JP3243129B2 - Fund management support equipment for automatic transaction equipment - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an automatic transaction system which stores time-series data of past transaction information of an automatic transaction device and environmental information at the time of occurrence of a transaction, and uses the data to predict the required amount of funds for a prediction period. The present invention relates to a fund management support device for a device.

[0002]

2. Description of the Related Art Bank ATM as a conventional automatic transaction device
However, it was possible to manage funds to some extent efficiently by reusing funds deposited within a single bank ATM for withdrawal. However, the total amount of withdrawal exceeded the total amount of
If the funds in the TM exceed the minimum amount of funds in the safe, it is necessary to replenish funds by a staff member. Similarly, if the total amount of withdrawal exceeded the total amount of deposit and the funds exceeded the maximum storage amount of the safe, it was necessary for the staff to collect the funds.

[0003] However, the conventional timing of replenishment and collection is determined in advance or when a bank ATM runs out of funds. The amount of funding for replenishment and collection is determined by the experience of a person in charge of replenishment and collection. I was Before the holidays, they often replenish their funds with sufficient time to avoid running out of funds during the holidays.

As described above, conventionally, a bank ATM
Replenishment and withdrawal of funds within the facility depended on staff experience. However, the amount of funds in the bank ATM was often set relatively large because of fear of the bank ATM running out of funds. This is Bank AT
A large amount of money will be stored in M, and efficient management of funds will be hindered.

[0005] Further, in this case, there is a high possibility that inexperienced staff members cannot determine the amount of funds well, so that there is a problem that the funds are stolen or conversely the bank ATM has to be stopped because of insufficient funds. Become. In addition, in the case of a new store, even an experienced staff at another store has a disadvantage that it takes time to obtain an experience for determining an appropriate amount of funds.

[0006]

SUMMARY OF THE INVENTION The object of the present invention is to eliminate the above-mentioned drawbacks, and to rely on the experience of staff in the past by estimating the future required amount of funds using time-series data of past transaction information. It can accurately indicate the amount of funds for replenishment / collection to the automated teller machine, can reduce the required amount of funds, and even inexperienced staff can replenish / replenish the appropriate amount of funds.
It is an object of the present invention to provide a fund management support device for an automatic transaction device that can be collected.

[0007]

A fund management support apparatus for an automatic transaction apparatus according to the present invention comprises: first input means for inputting time-series data of transaction information of the automatic transaction apparatus; Second input means for inputting date and time, day of the week, and weather information of the weather , time series data input by the first input means and environmental information input by the second input means are stored in association with each other. first storage means, clustering means for obtaining a time series pattern for each transaction type by clustering the time-series data stored in the first storage means, the time series for each transaction type obtained by the clustering unit A second storage unit that stores the pattern in association with the environment information input by the second input unit, and stores the pattern corresponding to the prediction period using the environment information of the prediction period. Search means for retrieving the time series pattern for each transaction type and its environment information from the second storage means, and combining a time series pattern for each of a plurality of transaction types retrieved by the search means with the environment information It is composed of prediction means for predicting the required amount of funds for the prediction period based on the information, and output means for outputting information on the required amount of funds predicted by the prediction means.

[0008]

SUMMARY OF THE INVENTION This invention inputs the time-series data of the transaction information of the automatic teller machine in the first input unit, the date and time of the transaction event of the automatic transaction apparatus, day, weather environmental information and the second input means And stores the time series data input by the first input means and the environment information input by the second input means in the first storage means in association with each other. the resulting time series pattern for each transaction type by clustering the series data collected by the resulting
The time series pattern for each type of discount is stored in the second storage unit in association with the environment information input by the second input unit, and the past transactions corresponding to the prediction period are stored using the environment information of the prediction period. A time-series pattern for each type and its environment information are retrieved from the second storage means, and the required time of the forecast period is estimated based on the combination of the retrieved time-series patterns for each of the plurality of transaction types and the environment information. The amount is predicted, and information on the predicted required amount of funds is output.

[0009]

An embodiment of the present invention will be described below with reference to the drawings. This embodiment is a fund management support apparatus for an automatic transaction apparatus (bank ATM: automatic tellers machine) of the present invention. In this method, time-series data of transaction information such as deposits and withdrawals of a bank ATM is input, clustered, and stored together with environmental information such as date, day of the week, weather, and the like at the time of transaction. Then, an appropriate past time-series pattern is searched based on the environmental information of the prediction period, and the required amount of funds in the prediction period is predicted based on the search.

FIG. 1 is a diagram showing an example of an A comprising a plurality of banks ATM1a,... As an automatic transaction apparatus realized by using the present invention.
FIG. 1 is a schematic configuration diagram of an ATM fund management support device 11 as a fund management support device for a TM group 1 and each bank ATM 1a,. Here, the thick arrow represents the flow of data.

The overall processing flow of the ATM fund management support apparatus 11 and the schematic operation of each unit will be described with reference to FIG. That is, transaction information such as deposits and withdrawals from a plurality of banks ATM1a,... Is input to the time-series data input device 2 in a time-series manner. On the other hand, the date, day of the week,
Environmental information such as weather is input from the environmental information input device 3 regardless of whether it is manual or automatic. These are both stored in the time-series information database 4.

Next, when the time-series data is collected in a unit time, it is sent to the clustering device 5 together with the environmental information. The clustering device 5 classifies data having similar time series patterns as the same cluster with respect to all the transmitted time series data. The clustered time-series pattern is sent to the pattern database 6 and stored as clustering information together with the environment information.

On the other hand, the prediction period environment information prediction device 7 uses the environment information obtained from the environment information input device 3 and the pattern database 6 to obtain fixed information such as the day of the week for the prediction period, and to change the weather and the like. Is predicted and sent to the pattern search device 8 as prediction period environment information. The pattern search device 8 searches the pattern database 6 for the most appropriate reference time-series pattern for the prediction period based on the transmitted prediction period environment information.

The required fund amount prediction device 9 performs noise removal of the reference time series pattern based on the retrieved reference time series pattern, uses similarities and differences between the prediction environment information and the reference environment information, and the like. Then, the required amount of funds in the prediction period is predicted. The required fund amount output device 10 outputs information on the required fund amount based on the fund amount predicted by the necessary fund amount prediction device 9 and the predicted environment information. Is a claim.

The details of each section will be described below. Each bank AT
M1a,... Sends to the time-series data input device 2 the unique information of each bank ATM 1a,.
Here, the model is the model name and model number of the bank ATM1a,... And the number assigned to the bank ATM1a,. These are used as unique information. Furthermore, the serial number is a serial number assigned to each transaction, the transaction time is the time when the transaction occurred, the transaction type is the type of transaction such as deposit, withdrawal, etc., the transaction amount is the actual amount of cash transaction, The number of each transaction is the actual number of transactions for each banknote and coin, and the number in each device is the amount of each banknote and coin present in each bank ATM1a,. These are the information at the time of the transaction.

The environment information input device 3 inputs environment information at the time of occurrence of a transaction as shown in FIG. The date, day of week, holiday flag, day before holiday flag, temperature, humidity, wind speed, and barometric pressure are the date, day of week, flag of holiday, flag of day of holiday, temperature and humidity at the time of transaction. , Wind speed and pressure. These may be entered by an attendant from an input device such as a keyboard, but the date, day of the week, holiday flags, flags before holidays, etc. may be entered from a timing device,
Humidity, wind speed, air pressure, etc. may be automatically input from the sensor, in which case the labor of the attendant can be saved.

The information at the time of the transaction sent from each of the ATMs 1a,... And the information sent from the environment information input device 3 are both in the format as shown in FIG. This is temporarily stored in the time-series information database 4 for each model number in a format as shown in FIG.

When the time series data is collected in a certain time unit, it is sent to the clustering device 5. The clustering device 5 converts the transmitted time-series data for each model number into pattern data for each transaction type. FIG. 6 shows an example of payment pattern data.

Next, a reference pattern for estimating the required amount of funds is obtained based on these patterns. For example,
Now, assuming that the transition of the amount of money is used, the transition pattern of the amount of money may be used as it is as the reference pattern. Alternatively, a pattern in which payment is subtracted from payment may be generated using a payment pattern and a payment pattern instead of the change in the amount of money, and this pattern may be used as a reference pattern. Further, each pattern for each transaction type necessary for prediction may be used as a reference pattern. The advantage in this case will be described later. The reference pattern has the same format as that of FIG.

Finally, the reference patterns are classified into similar patterns using a clustering technique. Then, an average of the patterns in the same cluster is calculated, and this is set as a reference representative pattern. The data format is the same as in FIG. By performing the clustering in this way, when a certain bank ATM 1a breaks down at a certain time and a transaction is stopped, a sudden disturbance such as an increase in transactions of another bank ATM 1a confuses prediction. Can be prevented. Of course, by changing the parameters at the time of clustering, for example, one pattern per cluster can be obtained. This means that one cluster includes only one bank ATM1a pattern. In this case, it is vulnerable to disturbance, but bank A of that model number
It is possible to make an accurate prediction most suitable for TM1a.

FIG. 7 shows the transition of the amount of money as a reference pattern. When four patterns A, B, C, and D are given, A and B are clustered into cluster 1, C and D are clustered into cluster 2. This is an example when reference representative patterns 1 and 2 are obtained, respectively.

The obtained reference representative patterns are sent to and stored in the pattern database 6 together with the environment information. By storing the reference representative patterns, the number of data can be reduced, and the data search speed can be increased.

On the other hand, the prediction period environment information prediction device 7 obtains substantially fixed information such as a day of the week or a holiday for a given prediction period, and predicts information that changes such as temperature and humidity. In the former case, a clerk may input from an input device such as a keyboard, or may have a timing device inside and determine automatically. In the latter case, the attendant may look at the weather forecast and input from an input device such as a keyboard,
For example, it is also possible to make predictions based on past transition patterns of temperature, humidity, and the like, or to make inferences using knowledge such as "when the atmospheric pressure falls and the humidity rises, it will rain". These pieces of information are sent to the pattern search device 8 in the same format as in FIG.

The pattern search device 8 is a past reference which is most appropriate for predicting the amount of funds in the prediction period based on the prediction period sent from the prediction period environment information prediction device 7 and the corresponding prediction period environment information. A period is obtained, and a corresponding reference time-series pattern is searched from the pattern database 6.

FIG. 8 shows the simplest algorithm for realizing the pattern search device 8. First, when a prediction period is input (ST1), a reference reference period is determined based on the prediction period (ST2). This may be given in advance as the same time in the same month of the previous year, the same time in the previous month, the same time in the previous month, the same time in the previous day, etc., may be determined using the knowledge given in advance, or gradually learn the appropriate ones You may make it.

Next, this reference reference period is temporarily set as a reference period (ST3), and it is checked whether or not the reference period corresponds to a transaction suspension period (ST4). If the reference period corresponds to the transaction suspension period, the reference period is added to a reference period list, which is a list of periods that have already been referred to (ST5), and the reference period is shifted to a period not included in the already-referenced period list (ST6). Return to 4.

If the reference period does not correspond to the transaction suspension period, a time-series pattern including the reference period as a component of the cluster is searched from the pattern database 6 (ST).
7), and output (ST8).

What has been described above is an example of the simplest algorithm. Next, a case where a classification tree given in advance is used for searching for a time-series pattern corresponding to a reference period will be described.

The pattern search device 8 prepares a classification tree from past time-series data using environment information as a parameter in advance, and has information classified in the form of a classification tree. FIG. 9 shows an example of a classification tree when the length of the reference period is set to one day. In FIG. 9, “bonus day”, “payday”, etc. are environmental information. A vertical line attached to the shoulder indicates that the environmental information is satisfied, and a vertical line not attached does not satisfy the environmental information. It will be shown.

When performing a search using this classification tree, it is first checked whether or not the prediction period is a "bonus day". In the case of the “bonus day”, 6/10 and 12/10 correspond to the reference period, and a time-series pattern including these as a component of the cluster is searched from the pattern database. If it is not "bonus day" and it is "payday", the time series data of the cluster including 1/25,.

Further, if it is not "payday", it is divided into "Friday", "Saturday, Sunday" and other days, and days other than "Friday", "Saturday, Sunday" are "fine weather", "cloudy weather", "rainy weather". And the corresponding time-series data is searched. By using environmental information,
It becomes possible to search for appropriate time-series data corresponding to the prediction period.

The classification tree may be learned in the pattern search device 8 at regular intervals. This generates and records a new classification tree by classifying time-series data at regular intervals based on environment information input from the environment information input device 3. The pattern search device 9 is advantageous in the following points by searching for a time-series pattern using a newly generated classification tree.

First, by increasing the number of time-series patterns used for generating a classification tree, a more appropriate time-series pattern can be searched. Next, since the latest environmental information can be used, it is possible to cope with a change in the environment, for example, when a company or an apartment complex is formed in the neighborhood.

Further, in the search for the time-series pattern by the pattern search device 8, knowledge describing rules or the like may be used. FIG. 10 shows an example in which rules corresponding to the classification tree shown in FIG. 9 are described. In this example, one rule starts with a rule number such as RULE1: and the rules from IF to THEN are not, and, or
Is the description of the condition part by the attribute / attribute value pair connected by. The attribute here is one of the environmental information.

The list below THEN indicates an appropriate prediction period corresponding to the condition part. For example, RULE
3 is that the environmental information "day" is not "bonus day",
If “day” is not “payday” and “day” is “before and after consecutive holidays”, a time-series pattern including “4/28” and “5/6” as cluster components is used as a reference time-series pattern. Demonstrate knowledge that it is appropriate to use.

If these rules are described using expert knowledge obtained in advance or knowledge obtained using a classification tree, the empirical knowledge of the expert can be entered. It is possible to search for a time series pattern. In addition, when converting the classification tree that has been gradually learned into rules or when automatically extracting knowledge such as rules using the inductive learning method, the knowledge is automatically refined and can respond to changes in the environment. There are advantages.

On the other hand, in the time series pattern of the past transaction information, it may be understood that the time series pattern is clearly different for each transaction type. For example, payments tend to be more on a particular day, but payments are mainly on Fridays. In this case, a time-series pattern for each transaction type required for fund amount prediction may be searched.

The required fund amount prediction device 9 predicts the required fund amount in the prediction period based on the reference time series pattern searched by the pattern search device 8. As a reference pattern, there are cases where a single type of pattern such as a time-series pattern of the amount of money or a time-series pattern in which a deposit is subtracted from a withdrawal is used, and a case where a time-series pattern for each transaction type necessary for fund amount prediction is used. is there.

The former is a case where there is no clear difference for each transaction type or a case where no clear difference is found, and the latter is a case where there is a clear difference. In the latter case, the time series pattern for each transaction type can be predicted more precisely, and the combination thereof is used to predict the required amount of funds, so that highly accurate prediction can be performed. For example, the time-series patterns of deposit and withdrawal are separately predicted, and the predicted required fund amount is determined using a pattern obtained by subtracting the deposit pattern from the withdrawal pattern.

Several examples of the required fund amount predicting device 9 will be described below. Now, the period from the transaction start time to the transaction end time on May 25 is set as the prediction period. At this time, the pattern search device 8 first uses January 25, February 25
It is assumed that a period from the transaction start time to the transaction end time on March 25, April 25, and April 25 is obtained as a reference period.
Further, the time-series patterns on January 25 and February 25 and the time-series patterns on March 25 and April 25 are clustered in advance into the same cluster as Cluster 1 and Cluster 2, respectively. It is assumed that it is stored.

The pattern retrieval device 8 retrieves the reference representative time-series patterns of the clusters 1 and 2 from the pattern retrieval device 6 to obtain patterns 1 and 2 as shown in FIG. At this time, for example, when a single pattern is obtained, prediction may be performed based on the pattern. However, when a plurality of patterns are obtained, a single pattern is obtained using the following method or the like.

First, as shown in FIG. 12, there is a method of averaging a plurality of obtained patterns. This allows
The noise between the reference representative time-series patterns can be averaged, and the bank ATM happens to be at a certain time when the time-series pattern referred to happens.
It is possible to reduce the influence of a sudden abnormal situation, such as when 1a,... Has failed and stopped, and it is effective to reduce the required funding amount.

As shown in FIG. 13, there is also a method of taking the maximum value among the obtained plural patterns. In this case, the required amount of funds is slightly increased, but it is possible to reduce situations such as suspension of transactions due to lack of funds.

Further, as shown in FIG. 14, there is a method of obtaining the minimum value among the obtained plural patterns. in this case,
The possibility of suspension of trading due to lack of funds increases, but the required amount of funds can be reduced most.

A single pattern is obtained by using the above method and the like, and based on this, the final required amount of funds is predicted by the following method. First, the required amount of funds during the reference period is obtained using the obtained single time-series pattern. This is the sum of the required funds for each transaction, and this value is then added to the bank ATM1a,.
Multiply the safety factor or add the minimum required amount so that the funds do not run out, and make the final required funds. At this time, for example, if it is known in advance that a situation requiring a large amount of funds will occur, such knowledge is input in advance, so that a more accurate prediction of the amount of funds can be made.

Here, furthermore, each predicted bank ATM
By calculating the sum of the required funds for each of the banks 1a,..., The required total funds of the banks ATM1a,. Thus, the amount of funds required for the entire bank ATM 1a,... Can be understood at a glance, and the request for the amount of funds for cash transportation is simplified.

Finally, the required fund amount output device 10 outputs information on the required fund amount based on the required fund amount and the predicted environment information predicted by the required fund amount prediction device 9 as shown in FIG. 15, for example. I do. Each column such as ten thousand yen and one thousand yen is the number of the bills and coins, and the number is shown as it is in case of replenishment, and a minus sign is shown in case of collection.

The total is the amount of funds required for each model number.
Also, if necessary, for example, by clicking the model number and pressing the button with the description, as shown in FIG.
The reason for the fund amount prediction may be output based on the predicted environment information. As a result, the attendant can be convinced of the required amount of money indicated by the bank ATM operation support device 11, and if necessary, the attendant can discover a clear mistake or lack of knowledge of the device, and ultimately, The necessary amount of funds can be corrected.

Further, at this time, for example, by pressing an advisory button, advice necessary for replenishment / recovery as shown in FIG. 11 may be given based on the required amount of money of each bank ATM 1a,. This is the case where the model number 005 has a large amount of withdrawal and must be replenished, while the model number 004 and the model number 008 have a large number of deposits and must collect funds.

At this time, the necessary fund amount output device 10 collects the necessary amount of funds from the model numbers 004 and 008 based on the estimated replenishment / collection fund amounts of the respective ATMs 1a,. This is an example of outputting an advice that it is better to replenish to 005. Thereby, the labor of replenishment / recovery work of the staff can be reduced.

As described above, by having the transaction information used for the prediction clustered as time-series data, the influence of data disturbance can be prevented and the storage capacity can be reduced. In addition, prediction is performed by directly using the searched data by using environmental information such as date and time, day of the week, and weather in searching for past time-series data.

Thus, by predicting the required amount of funds based on past transaction information of the automatic transaction apparatus and environmental information such as date and time, day of the week, weather, etc., the amount of funds is reduced, and the work load of the staff is reduced. Can be reduced.

That is, the time series data of past transaction information and the environment information at that time are input, clustering is performed on the time series data for a certain period, and the cluster is stored together with the environment information. Then, by searching for the past pattern that seems to be most suitable for the transaction status in the forecast period and using it to accurately predict the required fund amount in the forecast period, it is possible to reduce the amount of money required for the bank ATM. At the same time, even inexperienced staff can set an appropriate amount of funds.

[0054]

According to the present invention, by using the time-series data of past transaction information to predict the required amount of funds in the future, it is possible to accurately indicate the amount of funds for replenishment / recovery to an automatic transaction device which has conventionally relied on the experience of a staff member. It is possible to provide a fund operation support device of an automatic transaction device that can reduce the required amount of funds and can replenish / recover an appropriate amount of funds even with inexperienced staff. In addition, the type of transaction required for forecasting the amount of funds
Each time series pattern is searched, and the searched funds
Forecast based on time series pattern for each transaction type required for volume forecast
By estimating the required amount of funds for the period,
Time-series patterns can be predicted more precisely, and
Predict the required amount of funds in combination with the forecast
It can be performed. For example, time series pattern of deposit and withdrawal
Forecasts separately from payment patterns
The estimated required fund amount is calculated using the pattern
You.

[Brief description of the drawings]

FIG. 1 is a diagram showing a schematic configuration of a fund management support device of an automatic transaction device for explaining an embodiment of the present invention.

FIG. 2 is a diagram showing an example of transaction information sent from an ATM group each time a transaction occurs.

FIG. 3 is a diagram illustrating an example of environment information input to the environment information input device when a transaction occurs.

FIG. 4 is a diagram showing an example of information at the time of a transaction transmitted from each bank ATM and an environment information input device.

FIG. 5 is a diagram showing an example of time series information for each model number stored in a time series information database.

FIG. 6 is a diagram illustrating an example of pattern data for each transaction type for each model number.

FIG. 7 is a diagram showing an example of reference patterns and reference representative patterns when they are clustered.

FIG. 8 is a diagram illustrating an example of an algorithm for realizing the pattern search device.

FIG. 9 is a diagram illustrating an example of a classification tree used for searching for a time-series pattern in the pattern search device.

FIG. 10 is a diagram illustrating an example of a rule used for searching for a time-series pattern in the pattern search device.

FIG. 11 is a diagram illustrating an example of a time-series pattern searched by the pattern search device.

FIG. 12 is a diagram illustrating an example of obtaining a single pattern by averaging a plurality of time-series patterns when there are a plurality of time-series patterns searched by the pattern search apparatus.

FIG. 13 is a diagram illustrating an example of obtaining a single pattern from the maximum values of a plurality of time-series patterns when there are a plurality of time-series patterns searched by the pattern search apparatus.

FIG. 14 is a diagram illustrating an example of obtaining a single pattern from the minimum value of a plurality of time-series patterns when there are a plurality of time-series patterns searched by the pattern search apparatus.

FIG. 15 is a diagram illustrating an example of information on a required fund amount and a predicted fund amount output by a necessary fund amount output device.

FIG. 16 is a diagram illustrating an example of a reason that is a basis for estimating the amount of funds output by the required amount of funds output device.

FIG. 17 is a diagram showing an example of advice on replenishment and collection of funds output by the necessary fund amount output device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... ATM group 1a ... Bank ATM (automatic transaction apparatus) 2 ... Time-series data input device (1st input means) 3 ... Environmental information input apparatus (2nd input means) 4 ... Time-series information database (No. 1 storage unit) 5 clustering device (clustering unit) 6 pattern database (second storage unit) 7 forecast period environment information prediction device 8 pattern search device (search unit) 9 necessary fund amount prediction device (prediction) Means) 10: Required fund amount output device (output means) 11: Fund management support device of automatic transaction device

──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Kazunori Konda 70, Yanagicho, Saiwai-ku, Kawasaki-shi, Kanagawa Prefecture Inside the Toshiba Yanagicho Plant (56) References 6-20134 (JP, A) JP-A-3-19079 (JP, A) Abe, Nagashima, Maeda, Nagano, "Use of POS data for inference based on patterns automatically detected from time-series data" Application, ”Information Processing Society of Japan 48th (early 1994) National Convention Lecture Papers, No. 2, March 23, 2006, p. 2-175 to 2-176 (58) Fields investigated (Int. Cl. ⁷ , DB name) G06F 17/60 G06F 19/00 G07D 9/00 JICST file (JOIS)

Claims (1)

(57) [Claims] Enter the 1. A first input means for inputting the time series data of the transaction information of the automatic teller machine, date and time of transaction upon occurrence of the automatic teller machine, day of the week, a ring <br/> boundary information weather A second input unit; a first storage unit that stores the time series data input by the first input unit and the environment information input by the second input unit in association with each other; Clustering means for obtaining a time-series pattern for each transaction type by clustering the time-series data stored in the storage means, and a time-series pattern for each transaction type obtained by the clustering means. A second storage unit for storing the environment information input by the input unit 2 in association with the environment information of the prediction period, and a time series pattern for each past transaction type corresponding to the prediction period and its environment using the environment information of the prediction period. A search means for searching for boundary information from the second storage means; and a combination of time series patterns for a plurality of transaction types searched for by the search means and a required amount of funds for a prediction period based on environmental information. A fund management support device for an automatic transaction apparatus, comprising: a prediction unit for predicting; and an output unit for outputting information relating to the required amount of funds predicted by the prediction unit.