JP5997977B2 - Demand forecasting device, demand forecasting method, and program - Google Patents

Description

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

  In recent years, automated transaction equipment (ATM) has been installed in various places such as banks, station premises, and convenience stores. This automatic transaction device can perform various transactions such as withdrawals, deposits and balance inquiries automatically by the user's own operations, and is still in operation after the bank window is closed. Is expected to increase.

  It is extremely important for financial institutions such as banks to manage the cash demand of such automatic transaction devices and perform efficient operations. For example, Patent Document 1 discloses a cash processing system that manages cash in an automatic transaction apparatus in real time. However, even if the technology can prompt cash replenishment before the cash in the automatic transaction apparatus is emptied, it is urged to prepare a sufficient amount of cash so that the cash in the automatic transaction apparatus does not become empty. I couldn't.

  Therefore, Patent Document 2 discloses a cash management system that predicts the cash demand of an automatic transaction apparatus based on accumulated transaction data. Furthermore, Patent Literature 3 discloses a technique for excluding data on a date that has a special factor such as the vicinity of a salary day and the cash demand largely fluctuates when forecasting cash demand of an automatic transaction apparatus. .

  According to the techniques described in Patent Documents 2 and 3 above, since the cash demand of the automatic transaction apparatus can be predicted based on the accumulated transaction data, a sufficient amount of cash is prepared in the automatic transaction apparatus in advance. be able to. Further, if data on a specific date on which the cash demand largely fluctuates, such as around the salary day, is excluded, the cash demand can be predicted with higher accuracy.

JP-A-8-320966 Japanese Patent Laid-Open No. 11-86089 JP-A-8-123984

  However, with the technology described in Patent Document 3, even though it is possible to eliminate data having a special factor, it has not been possible to exclude expensive transactions that occurred suddenly in individual transactions. In such a case, there is a problem in that the accuracy of subsequent demand prediction is greatly lowered due to the influence of the expensive transaction.

  Therefore, the present invention has been made in view of the above problems, and an object of the present invention is to provide a new and improved demand forecast capable of improving the cash demand forecast accuracy of an automatic transaction apparatus. An apparatus, a demand prediction method, and a program are provided.

  In order to solve the above-described problem, according to one aspect of the present invention, a demand prediction device for predicting cash demand of an automatic transaction device, statistically processing a transaction for a past period, and calculating a statistical outlier. A demand prediction device is provided that includes an outlier extraction unit that extracts an existing transaction as an unexpected transaction, and a demand prediction unit that performs a demand prediction using the transaction for the one period excluding the unexpected transaction.

  The demand prediction apparatus may further include a regularity determination unit that determines whether or not there is regularity in a transaction occurrence day of the sudden transaction, and excludes a transaction having regularity in the transaction occurrence day from the sudden transaction.

  The regularity determination unit may determine that there is regularity when the sudden transaction occurs on the same day of the month or the day before and after.

  The regularity determination unit may determine that there is regularity when the sudden transaction occurs on a fixed cycle day.

  The regularity determination unit may determine that there is regularity when the sudden transaction occurs on a specific day of the week.

  The regularity determination unit may determine that there is regularity when the sudden transaction occurs at a certain frequency.

  The outlier extraction unit may log-transform the number of transactions for each transaction, determine an outlier based on an average and standard deviation of the logarithmically converted number of transactions, and extract a transaction having the outlier.

  The demand prediction device is a trend extraction unit that extracts a transaction trend by comparing a past demand prediction predicted using a transaction for one period excluding the sudden transaction and a demand actual result for a period during which the demand prediction is performed. And a trend reflecting unit that corrects the demand forecast based on the transaction trend.

  The trend extracting unit determines the degree of divergence between the past demand forecast and the demand record when the demand record is more or less than the past demand forecast at a predetermined rate. You may extract as a tendency.

  The trend reflecting unit may reduce the amount of correction to the demand prediction based on the transaction trend according to the period in which the transaction trend is extracted.

  The outlier extraction unit may extract the sudden transaction for each transaction of deposit and withdrawal.

  The demand prediction unit may perform prediction of the cash demand for each denomination banknote.

  Moreover, in order to solve the above-mentioned problem, according to another aspect of the present invention, there is a demand prediction method for predicting cash demand of an automatic transaction apparatus, which statistically processes transactions in the past period, There is provided a demand prediction method comprising: extracting a transaction having an outlier as an unexpected transaction; and performing a demand prediction using the transaction for the one period excluding the unexpected transaction.

  Moreover, in order to solve the said subject, according to another viewpoint of this invention, it is a program for functioning a computer as a demand prediction apparatus which estimates the cash demand of an automatic transaction apparatus, Comprising: The past one period Statistically processing the transaction, and an outlier extraction unit that extracts a transaction having a statistical outlier as an unexpected transaction; a demand prediction unit that performs a demand prediction using the transaction for the one period excluding the unexpected transaction; A program for functioning as a server is provided.

  As described above, according to the present invention, it is possible to improve the cash demand prediction accuracy of the automatic transaction apparatus.

It is explanatory drawing which showed the structure of the demand prediction system which concerns on this invention. It is the block diagram which showed the internal structure of the demand prediction apparatus which concerns on 1st Embodiment. It is explanatory drawing which showed an example of the specific example of the transaction data which is the past transaction performance. It is the graph which showed an example of the specific example which made the histogram the number of outflows of 10,000 tickets in the past one period. It is a flowchart figure which shows operation | movement of the demand prediction apparatus which concerns on this embodiment. It is the block diagram which showed the internal structure of the demand prediction apparatus which concerns on 2nd Embodiment. It is a flowchart figure which shows operation | movement of the demand prediction apparatus which concerns on this embodiment. It is a block diagram which shows the hardware constitutions of the demand prediction apparatus which concerns on this invention.

  Exemplary embodiments of the present invention will be described below in detail with reference to the accompanying drawings. In addition, in this specification and drawing, about the component which has the substantially same function structure, duplication description is abbreviate | omitted by attaching | subjecting the same code | symbol.

<1. Configuration of demand forecasting system>
The present invention can be implemented in various forms, as will be described below by way of example. In addition, the demand prediction device 20 according to the embodiment of the present invention includes:
A. An outlier extraction unit 230 that statistically processes transactions in the past period and extracts transactions having statistical outliers as sudden transactions;
B. A demand prediction unit 250 that performs demand prediction using the transaction of the one period excluding the sudden transaction;
Is provided.

  In the following, first, the basic configuration of a demand prediction system including such a demand prediction device 20 will be described with reference to FIG.

  FIG. 1 is an explanatory diagram showing a configuration of a demand prediction system according to the present invention. As shown in FIG. 1, the demand prediction system according to the present invention includes a financial institution host 10, a dedicated network 12, an automatic transaction apparatus central management unit 14, a demand prediction apparatus 20, an automatic transaction apparatus 30, and terminal management. Unit 32.

  The automatic transaction apparatus 30 can automatically perform transactions such as deposits and withdrawals by operations of customers of financial institutions. The automatic transaction apparatus 30 is installed in various facilities such as a branch of a financial institution, a convenience store, a station premises, a department store, a hotel, and an office building. For example, as shown in FIG. 1, automatic transaction apparatuses 30A to 30N are installed in the A branch 16 of the financial institution, and communication with the financial institution host 10 and the automatic transaction apparatus central management unit 14 and the automatic transaction apparatus 30A are performed. A terminal management unit 32 for managing ˜30N is installed. Similarly, a plurality of automatic transaction apparatuses 30 and a terminal management unit 32 are installed in a branch B 18 other than the branch A 16 (not shown).

  The dedicated network 12 is a dedicated network owned by a financial institution, and is configured by, for example, an IP-VPN (Internet Protocol-Virtual Private Network). The financial institution host 10 and the automatic transaction apparatus central management unit 14 can communicate with the automatic transaction apparatus 30 via the dedicated network 12.

  The financial institution host 10 controls various transactions of the automatic transaction apparatus 30 by communicating with the automatic transaction apparatus 30 via the dedicated network 12. For example, the financial institution host 10 authenticates a customer who operates the automatic transaction apparatus 30, or executes transaction processing such as deposit / withdrawal and transfer instructed by the customer's operation in the automatic transaction apparatus 30. . The financial institution host 10 manages customer information such as an account number, personal identification number, name, address, age, date of birth, telephone number, occupation, family structure, annual income, and deposit balance as an account ledger.

  The automatic transaction apparatus central management unit 14 monitors the operating status of each automatic transaction apparatus 30 by communicating with the automatic transaction apparatus 30 via the dedicated network 12. For example, the automatic transaction apparatus central management unit 14 monitors the amount of cash remaining in the automatic transaction apparatus 30, the status of transactions, the presence or absence of errors (out of cash, banknote jam, etc.), and the like. Moreover, the day when there is no cash in the automatic transaction apparatus 30 can be predicted from the demand for cash predicted by the demand prediction apparatus 20 and the amount of cash remaining in the automatic transaction apparatus 30, so that replenishment can be performed in advance. Make a “cash loading plan” and give instructions.

  The demand prediction device 20 is a prediction device that predicts the future cash demand of each automatic transaction device 30. In the automatic transaction apparatus 30, for example, cash flows out due to a withdrawal transaction. Therefore, it is necessary to load the automatic transaction apparatus 30 with cash at an appropriate timing. For this reason, the demand prediction device 20 predicts future cash demand based on past transaction results.

  In addition, as disclosed in Patent Document 3, data on a date that causes a significant increase in cash demand due to a special factor such as the vicinity of a salary transfer date affects the demand prediction. The demand prediction may be performed by distinguishing data of the date. Furthermore, when a system abnormality such as a power failure or a miscount at the time of cash loading occurs, the demand prediction device 20 performs the demand prediction by excluding the data on the date when the abnormality occurred as abnormal data. Also good.

  Since the amount of cash outflow varies from branch to branch, the demand prediction device 20 may predict the demand for cash from branch to branch. Moreover, even if it is the automatic transaction apparatus 30 in the same branch, the outflow amount of cash differs. For example, the outflow amount of the automatic transaction apparatus 30 near the entrance of the branch tends to be larger than the outflow amount of the automatic transaction apparatus 30 far from the entrance. For this reason, the demand prediction device 20 may predict the cash demand for each automatic transaction device 30.

  Moreover, as for the past transaction performance which the demand prediction apparatus 20 uses for prediction of cash demand, it is desirable to use the transaction performance of the automatic transaction apparatus 30 which performs the said prediction. However, when the demand prediction device 20 predicts the cash demand for each branch, all transaction results of all automatic transaction devices 30 in the branch performing the prediction may be used.

  The demand prediction system according to the present invention has been described above with reference to FIG. According to this demand prediction system, it is possible to predict future cash demand based on past transaction results.

  However, in the transaction of the automatic transaction apparatus 30, a rare transaction that cannot normally occur may occur. For example, even in the automatic transaction apparatus 30 that normally performs transactions of tens of thousands to hundreds of thousands of yen, transactions of several million yen may occur. As a result, the actual transaction amount on the date on which the transaction occurred is significantly greater than the actual transaction amount on other dates. Here, since the demand prediction device 20 predicts future cash demand based on past transaction results, if there is such an unexpectedly expensive transaction, more predictions of cash demand will be calculated. . However, since such a sudden transaction is not reproduced, as a result, when a sudden high-priced transaction occurs, the prediction accuracy of the cash demand of the demand prediction device 20 is lowered.

  Therefore, the demand prediction device 20 according to the present invention has been created by paying attention to the above problem. Since the demand prediction device 20 according to the present invention can predict the cash demand by extracting and eliminating the transaction as a sudden transaction even if the transaction is suddenly generated, the cash demand can be predicted accurately. It is possible to improve. Hereinafter, specific embodiments of the demand prediction apparatus 20 according to the present invention will be described in detail.

<2. First Embodiment>
[Configuration of Demand Prediction Device 20A]
First, the internal configuration of the demand prediction apparatus 20A according to the first embodiment will be specifically described with reference to FIG. FIG. 2 is a block diagram showing an internal configuration of the demand prediction apparatus 20A according to the present embodiment.

  As illustrated in FIG. 2, the demand prediction apparatus 20A according to the present embodiment includes an outlier extraction unit 230, an outlier extraction parameter storage unit 235, a regularity determination unit 240, a regularity determination parameter storage unit 245, A demand prediction unit 250 and a demand prediction parameter storage unit 255 are provided.

(Outlier extraction unit, outlier extraction parameter storage unit)
The outlier extraction unit 230 acquires transaction data that is a past transaction record, and extracts an unexpected transaction having an outlier based on the outlier extraction parameter stored in the outlier extraction parameter storage unit 235. Here, prior to detailed description of the outlier extraction unit 230 and the outlier extraction parameter storage unit 235, a specific example of transaction data that is past transaction results will be described with reference to FIG.

  FIG. 3 is an explanatory view showing an example of transaction data which is a past transaction record. As shown in FIG. 3, the transaction data includes a device ID for identifying the automatic transaction device 30, a transaction ID, a date and time when the transaction occurred, a time, and the number of outflows of each banknote.

  In FIG. 3, since the device ID is “10000101”, the transaction data shown in FIG. 3 indicates the past transaction record of the automatic transaction apparatus 30 whose device ID is “10000101”. According to the transaction data shown in FIG. 3, the transaction with the transaction ID “AB15000001” occurred at 9:11:15 on December 1, 2011, indicating that 3 million tickets were withdrawn. Show. In addition, the transaction with the transaction ID “AB15000002” occurred at 9:11:58 on December 1, 2011, and indicates that 1 million tickets and 10 thousand tickets were withdrawn. Further, the transaction with the transaction ID “AB15000003” occurred at 9:13:21 on December 1, 2011, and since the outflow number is a negative number, it indicates that 5 million tickets have been deposited.

  The outlier extraction unit 230 statistically processes the transaction data as shown in FIG. 3 described above, thereby detecting an outlier and extracting a transaction having the outlier as an unexpected transaction. The outlier extraction parameter storage unit 235 stores a statistical processing method performed by the outlier extraction unit 230, parameters for determining a threshold for detecting an outlier, and the like.

  The period during which the outlier extraction unit 230 statistically processes the transaction data may be a past period. Specifically, in April, June and December, withdrawal transactions increased compared to other months, and in January and July, deposit transactions increased. It is desirable to use the above period. For example, the outlier extraction unit 230 may statistically process transaction data for the latest three years.

  Here, a specific example of the outlier extraction method of the outlier extraction unit 230 will be described with reference to FIG. FIG. 4 is a graph showing an example of a specific example in which the number of outflows of 10,000 tickets in the past period is a histogram. As shown in FIG. 4, in the past period, the number of outflows of ten thousand tickets is taken as the number of transactions and the horizontal axis is taken for every five sections, and the number of transactions is taken as the frequency on the vertical axis. Here, a transaction with a positive number of transactions is a withdrawal transaction, and a transaction with a negative number of transactions is a deposit transaction. In addition, as shown in FIG. 4, the logarithmic normal distribution approximate curve is drawn with a solid line for each of the withdrawal transaction and the receipt transaction, and the histogram of the transaction data is logarithmic normal for each of the withdrawal transaction and the receipt transaction. It is shown that the distribution can be approximated.

  In the outlier extraction, first, the outlier extraction unit 230 divides the transaction data of the past one period into a withdrawal transaction and a deposit transaction, sets a threshold value for judging the outlier for each transaction, and suddenly has an outlier. Extract transactions. In the following, we will mainly explain the outlier extraction of withdrawal transactions, but for deposit transactions as well, since the transaction data is negative, the outliers are essentially the same except that the absolute value of the number of transactions is used when performing logarithmic conversion. Extraction is performed. As shown in FIG. 4, since the withdrawal transaction can be approximated by a lognormal distribution, the outlier extraction unit 230 uses the mathematical formulas 1 to 3 to logarithmically transform the number of transactions x and the standard deviation y and standard deviation σ. Is calculated.

  Next, the outlier extraction unit 230 determines whether or not y obtained by logarithmically converting the number of transactions x for each transaction in the past one period exceeds a threshold μ + mσ. When y obtained by logarithmically converting the number of transactions x exceeds a threshold μ + mσ, the outlier extraction unit 230 extracts a transaction having the transaction number x as an unexpected transaction. Here, m is a predetermined real number for detecting an outlier. For example, m may be 2 or 3.

  Further, the outlier extraction unit 230 also extracts an unexpected transaction having an outlier using the outlier extraction method described above, similarly to the withdrawal transaction. Similarly, the outlier extraction unit 230 can extract an unexpected transaction in both the deposit transaction and the withdrawal transaction for the number of outflows of thousand tickets.

  As described above, the outlier extraction unit 230 can extract an unexpected transaction having an outlier based on the outlier extraction parameter stored in the outlier extraction parameter storage unit 235. In the above embodiment, the outlier detection is performed based on the average and the standard deviation assuming that the histogram of the number of transactions can be approximated by a logarithmic normal distribution, but the present invention is not limited to such an example. For example, the outlier extraction unit 230 may detect outliers by approximating the histogram of the number of transactions to a normal distribution or Poisson distribution, and based on the degree of deviation from the average without using the average and the standard deviation. Outliers may be detected. Further, the outlier extraction unit 230 may extract an unexpected transaction having an outlier using another outlier detection method such as outlier detection based on clustering or outlier detection based on density ratio estimation.

(Regularity judgment unit, regularity judgment parameter storage unit)
The regularity determination unit 240 determines whether or not the sudden transaction having the outlier extracted by the outlier extraction unit 230 has regularity in the transaction occurrence day of the unexpected transaction. Is excluded from sudden transactions. The regularity determination parameter storage unit 245 stores regularity definitions, determination parameters, and the like for the regularity determination unit 240 to determine whether there is regularity in the transaction occurrence day of the sudden transaction. .

  Even if it is a high-value transaction that is extracted as an unexpected transaction by the outlier extraction unit 230, when the transaction is reproduced with a certain regularity, it is better to use it as transaction data for cash demand prediction. Can be improved. Therefore, the regularity determination unit 240 determines whether or not there is regularity in the transaction occurrence day of the sudden transaction by paying attention to, for example, the common attribute, period and frequency of the transaction occurrence date.

  Here, the regularity determination unit 240 may determine that there is regularity when an unexpected transaction occurs on the same day of each month and on the day before and after that. For example, when an unexpected transaction occurs on “June 15” and thereafter an unexpected transaction occurs again on any of “July 14-16”, the regularity determination unit 240 determines that the aforementioned unexpected transaction is a rule. You may judge that it has sex. Here, even if the months in which sudden transactions occur are not consecutive, the regularity determination unit 240 may determine that the aforementioned sudden transactions have regularity. In the above example, even if an unexpected transaction occurs on “June 15” and an unexpected transaction occurs again on any of “August 14-16”, the regularity determination unit 240 may The transaction may be determined to have regularity. In addition, the present invention is not limited to the above-described example in which it is determined that there is regularity when an unexpected transaction occurs on the same day of the month or on the day before or after that day. For example, the regularity determination unit 240 has regularity when the sudden transaction occurs on any of the five days “July 13 to July 17” on the same day of the month and the two days before and after the same. You may judge.

  In addition, the regularity determination unit 240 may determine that there is regularity when sudden transactions occur at a constant period. For example, if an unexpected transaction occurs on “June 15” and then an unexpected transaction occurs again on “June 25” and “July 5”, the unexpected transaction occurs on a 10-day cycle. Therefore, the regularity determination unit 240 may determine that the above-described sudden transaction has regularity.

  Note that the regularity determination unit 240 may determine that there is regularity when an unexpected transaction occurs on a specific day of the week. For example, if “June 8” of a certain year is Friday, “June 15”, “June 22”, and “June 29” are Friday. Here, when an unexpected transaction occurs on “June 8”, “June 15”, and “June 29”, since the unexpected transaction occurs on a specific day of the week, it is a regularity judgment. The unit 240 may determine that the aforementioned unexpected transaction has regularity.

  Furthermore, the regularity determination unit 240 may determine that there is regularity when sudden transactions occur at a certain frequency. For example, sudden transactions occurred on “June 10”, “June 15”, “June 17”, “June 20”, “June 25”, and “June 27” In this case, since the sudden transaction has occurred at a constant frequency of 3 times in 10 days, the regularity determination unit 240 may determine that the aforementioned unexpected transaction has regularity. Here, in the above example, when sudden transactions occur at a certain frequency, the regularity determination unit 240 determines that there is regularity, but the present invention is not limited to such an example. Even when an unexpected transaction occurs at a frequency equal to or higher than a threshold, the regularity determination unit 240 may determine that the aforementioned unexpected transaction has regularity. Here, the threshold value of the frequency at which the regularity determination unit 240 determines that the regularity determination unit 240 has regularity can be stored in the regularity determination parameter storage unit 245. In addition, the regularity determination unit 240 may separately count the sudden transactions when the sudden transactions occur a plurality of times on the same day.

  In the above example, the number of repeated sudden transactions that the regularity determining unit 240 determines to have regularity is set to 2 or 3, but the present invention is not limited to such number. The number of repetitions that the regularity determination unit 240 determines to have regularity may be four or five, and a threshold value of an appropriate number of repetitions can be stored in the regularity determination parameter storage unit 245. .

  The regularity determination unit 240 excludes all of the transactions that are determined to have regularity on the transaction occurrence date from the sudden transaction. The excluded transaction may be used as transaction data for cash demand prediction in a demand prediction unit 250 described later, as in a normal transaction. In addition, the regularity determination unit 240 defines a transaction that is determined to be regular on the transaction occurrence date as a regular transaction, and calculates the rule for cash demand prediction calculated from transaction data excluding sudden transactions and the regular transactions. You may make it apply the correction | amendment based on sex transactions.

(Demand forecasting unit, demand forecasting parameter storage unit)
Based on the demand prediction parameters stored in the demand prediction parameter storage unit 255, the demand prediction unit 250 determines the future cash demand of the automatic transaction apparatus having the transaction results as past transaction results excluding sudden transactions. Predict. The demand prediction parameter storage unit 255 stores a cash demand prediction method, parameters used for cash demand prediction, and the like.

  Here, the well-known various method can be employ | adopted for the prediction method of the cash demand of the demand prediction part 250. FIG. For example, the demand prediction unit 250 may predict the cash demand by averaging past transaction results, but the present invention is not limited to such an example.

[Operation of Demand Prediction Device 20A]
The configuration of the demand prediction device 20A according to this embodiment has been described above. Next, a specific example of the operation of the demand prediction device 20A according to the present embodiment will be described with reference to FIG.

  FIG. 5 is a flowchart showing the operation of the demand prediction apparatus 20A according to this embodiment. As illustrated in FIG. 5, the demand prediction device 20A executes the processes of S300 to S340 including a transaction type and a denomination loop.

  Specifically, first, the demand prediction device 20A acquires transaction data of the automatic transaction device 30 that is a target for predicting cash demand (S300). Next, the outlier extraction unit 230 obtains the data of the transaction number from the transaction data of the past one period, and calculates the average and the standard deviation on the assumption that the histogram of the transaction number can be approximated to a lognormal distribution. (S310). Furthermore, the outlier extraction unit 230 determines a threshold based on the calculated average and standard deviation, and extracts a transaction having an outlier that exceeds the threshold as an unexpected transaction (S320).

  Subsequently, the regularity determination unit 240 determines whether there is regularity in the transaction occurrence day for the unexpected transaction extracted by the outlier extraction unit 230, and the regular transaction is excluded from the unexpected transaction. (S330). The outlier extraction unit 230 and the regularity determination unit 240 execute the processes of S310 to S330 in a loop for each of the withdrawal transaction and the deposit transaction.

  By the above S310-S330, the demand prediction device 20A can calculate transaction data of the withdrawal transaction and the deposit transaction from which the unexpected transaction is excluded. The demand prediction unit 250 predicts the cash demand based on the transaction data (S340). Here, 20 A of demand prediction apparatuses can estimate the future cash demand of the said automatic transaction apparatus by performing the process of S310-S340 to each denomination of 10,000 tickets and thousand tickets.

  As described above, the demand prediction device 20A according to the present embodiment can predict a cash demand by excluding the transaction as a sudden transaction even if an unexpectedly expensive transaction occurs. Prediction accuracy can be improved. In addition, if there is regularity on the transaction occurrence date and reproducibility is seen in the sudden transaction, it is possible to exclude the regular transaction from the sudden transaction by judging whether there is regularity. It is possible to further improve the accuracy of demand prediction.

[Modification of First Embodiment]
Subsequently, a modification of the first embodiment will be described below. In the modified example of the first embodiment, the demand prediction device 20A performs outlier extraction based on the number of transactions per day instead of the number of transactions, and designates the day having the number of transactions as an outlier as an abnormal day. Transaction data can be excluded from transaction data used to predict cash demand.

  Specifically, the outlier extraction unit 230 calculates the average μ ′ and the standard deviation σ ′, assuming that the number of transactions per day is approximated to a lognormal distribution, not the number of transactions for each transaction. Next, the outlier extraction unit 230 extracts a day that is the number of transactions exceeding μ ′ + m′σ ′ as an abnormal day by using μ ′ + m′σ ′ as a threshold. Here, m ′ is a predetermined real number for detecting an outlier. In addition, the regularity determination unit 240 determines whether or not there is regularity in an abnormal day by a method similar to the method for determining the regularity in the transaction occurrence day of the sudden transaction, and has regularity in the abnormal day Excludes all days with such regularity from abnormal days. Furthermore, the demand prediction unit 250 performs cash demand prediction based on the transaction data excluding the transaction data on the abnormal day.

  Note that the modification of the first embodiment can be implemented alone, but may be implemented in combination with the first embodiment. In such a case, the prediction accuracy of cash demand can be further improved.

<3. Second Embodiment>
The first embodiment that can improve the prediction accuracy of cash demand has been described in detail above. However, in the first embodiment, since cash demand is predicted based on past transaction results over a long period of time, for example, a tourist spot or an office building is formed near the branch where the automatic transaction apparatus 30 is installed. When the cash demand suddenly increases, the trend change of the cash demand cannot be immediately reflected in the demand forecast.

  On the other hand, in order to reflect the sudden cash demand in the demand forecast, when the demand forecasting device 20 makes a cash demand forecast based on the latest past short-term transaction performance, the long-term transaction performance is used. The monthly characteristics that cannot be extracted without it cannot be reflected in the demand forecast. For example, in order to reflect the impact of seasonality on demand forecasts, such as April, June and December, where cash demand is higher than in other months and deposits are increased in January and July, preferably one year or more It is necessary to forecast cash demand based on the transaction performance during the period.

  Therefore, the second embodiment of the present invention has been created by paying attention to the above problem in addition to the problem to be solved by the first embodiment. The demand prediction device 20B according to the present embodiment first makes a cash demand prediction based on the past long-term transaction performance, and further corrects the demand prediction based on the latest trend change in the past cash demand. Thereby, in addition to the influence of long-term seasonality, the demand prediction apparatus 20B can predict the cash demand corresponding to the sudden trend change of the cash demand. Below, the demand prediction apparatus 20B of such 2nd Embodiment is demonstrated in detail.

[Configuration of Demand Prediction Device 20B]
First, the internal configuration of the demand prediction apparatus 20B according to the second embodiment will be specifically described with reference to FIG. FIG. 6 is a block diagram showing an internal configuration of the demand prediction apparatus 20B according to the present embodiment.

  As shown in FIG. 6, the demand prediction apparatus 20B according to the present embodiment includes an outlier extraction unit 230, an outlier extraction parameter storage unit 235, a regularity determination unit 240, a regularity determination parameter storage unit 245, A demand prediction unit 250, a demand prediction parameter storage unit 255, a trend extraction unit 260, a trend extraction parameter storage unit 265, and a trend reflection unit 270 are provided.

  The outlier extraction unit 230, the outlier extraction parameter storage unit 235, the regularity determination unit 240, the regularity determination parameter storage unit 245, the demand prediction unit 250, and the demand prediction parameter storage unit 255 are the first Since the configuration is the same as or substantially the same as that of the embodiment, a detailed description thereof is omitted here. Hereinafter, the trend extraction unit 260, the trend extraction parameter storage unit 265, and the trend reflection unit 270 that should be specially described in the second embodiment will be specifically described.

  The trend extraction unit 260 compares the cash demand prediction predicted based on the transaction data excluding sudden transactions with the actual transaction performance in the prediction period, and sets the trend extraction parameter stored in the trend extraction parameter storage unit 265. Based on this, the trend change in cash demand is extracted as a transaction trend. In addition, the trend extraction parameter storage unit 265 stores a transaction trend determination algorithm and a determination threshold for extracting a transaction trend.

  Specifically, the trend extracting unit 260 suddenly detects the demand forecast predicted from the past transaction data excluding the sudden transactions and the actual transaction results in the relevant period during the most recent past one week to three months. Compare the transaction results excluding the transaction. In addition, the trend extraction unit 260 calculates and compares the number of days when the transaction performance has exceeded the demand forecast and the date when the transaction has fallen. The trend extraction unit 260 determines whether the transaction result exceeds the threshold when either one of the days when the transaction result is higher or the day when the transaction result is lower than a predetermined ratio, and the degree of deviation of the transaction result from the demand forecast exceeds the threshold. Judge that there has been a change in the trend of demand, and extract transaction trends.

  For example, the trend extraction unit 260 has twice the number of days when the number of days when the transaction performance has exceeded the demand forecast is less than the number of days, and the average cash demand for the day when the transaction performance has exceeded When the average of the demand prediction is 1.2 times, “transaction result is 1.2 times higher than the demand prediction” is extracted as a transaction tendency.

  If the period for calculating the transaction trend is short, if there is an unexpected transaction, the transaction tendency is easily affected, and the accuracy of the transaction trend extracted by the trend extraction unit 260 is reduced. On the other hand, since the demand prediction apparatus 20B according to the present embodiment calculates a transaction trend from the transaction results and the demand prediction from which the unexpected transaction is excluded by the outlier extraction unit 230, even if the period during which the transaction trend is calculated is short, the accuracy can be improved. A high transaction tendency can be calculated. Therefore, even if the cash demand trend changes suddenly, the demand prediction device 20B can correct the demand prediction quickly.

  The trend reflection unit 270 corrects the future cash demand prediction predicted by the demand prediction unit 250 based on the transaction trend extracted by the trend extraction unit 260. The trend reflection unit 270 may decrease the correction amount according to the length of the period in which the transaction trend is extracted when the cash demand prediction is corrected based on the transaction trend extracted by the trend extraction unit 260. In other words, the transaction trend extracted by the trend extraction unit 260 becomes less reliable as the period during which the transaction trend is extracted is shorter. Therefore, the trend reflecting unit 270 may reduce the correction amount when correcting the demand for the extracted transaction tendency so as not to be overcorrected when the period in which the transaction tendency is extracted is short and the reliability is low. Good. The trend reflecting unit 270 may provide an upper limit for the correction amount for correcting the demand prediction.

[Operation of Demand Prediction Device 20B]
The configuration of the demand prediction device 20B according to this embodiment has been described above. Subsequently, a specific example of the operation of 20B according to the present embodiment will be described with reference to FIG.

  FIG. 7 is a flowchart showing the operation of the demand prediction apparatus 20B according to this embodiment. As illustrated in FIG. 7, the demand prediction device 20 </ b> B includes a transaction type and a denomination loop as in the first embodiment, and executes the processes of S <b> 300 to S <b> 370. Since the processes from S300 to S340 are the same as or substantially the same as those in the first embodiment, a detailed description thereof will be omitted, and the processes after S350 that should be noted in the present embodiment will be described in detail. .

  Specifically, after the demand prediction is performed by the demand prediction unit 250 for the target denomination in the process of S340, the trend extraction unit 260 predicts the demand for the past period predicted from the transaction data excluding unexpected transactions. Is acquired (S350). Next, the trend extraction unit 260 acquires a transaction record that excludes sudden transactions in the past period, compares the demand prediction with the transaction record, and extracts a transaction trend (S360). Further, the trend reflection unit 270 corrects the prediction of the future cash demand predicted by the demand prediction unit 250 in S340 based on the transaction trend extracted by the trend extraction unit 260 (S370).

  Here, in the process of the demand prediction device 20A according to the first embodiment, S310 to S340 are set as a loop based on denominations of ten thousand and thousand bills, and the demand for cash is predicted for each denomination. In the process of the demand prediction device 20B according to the second embodiment, S310 to S370 are set as a loop based on the denominations of ten thousand and one thousand denominations, and corrections based on cash demand predictions and transaction trends are performed for each denomination. .

  As described above, in addition to the effects of the first embodiment, the demand prediction device 20B according to the present embodiment corrects the demand prediction based on the trend change extracted by comparing the past demand prediction and the transaction results. By doing so, it is possible to predict cash demand corresponding to sudden changes in cash demand.

<4. Hardware configuration>
The embodiment of the present invention has been specifically described above. Information processing such as extraction of unexpected transactions having outliers, determination of regularity of occurrence date of unexpected transactions, demand prediction, and extraction of transaction trends is performed by software and hardware of the demand prediction apparatus 20 described below. It is realized by collaboration.

  FIG. 8 is a block diagram illustrating a hardware configuration of the demand prediction apparatus 20. The demand prediction device 20 includes a CPU (Central Processing Unit) 201, a ROM (Read Only Memory) 203, a RAM (Random Access Memory) 205, and an internal bus 207. The demand prediction device 20 includes an input / output interface 209, a display unit 211, an input unit 213, an audio output unit 215, a HDD (Hard Disk Drive) 217, a drive 219, a network interface 221, and an external interface. 223.

  The CPU 201 functions as an arithmetic processing device and a control device, and controls the overall operation in the demand prediction device 20 according to various programs. Further, the CPU 201 may be a microprocessor. The CPU 201 corresponds to the outlier extraction unit 230, the regularity determination unit 240, the demand prediction unit 250, the trend extraction unit 260, and the trend reflection unit 270, and realizes the functions of the respective units.

  The ROM 203 stores programs used by the CPU 201, calculation parameters, and the like. The RAM 205 temporarily stores programs used in the execution of the CPU 201, parameters that change as appropriate during the execution, and the like. These are connected to each other by an internal bus 207, and further connected to a display unit 211 (to be described later), an input unit 213, an audio output unit 215, an HDD 217, a drive 219, a network interface 221, and an external interface 223 via an input / output interface 209. Is done.

  The display unit 211 includes display devices such as a liquid crystal display (LCD) device, an OLED (Organic Light Emitting Diode) device, and a CRT (Cathode Ray Tube) display device. The audio output unit 215 includes an audio output device such as a speaker and headphones, for example, and converts audio data or the like into audio and outputs the audio.

  The input unit 213 includes input means for a user to input information, such as a mouse, a keyboard, a touch panel, a button, a microphone, a switch, and a lever, an input control circuit that generates an input signal based on the input by the user, and outputs the input signal to the CPU 201. It is composed of The user of the demand prediction device 20 can input various data to the demand prediction device 20 or instruct a processing operation by operating the input unit 213.

  The HDD 217 is a data storage device configured as an example of a storage unit of the demand prediction device 20 according to the present embodiment. The HDD 217 includes a storage medium, a recording device that records data in the storage medium, a reading device that reads data from the storage medium, and a deletion device that deletes data stored in the storage medium. The HDD 217 stores programs executed by the CPU 201 and various data. The HDD 217 corresponds to the outlier extraction parameter storage unit 235, the regularity determination parameter storage unit 245, the demand prediction parameter storage unit 255, and the trend extraction parameter storage unit 265, and implements the functions of the respective units.

  The drive 219 is a storage medium reader / writer, and is built in or externally attached to the demand prediction apparatus 20. The drive 219 reads information stored in a mounted removable storage medium such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory and outputs the information to the RAM 203. The drive 212 can also write information to a removable storage medium.

  The network interface 221 is a communication interface configured by a communication device or the like for connecting to the dedicated network 12, for example. The network interface 221 may be a wired LAN (Local Area Network) or a wireless LAN compatible communication device, or may be a wire communication device that performs wired communication.

  The external interface 223 is a connection port for connecting an external connection device such as a USB (Universal Serial Bus) port, an IEEE 1394 port, a SCSI (Small Computer System Interface), an RS-232C port, or an optical audio terminal. A configured connection interface.

<5. Summary>
As described above, the demand prediction device 20 according to the present invention can extract and eliminate the transaction as a sudden transaction even if a large amount of transaction occurs unexpectedly, so that the demand for cash can be predicted. It is possible to improve the accuracy of prediction. In addition, after eliminating unexpected transactions from transaction data, comparing the past demand forecast based on the transaction results for the period concerned, and applying the correction based on the trend change of the extracted cash demand to the demand forecast, It is possible to predict cash demand corresponding to changes in cash demand trends.

  The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to such examples. It is obvious that a person having ordinary knowledge in the technical field to which the present invention pertains can come up with various changes or modifications within the scope of the technical idea described in the claims. Of course, it is understood that these also belong to the technical scope of the present invention.

  In addition, it is possible to create a computer program for causing hardware such as the CPU 201, the ROM 203, and the RAM 205 built in the demand prediction apparatus 20 according to the present invention to perform the same functions as the components of the demand prediction apparatus 20 described above. is there. A storage medium storing the computer program is also provided.

DESCRIPTION OF SYMBOLS 10 Financial institution host 12 Dedicated network 14 Automatic transaction apparatus central management part 20 Demand prediction apparatus 30 Automatic transaction apparatus 32 Terminal management part 201 CPU
203 ROM
205 RAM
207 Internal bus 209 Input / output interface 211 Display unit 213 Input unit 215 Audio output unit 217 HDD
219 Drive 221 Network interface 223 External interface 230 Outlier extraction unit 235 Outlier extraction parameter storage unit 240 Regularity judgment unit 245 Regularity judgment parameter storage unit 250 Demand prediction unit 255 Demand prediction parameter storage unit 260 Trend extraction unit 265 Trend extraction parameter Storage unit 270 Trend reflection unit

Claims (13)

A demand prediction device for predicting cash demand of an automatic transaction device,
An outlier extraction unit that statistically processes transactions in the past period and extracts transactions having statistical outliers as sudden transactions;
Determining whether or not there is regularity in the transaction occurrence day of the sudden transaction, and a regularity determination unit for excluding transactions having regularity in the transaction occurrence day from the sudden transaction;
A demand prediction device comprising: a demand prediction unit that performs demand prediction using the transaction for the one period excluding the sudden transaction. The demand prediction device according to claim 1 , wherein the regularity determination unit determines that there is regularity when the sudden transaction occurs on the same day of the month or the day before and after. The demand forecasting device according to claim 1 or 2 , wherein the regularity determination unit determines that there is regularity when the sudden transaction occurs on a fixed cycle day. The demand forecasting device according to any one of claims 1 to 3 , wherein the regularity determination unit determines that there is regularity when the sudden transaction occurs on a specific day of the week. The demand forecasting device according to any one of claims 1 to 4 , wherein the regularity determination unit determines that there is regularity when the sudden transaction occurs at a constant frequency. The outlier extraction unit logarithmically converts the number of transactions for each transaction, determines an outlier based on an average and standard deviation of the logarithmically converted number of transactions, and extracts a transaction having the outlier. The demand prediction apparatus according to any one of 5 . A trend extractor that extracts a transaction trend by comparing a past demand forecast predicted using a transaction for one period excluding the sudden transaction and a demand actual for the period during which the demand prediction was performed, and the transaction trend The demand forecasting device according to any one of claims 1 to 6 , further comprising a trend reflecting unit that corrects demand forecast based on the trend forecasting unit. The trend extracting unit determines the degree of divergence between the past demand forecast and the demand record when the demand record is more or less than the past demand forecast at a predetermined rate. The demand prediction apparatus according to claim 7 , which is extracted as a trend. The demand prediction device according to claim 7 or 8 , wherein the trend reflection unit reduces a correction amount to the demand prediction based on the transaction tendency according to a period in which the transaction tendency is extracted. The demand prediction device according to any one of claims 1 to 9 , wherein the outlier extraction unit extracts the sudden transaction for each transaction of deposit and withdrawal. The said demand prediction part is a demand prediction apparatus as described in any one of Claims 1-10 which each performs the prediction of the said cash demand about the banknote of each denomination. A demand prediction method for predicting cash demand of an automatic transaction apparatus,
Statistically processing transactions in the past period and extracting transactions having statistical outliers as sudden transactions;
Determining whether there is regularity in the transaction occurrence day of the sudden transaction, and excluding transactions having regularity in the transaction occurrence day from the sudden transaction;
A demand prediction method comprising: performing a demand prediction using the transaction of the one period excluding the sudden transaction. Computer
A program for functioning as a demand prediction device for predicting cash demand of an automatic transaction device,
An outlier extraction unit that statistically processes transactions in the past period and extracts transactions having statistical outliers as sudden transactions;
Determining whether or not there is regularity in the transaction occurrence day of the sudden transaction, and a regularity determination unit for excluding transactions having regularity in the transaction occurrence day from the sudden transaction;
A program for functioning as a demand prediction unit that performs demand prediction using the transaction for the one period excluding the sudden transaction.

Images