at the end of the week (YES in S806), the days following that day are sequentially viewed to determine whether the error area is “1” in succession. (S808). The predetermined number is also set to 7, for example, as in the above case. If there are more than a predetermined number of errors (YES in S808), data for this week is acquired (S803). If the predetermined number or more is not continuous (NO in S808), the data for this week is not acquired, and the process returns to Step S801.

  On the other hand, if the store server 1 determines in step S804 that the acquired data is greater than or equal to a predetermined number (YES in S804), the store server 1 sets a predetermined number of data closer to the date from the predicted execution date as acquired data (S809).

  FIG. 19 shows only the data stored in the error area before the prediction execution date arranged in chronological order. What kind of week is acquired will be described with reference to FIG. Here, the continuous number of errors as the predetermined number is 7. First, since the first week (a) including the predicted execution date does not include an error, it is acquired as it is. The week (b) has no error of 7 or more within the week, but since the error is 7 or more continuous over the week (c), this week is acquired. Week (c) is acquired because there are seven or more days with errors in the week. Although there is an error, week (d) is not continuous for 7 or more, and is not continuous for the preceding and succeeding weeks, so this week is not acquired.

  Since POS data is recorded in time series in the table T ′ of the recording unit 15, when data is used as an explanatory variable, when it is determined that there is an error (abnormal value) when creating a multiple regression equation Although it is necessary to prepare POS data that has not been determined to be an error (POS data that is temporally retroactive from the POS data that was originally scheduled to be used), the temporal arrangement of data is important for data stored in time series Therefore, if data that goes back in time is used, that is, if data that is separated in time from the prediction date is used, the prediction accuracy may deteriorate. However, according to this embodiment, by determining the error between the predicted number of customers and the actual number of customers, and when the error is more than a predetermined number, by making a prediction based on the actual number of customers, Whether or not there is an error in the POS data used as the explanatory variable can be determined easily and flexibly, and even when it is determined that there is an error in the POS data, the prediction accuracy can be prevented from being lowered.

  In other words, if errors of a predetermined number or more continue, it is possible to acquire weekly data indicating a long-term trend even if there is a large error by acquiring data for that week.

  The number of customers visiting the store, which is the prediction result predicted as described above, can be used as a reference when purchasing personnel determine the order quantity of products, and can also be used as a reference for staff shift management such as part-time jobs. .

  In this embodiment, the function for carrying out the invention is recorded in advance in the apparatus. However, the present invention is not limited to this, and the same function may be downloaded from the network to the apparatus. You may install what was memorize | stored in the apparatus. The recording medium may be in any form as long as it can store a program such as a CD-ROM and can be read by the apparatus. As described above, a function obtained by installation or download in advance may be realized in cooperation with an operating system (OS) or the like inside the apparatus.

  Note that the present invention is not limited to the above-described embodiments as they are, and can be embodied by modifying constituent elements without departing from the scope of the invention in the implementation stage.

1 is a diagram schematically showing a network system in an embodiment of the present invention. The block diagram which shows the principal part structure of the store server in the embodiment. The block diagram which shows the principal part structure of the POS terminal device in the embodiment. The flowchart which shows the POS data acquisition process in the embodiment. The table which shows the customer number data in the same embodiment. The flowchart which shows the store visitor number prediction process in the embodiment. The figure for demonstrating the case where there is no peculiar day in one week in which the prediction date in the same embodiment is contained. The sub-flowchart of step S205 in the same embodiment. The figure which shows the example of a display of the prediction result in the same embodiment. The figure which shows the relationship between the prediction execution date and prediction date in the embodiment. Sub-flowchart of step S209 in the same embodiment The figure for demonstrating the case where the prediction day in the embodiment exists in a week containing a specific day or a specific day. The sub-flowchart of step S210 in the embodiment. The figure which shows the relationship between the prediction execution date and prediction date in the embodiment. The sub-flowchart of step S216 in the embodiment. 6 is a flowchart showing processing for recording POS data in a recording unit in the embodiment. The table which shows the customer number data recorded on the recording part in the embodiment. The flowchart which shows the process which acquires the number-of-customer data in case the error of the number-of-customer predicted value and a track record value in the embodiment is large. The figure which arranged only the data memorize | stored in the area of the error before the prediction execution date in the same embodiment in time series.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Store server, 2 ... POS terminal device, 3 ... Network, 11, 21 ... CPU, 12, 22 ... ROM, 13, 23 ... RAM, 14, 24 ... Communication control part, 15, 25 ... Recording part, 16, 26 ... Display unit, 17, 27 ... Input unit, 18, 29 ... Clock unit, 28 ... Bar code connection unit, T, T '... Table

Claims (6)

An acquisition means for acquiring the number of visits to the store, which is stored in units of one day in time series;
A store visitor number prediction server comprising: a calculation unit that calculates an average number of store visitors for a predetermined period before the prediction target date from the actual store visitor number acquired by the acquisition unit.   2. The store visitor number predicting server according to claim 1, further comprising a predicting unit that predicts the store visitor day on the basis of the average store visitor calculated by the calculating unit.   3. The store visitor number prediction server according to claim 1, wherein the predetermined period is a week unit.   3. The store visitor number prediction server according to claim 1 or 2, wherein the predetermined period is a week without a peculiar day, which is a day with different tendencies of store visitor numbers. On the computer,
An acquisition function that allows you to acquire the actual number of customers visiting the store in a time series,
A store visitor number prediction program that realizes a calculation function for calculating an average store visitor number for a predetermined period before the prediction target date from the actual store visitor number acquired by the acquisition function.   The store visitor number prediction program according to claim 5, further realizing a prediction function for predicting the store visitor number on the prediction target day based on the average visitor number calculated by the calculation function.

Images