JPH07234861A - Data monitoring system - Google Patents

Abstract

PURPOSE:To automatically recognize data patterns of a data set of time-series data, etc., to be monitored by a data monitoring system which utilizes an infer ence system, and derive a proper inference result on the basis of the data pat tern. CONSTITUTION:A pattern extraction part 2 extracts data patterns of the time- series data for learning consisting of, for example, past time-series data. A pattern recognition part 5 recognizes a data pattern that the time-series data to be monitor is similar to, among the extracted data patterns. An inference system 6 outputs an inference result corresponding to the time-series data to be monitored on the basis of the data pattern recognized by the pattern recognition part 5 by utilizing knowledge information for deriving the inference result based upon a specific rule from the extracted data patterns. Therefore, the inference process utilizing the knowledge information for deriving the inference result based upon the specific rule from the data pattern can be performed without being well acquainted with the data pattern generated in the time-series data to be monitored in advance.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data monitoring system for monitoring various time-series data such as POS data of a sales information management system used in a retail business by an inference method utilizing a knowledge database.

[0002]

2. Description of the Related Art Conventionally, a computer system has been used to monitor time-series data such as weather data, stock price data, medical-related diagnostic data, etc. Data monitoring systems have been developed. With such a system, for example, in a hospital or the like, the state of the electrocardiogram is monitored in time series by the collected diagnostic data.

Recently, POS has been introduced especially in the retail business.
A sales information management system called a (point of sales) system is a kind of data monitoring system, and is effective in sales management of products. This PO
In the S system, time-series data related to merchandise sales management called POS data is monitored.

Some data monitoring systems such as POS systems use an inference system that incorporates the results of recent AI (artificial intelligence) technology development. Such a system uses a rule-based inference method or the like to
The state of transition of time-series data such as data is monitored, and an inference result in the form of "if the data pattern of A occurs, give advice of B" is output to the time-series data. The inference system is roughly divided into a knowledge database and an inference section (inference mechanism),
It is a system that uses knowledge information (rules) in a knowledge database to derive an appropriate inference result (a warning or other action) from a rule corresponding to a data pattern.

Specifically, for example, the transition of time-series data relating to the sales of a certain product is monitored, and if the time-series data shows "a data pattern of A", for example, "increase the purchase amount of the product. This is the case where the advice such as "is issued as the inference result.

By the way, in a system using the inference system as described above, it is necessary to describe in the form of a rule as knowledge information, "if a data pattern of A occurs,". However, before describing in the rule format, it is necessary to be sufficiently familiar with whether the data pattern of A is a similar pattern. For example, when targeting time-series data such as stock prices, a knowledge database is constructed by describing rules based on research results on stock price patterns such as Granville's law.

[0007]

Conventionally, in a sales information management system or the like, in a data monitoring system using an inference system, it is necessary to be familiar with knowledge about a data pattern generated in time-series data to be monitored. As a premise, the knowledge database described in the rule format has been constructed. In other words, without knowing the characteristics of the transition of the time-series data to be monitored, the knowledge database could not be constructed and the inference system could not be effectively utilized.

An object of the present invention is to automatically recognize a data pattern of a data set such as time-series data to be monitored in a data monitoring system using an inference system, and obtain an appropriate inference result based on this data pattern. It is to provide a system that makes it possible to derive.

[0009]

According to the present invention, there is provided a learning database means for storing a learning time series data set consisting of past time series data, and a pattern extracting means for extracting the learning time series data set data pattern. A data monitoring system comprising pattern recognition means for recognizing a data pattern of a time-series data set to be monitored and inference execution means for outputting an inference result for the time-series data set to be monitored based on the recognized data pattern. .

[0010]

In the present invention, the pattern extracting means extracts a data pattern of a learning time series data set consisting of past time series data, for example. The pattern recognition means recognizes, from the extracted data patterns, data patterns having similar time series data sets to be monitored. The inference execution means uses the knowledge information for deriving an inference result based on a predetermined rule from the extracted data pattern, and infers the time series data set to be monitored based on the data pattern recognized by the pattern recognition means. Output the result. Therefore, even if you are not familiar with the data patterns that occur in the time-series data to be monitored, you can perform inference processing that uses knowledge information to derive inference results based on predetermined rules from the data patterns. Becomes

[0011]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a basic configuration of a data monitoring system according to the embodiment, and FIG. 2 is a block diagram when the data monitoring system of the embodiment is applied to a sales information management system (POS system). FIG. 4 is a flow chart for explaining the operation of the same embodiment, and FIGS. 5 and 8 are conceptual diagrams showing an example of a display screen for explaining the operation of the data monitoring system of the same embodiment. (Basic Configuration) As shown in FIG. 1, the data monitoring system of the same embodiment includes a learning database 1, a pattern extraction unit 2, a pattern output unit 3, a monitoring target database 4, a pattern recognition unit 5, an inference system 6 and knowledge. It comprises an input unit 7.

The learning database 1 stores, for example, a time series data set in which past time series data is accumulated. The pattern extraction unit 2 extracts a data pattern of a predetermined format from the time series data set extracted from the learning database 1 by using, for example, a clustering method. The pattern output unit 3 is an element for displaying the data pattern extracted by the pattern extraction unit 2. The knowledge input unit 7 is an element for describing knowledge information for guiding an appropriate inference result from the displayed data pattern in a rule format and inputting it to the inference system 6.

On the other hand, the monitoring target database 4 stores, for example, a time series data set in which current monitoring target time series data is accumulated. The pattern recognition unit 5 recognizes the data patterns extracted by the pattern extraction unit 2 that have similar monitoring target time-series data sets extracted from the monitoring target database 4 and outputs them to the inference system 6. To do.

The inference system 6 executes an inference process for leading an inference result corresponding to the data pattern output from the pattern recognizing unit 5, using a knowledge database composed of knowledge information input by the knowledge input unit 7. The inference result is action contents such as appropriate warnings and advice obtained by analyzing the monitored time series data based on the recognized data pattern. (Specific Configuration Applied to POS System) As shown in FIG. 2, a system in which the data monitoring system having the above-described basic configuration is applied to a POS system has a control unit 10, an input / output unit 11, and learning POS data. Extraction unit 12, target P
OS data extraction unit 13, pattern database 14,
It has a knowledge database 15, an inference unit 16, and an inference result output unit 17.

The learning POS data extraction unit 12 stores the past POS data (mainly time-series data relating to product sales management) from the learning POS database 1.
The POS data in the range designated by the control unit 10 is taken out and output to the pattern extraction unit 2. On the other hand, the target POS
The data fetching unit 13 fetches the POS data in the range designated by the control unit 10 from the target POS database 4 in which the current POS data to be monitored is accumulated,
It is output to the pattern recognition unit 5.

The input / output unit 11 has a keyboard, a mouse and a display, and inputs a command to the control unit 10 by pointing an input item in a window (or icon) displayed on the screen of the display with the mouse. . The input / output unit 11 also displays the data pattern output from the pattern output unit 3 and the inference result output from the inference result output unit 17 on the screen of the display.

The pattern database 14 stores the data patterns extracted by the pattern extracting section 2. The pattern recognition unit 5 searches the pattern database 14 for a data pattern similar to the pattern of the target POS data extracted by the target POS data extraction unit 13 and outputs it as a recognition result.

The knowledge database 15 stores the knowledge information input by the knowledge input unit 7 through the input / output unit 11. The knowledge database 15, the inference unit 16, and the inference result output unit 17 are elements that make up the inference system shown in FIG. The inference unit 16 is an inference mechanism that executes inference processing by the rule type inference method, and uses the knowledge information of the knowledge database 15 described in the rule format to correspond to the data pattern recognized by the pattern recognition unit 5. Output an appropriate inference result. This inference result is a warning,
It consists of caution and advice. The inference result output unit 17 displays the inference result obtained by the inference unit 16 on the display of the input / output unit 11.

Next, the operation of the embodiment when applied to the POS system will be described. First, when the system is activated, the control unit 10 displays a menu screen on the display of the input / output unit 11 so as to prompt the user to select the learning mode or the monitoring mode. That is, as shown in FIG. 5A, the window 11a for the menu screen is displayed on the display. The user (operator) operates the mouse of the input / output unit 11 to move the cursor 11b to instruct a desired mode (step S in FIG. 3).
1).

When the "pattern learning" of the window 11a is instructed by operating the mouse, the learning mode is selected and the learning POS data extracting section 12 is activated by the control section 10 (YES in step S2, S3). here,
In the learning POS database 1, as shown in FIG.
The POS data in which the past sales number (the present year is 1993) and the purchase number are recorded on a daily basis is stored for each product.

The control unit 10 activates the learning POS data extraction unit 12 and also opens and closes a window (learning data setting screen) for designating the range of extraction data as shown in FIG. 5B. It is displayed on the display of 11. When the range of the fetched data is designated by this window, the learning POS data fetching unit 12 sets the learning P
The POS data in the designated range is retrieved from the OS database 1 (step S4). Here, as shown in FIG. 5B, all the individual items of the product “tofu” from April 1 to 30, 1992 are designated. Numbers such as the date are input by inputting / outputting the numbers on the keyboard or mouse of the input / output unit 11, and items such as the product “Tofu” are input by the menu function.

The learning POS data extracting unit 12 extracts the POS data in the designated range from the learning POS database 1 and, as shown in FIG. 6 (A), has a predetermined format (here, the purchase number to the sales number). Calculation result of subtracting
And output to the pattern extraction unit 2 (step S
5). The pattern extraction unit 2 is activated by instructing the "execution" mode with the mouse on the screen shown in FIG. 5B (step S6).

When the pattern extracting unit 2 receives the POS data in the format shown in FIG. 6A, the pattern extracting unit 2 executes a process of classifying the products into a predetermined group (corresponding to a pattern) by a clustering method, and the POS for each product. Extract the characteristic pattern of the data. Specifically, the difference between the purchase and the sales (that is, the number of unsold items) is treated as time-series data. For example, products that are routinely sold out are data pattern 3, and products that are routinely unsold are data pattern 2. Further, the products whose fluctuations in the unsold and sold out are large are classified into the data pattern 1. For example, as shown in FIG. 6B, a product (denoted as E) whose unsold amount is “1” or “2” throughout the year.
Are classified into data pattern 3.

Here, for example, in each of the products A to E, since 30 pieces of POS data from April 1 to 30 are used, they are assumed to be 30-dimensional data and clustered. As a result, when the product D and E are classified into a cluster having the elements D, E, a cluster having the products B and C as an element, and a cluster having the product A as an element, these three clusters are handled as the data patterns 1 to 3. Figure 8
(A) is a graph of such data patterns 1 to 3.

The pattern extraction unit 2 outputs the extracted data pattern for each product in the format shown in FIG. 6B and stores it in the pattern database 14 (step S7). Here, the pattern extraction unit 2 actually obtains the average value of each cluster obtained by the clustering and sets it as a data pattern. The pattern extraction unit 2 also outputs the extracted data pattern to the pattern output unit 3. As shown in FIG. 8A, the pattern output unit 3 displays the graphed extraction pattern on the screen of the display of the input / output unit 11 (step S8).

The user uses the graph displayed on the screen of the display of the input / output unit 11 to specify the specified POS.
You can check the pattern of the data. The user is
The knowledge input unit 7 is activated by operating the input / output unit 11 to input knowledge information for leading an appropriate inference result corresponding to the data pattern. Specifically, for each of the displayed data patterns 1 to 3, the knowledge information for pointing out warnings, cautions, etc., and obtaining advice is entered in a rule format by an editor (step S9). The knowledge information input by the knowledge input unit 7 is stored in the knowledge database 1
5 (step S10). The knowledge database 15 is stored as rule format information, for example, as shown in FIG. The rule shown in FIG. 8 (B) is
In the case of the data pattern being pattern 3, it is pointed out that "there is a tendency that there is a shortage of unsold stock" and the advice that "it is better to increase the order" is given.

On the other hand, the target POS database 4 is a base on which daily POS data is accumulated, and is specifically realized as a file of a computer system. The target POS database 4 is the past (here, 19
The POS database 1 for learning that stores the POS data of 1992) and the PO of the present (here, 1993)
It is a collection of S data.

Here, as shown in FIG. 5 (A), when the window 11a for the menu screen is displayed on the display and the "monitoring mode" is designated by operating the mouse,
The target POS data extraction unit 13 is activated by the control unit 10 (YES in step S11, S12). Control unit 10
Activates the target POS data extraction unit 13, and
As shown in FIG. 5C, a window (target data setting screen) for designating the range of fetched data is displayed on the display of the input / output unit 11.

When the range of the fetched data is designated by this window, the target POS data fetching section 13 fetches the designated POS data from the target POS database 4 (steps S13 and S14). Target POS
The data fetching unit 13 fetches data for the number of days (that is, for 30 days) which is the same number of dimensions of the data pattern stored in the pattern database 14 based on the input current date. Here, since the current date is “February 20”, as shown in FIG.
P from "January 21st" to "February 19th" a few minutes ago
Retrieve OS data. This POS data has the same format as the extracted learning POS data and is time-series data indicating the difference between the purchase and the sales (that is, the number of unsold items).

The pattern recognition unit 5 recognizes whether the target POS data extracted from the target POS database 4 is similar to the contact pattern among the data patterns stored in the pattern database 14 (step S1).
5). Specifically, the pattern recognition unit 5 calculates the Euclidean distance between the target POS data and each data pattern if the data pattern stored in the pattern database 14 is the average value of clustered clusters, and calculates the maximum distance. A data pattern close to is output as a recognition result.

The inference unit 16 infers an appropriate inference result corresponding to the data pattern of the target POS data recognized by the pattern recognition unit 5 based on the knowledge information stored in the knowledge database 15 (step S16). That is, for example, if the target POS data is the data pattern 3, it is pointed out by the knowledge information in the rule format shown in FIG. Output the advice as an inference result.
The inference result output unit 17 displays the inference result obtained by the inference unit 16 on the display of the input / output unit 11 (step S17).

In this way, when the present invention is applied to a POS system, daily POS data is monitored as time-series data in the form of, for example, "unsold data", and the data pattern of the POS data to be monitored in the specified range is set. The corresponding appropriate inference result can be obtained. That is, if the monitored POS data is the data pattern 3 which is close to "sold out", appropriate advice such as "increase order" can be obtained. Further, if the data pattern 2 has a large amount of unsold items on a daily basis, it is possible to obtain appropriate advice such as "reduce orders."

In this embodiment, the monitoring target P is based on the data pattern extracted from the past learning POS data.
An appropriate inference result can be obtained by recognizing the data pattern of the OS data and utilizing the inference mechanism by the knowledge database for the data pattern. Therefore, unlike the conventional method, it is not always necessary to be familiar with the knowledge regarding the data pattern generated in the time-series data to be monitored. That is, conventionally, when constructing a knowledge database described in the form of rules, it is necessary to grasp the degree of "unsold", time series tendency, etc. in advance. Otherwise, for example, you will set an appropriate threshold for the degree of "unsold",
It will be unclear whether it will be based on actual data. On the other hand, in this embodiment, since the data pattern of the POS data to be monitored is automatically extracted from the past time series data, the same effect as that in which the value according to the actual data is automatically accumulated There is.

Further, in the conventional method, even if a system in which knowledge information is described in accordance with a presumed data pattern is realized, an unexpected pattern occurs when past data is not analyzed. In some cases it will be completely ineffective. On the other hand, in this embodiment, since the past time series data is automatically analyzed, there is no omission except for the pattern that has occurred for the first time, and an effective monitoring system can be realized.

The pattern recognition section 5 of the above embodiment has been described for the case where the pattern recognition is performed by determining the Euclidean distance from the target POS data with the average value of the clustered clusters as the data pattern. It is not limited to this. For example, a pattern recognition method using the composite similarity method may be used. In the case of this method, it is necessary to obtain the eigenvalues and eigenvectors of a plurality of data of each clustered cluster and store them in the pattern database 14. Furthermore, in the case of a pattern recognition method using a neural network, it is necessary to train the result of clustering with a backpropagation learning algorithm and store the resulting network in the pattern database 14.

[0036]

As described above in detail, according to the present invention, in the data monitoring system using the inference system, the data pattern generated in the time-series data of the monitoring target is automatically recognized, so that the data pattern It is possible to realize the construction of a knowledge database necessary for an inference system without necessarily being familiar with the knowledge about. Therefore, an inference system that obtains an appropriate inference result based on the data pattern generated in the time-series data of the monitoring target is enabled without requiring complicated analysis work to grasp the characteristics of the transition of the time-series data of the monitoring target. Can be demonstrated. If the present invention is applied to, for example, a sales information management system, it is possible to monitor related time series data and efficiently obtain effective inference results regarding sales management from this data pattern.

[Brief description of drawings]

FIG. 1 is a block diagram showing a basic configuration of a data monitoring system according to an embodiment of the present invention.

FIG. 2 is a block diagram when the data monitoring system of the embodiment is applied to a POS system.

FIG. 3 is a flowchart for explaining the operation of the embodiment.

FIG. 4 is a flowchart for explaining the operation of the embodiment.

FIG. 5 is a conceptual diagram showing an example of a display screen for explaining the operation of the data monitoring system of the same embodiment.

FIG. 6 is a conceptual diagram for explaining an example of POS data of the same embodiment.

FIG. 7 is a conceptual diagram for explaining an example of POS data of the same embodiment.

FIG. 8 is a conceptual diagram showing an example of a display screen for explaining the operation of the data monitoring system of the embodiment.

[Explanation of symbols]

1 ... POS database for learning, 2 ... pattern extraction unit,
3 ... Pattern output unit, 4 ... Target POS database, 5
... pattern recognition unit, 6 ... inference system, 7 ... knowledge input unit, 10 ... control unit, 11 ... input / output unit, 12 ... learning PO
S data extraction unit, 13 ... Target POS data extraction unit,
14 ... Pattern database, 15 ... Knowledge database, 16 ... Inference unit, 17 ... Inference result output unit.

Claims (5)

[Claims] 1. A pattern extracting means for extracting a data pattern of a predetermined format from a selected data set based on a predetermined method, and data to be monitored in the data patterns extracted by the pattern extracting means. A pattern recognition means for recognizing a data pattern similar to the pattern of the set, and a knowledge set given in advance based on the data pattern recognized by the pattern recognition means, for the data set to be monitored. A data monitoring system comprising: an inference executing means for outputting an inference result. 2. A learning database means for storing a learning time series data set for determining a data pattern of a predetermined format, and the learning database means selected from the learning database means based on a predetermined method. Pattern extracting means for extracting the data pattern of the time series data set, target database means for storing the time series data set to be monitored, and among the data patterns extracted by the pattern extracting means, from the target database means Pattern recognition means for recognizing a data pattern in which the selected time series data sets are similar to each other, and knowledge information for guiding an inference result based on a predetermined rule from the data pattern extracted by the pattern extraction means. Knowledge input means for inputting, and the knowledge input by this knowledge input means And an inference execution unit that outputs inference results for the time-series data set to be monitored based on the data pattern recognized by the pattern recognition unit. Data monitoring system. 3. A data monitoring system for managing sales information, comprising learning database means for storing a learning time series data set related to past sales information, and the learning based on a predetermined method. Pattern extracting means for extracting a data pattern of a predetermined format from the learning time-series data set selected from the database for database, pattern database means for storing the data pattern set extracted by the pattern extracting means, and a monitoring target And, the target database means that stores a time series data set related to the current sales information, and the time series data selected from the target database means in the data pattern set stored in the pattern database means. A pattern recognition means for recognizing data patterns whose sets are similar Knowledge input means for inputting knowledge information for inferring a sales management result according to the sales information for each of the data patterns extracted by the pattern extraction means based on a predetermined rule, and the knowledge input means. Knowledge database means that stores the knowledge information set input by, and executes inference processing by using the knowledge information set, and based on the data pattern recognized by the pattern recognition means, the sales information at the present time is obtained. A data monitoring system comprising: inference executing means for outputting a sales management result for a related time series data set. 4. The data monitoring system according to claim 2, wherein said knowledge input means has a display means for displaying said data pattern recognized by said pattern recognition means. 5. A data monitoring system having a learning database means for storing a learning time series data set for determining a data pattern of a predetermined format and a target database means for storing a time series data set to be monitored, A step of extracting a data pattern of the learning time-series data set selected from the learning database means based on a predetermined method; and a step of selecting from the target database means among the extracted data patterns. Recognizing a data pattern in which the time series data sets are similar to each other; inputting knowledge information for deriving an inference result based on a predetermined rule from the extracted data pattern; Perform inference processing using information,
Outputting an inference result for the time-series data set to be monitored based on the recognized data pattern.