JP6633403B2 - Analysis target determination apparatus and analysis target determination method - Google Patents

Description

  The present invention relates to an analysis target determination apparatus and an analysis target determination method for determining data to be analyzed from a large amount of various data.

  Analyzing a large amount of data (big data) is performed to derive knowledge useful for business. 2. Description of the Related Art In the field of web services such as search, electronic commerce, and social media, a large amount of data such as search histories, purchase histories of products and digital contents, settlement information, and communication histories are accumulated. Further, in a machine or a facility in a factory, various sensors accumulate various and large amounts of data on the state (temperature, pressure, flow rate, etc.) of the machine or facility. As one of such big data analysis methods, a relational expression (model) that holds for a plurality of data items (variables) is derived by multiple regression analysis, and a relationship between the data items is captured.

  Patent Literature 1 describes a method of predicting a process state by performing multiple regression analysis on data accumulated in plant equipment. In this method, data including a plurality of data items is sampled in time series from one plant, and data items (explanatory variables) used for multiple regression analysis are narrowed down by a stepwise method at each sampling time. I have.

JP 2012-128800 A

  For example, when analyzing a plurality of machines, data including a plurality of data items is sampled from each machine, and an analysis process is performed on the obtained data of each machine. If the machines that sampled data include different types of machines or machines that have malfunctions, the data sampled from these machines will be different from the data sampled from other machines. Have different tendencies. For this reason, if the stepwise method is applied to the data of all the machines, the data items cannot be properly selected, and the machines cannot be analyzed properly.

  The present invention has been made in view of such circumstances, and a main object thereof is to provide an analysis object determination apparatus and an analysis object determination method capable of determining data suitable for analysis from a large amount of various data.

  In order to solve the above-described problem, an analysis target determination apparatus according to an aspect of the present invention is configured such that when analyzing data including a plurality of data items, an analysis using a model including at least some of the data items as variables is performed. An analysis target determination apparatus that determines an object of the above, wherein a selection unit that selects a part of data that is different each time from a plurality of data, based on the data selected by the selection unit, a data item suitable for generating a model is Determining means for determining whether selection is possible or not, and determining means for determining data to be analyzed based on the determination result by the determining means after repeating the processing by the selecting means and the determining means a plurality of times. And.

  In this aspect, the analysis target determination device repeats the processing by the selection unit and the determination unit a plurality of times, and then determines a data item used for analysis based on a determination result by the determination unit. May be further provided.

  Further, in the above aspect, the determination unit may be configured to determine a data item used for analysis based on the number of times for each data item selected by the determination unit.

  Further, in the above aspect, the analysis target determination apparatus, when the determination unit determines that the data item is selectable, an evaluation unit that evaluates an error of a model generated using the selectable data item. The second determining means determines a data item to be used for analysis based on the evaluation result by the evaluating means after repeating the processing by the selecting means, the determining means, and the evaluating means a plurality of times. It may be configured to do so.

  Further, in the above aspect, the analysis target determination device may further include an output unit that outputs data of the analysis target determined by the determination unit.

  In the above aspect, the analysis target determination device may further include an output unit that outputs a data item used for the analysis determined by the second determination unit.

  In the above aspect, the analysis target determination device may further include a second output unit that outputs data different from the data to be analyzed as data that is not suitable for analysis.

  Further, in the above aspect, the plurality of data is time-series data, and the second output unit determines that the data item is selectable for the plurality of continuous data by the determination unit, and the next continuous When it is determined that the data item cannot be selected for a plurality of data, the first data of the next successive plurality of data may be output as abnormal data.

  In the above aspect, the determination unit may be configured to determine whether a data item suitable for generating the model can be selected by a stepwise method.

  Further, in the above aspect, the analysis target determination apparatus determines, when the determination unit determines that the data item is selectable a plurality of times, the data commonly selected by the selection unit as the same group. The image processing apparatus may further include third determining means, and third output means for outputting data determined as the same group by the third determining means.

  According to another aspect of the present invention, there is provided an analysis target determination method for determining an analysis target using a model including at least some data items as variables when analyzing data including a plurality of data items. Wherein the computer selects different data each time from the plurality of data, and the computer determines whether a data item suitable for generating a model can be selected based on the selected data. Determining the data to be analyzed based on the determination result in the determining step after the computer repeats the selecting step and the determining step a plurality of times. .

  According to the present invention, data suitable for analysis can be determined from a large amount of data.

FIG. 1 is a schematic diagram illustrating an example of an analysis system including an analysis target determination device. The schematic diagram which shows the other example of the analysis system including the analysis target determination apparatus. FIG. 2 is a block diagram illustrating a configuration of an analysis target determination device according to the first embodiment. 5 is a flowchart showing a procedure of an analysis target determination process by the analysis target determination device according to the first embodiment. The figure which shows an example of a processing result screen. The figure which shows the other example of a processing result screen. 5 is a graph showing data used in a performance evaluation test. A graph for explaining an example of data selection. 9 is a graph showing the estimation result of an objective variable. Graph when this model is applied to the entire time series data. 7 is a graph showing a selection rate for each data item when the stepwise method is successful. 9 is a graph showing the selectivity of data items when the evaluation value of the model error is small. 5 is a graph showing a selectivity for each data when the stepwise method fails. 9 is a flowchart illustrating a procedure of an analysis target determination process performed by the analysis target determination device according to the second embodiment. The figure which shows the example of a processing result screen. 13 is a flowchart showing a procedure of an analysis target determination process by the analysis target determination device according to the third embodiment. The figure which shows the example of a processing result screen.

  Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings.

(Embodiment 1)
The configuration of the analysis target determination device according to the present embodiment will be described. The analysis target determination device according to the present embodiment collects a large amount of various data. FIG. 1 is a schematic diagram illustrating an example of an analysis system including the analysis target determination device according to the present embodiment, and FIG. 2 is a schematic diagram illustrating another example. For example, when analyzing a production facility, as shown in FIG. 1, the analysis target determining apparatus 1 is connected to a plurality of sensors 2, 2,. The time series data is collected by receiving the detection signal from. FIG. 1 shows an analysis system for analyzing a rolled steel plate manufacturing facility, and the sensors 2, 2,... Include a vibration sensor, a speed sensor, a tension sensor, a thickness sensor, a pressure sensor, a temperature sensor, and the like. The various types of sensors are included. When analyzing a plurality of machines, as shown in FIG. 2, the analysis target determination apparatus 1 is connected to a plurality of sensors 21, 21,... Provided for each of the plurality of machines 3, 3,. Receive the detection signals from the sensors 21, 21,... To collect data indicating the states of a plurality of machines. 2 shows an analysis system for analyzing a plurality of welding robots. The sensors 21, 21,... Include temperature sensors, pressure sensors, angle sensors, speed sensors, acceleration sensors, current sensors, and the like. Various types of sensors are included. The data collected by the analysis target determining apparatus 1 is not limited to sensor information, but may be command value information or error information of a control system.

  In the case of an analysis system in the web service field such as search, electronic commerce, social media, and the like, the analysis target determination device 1 receives data such as search history, product / digital content purchase history, settlement information, and communication transmission history from the server. You may do so. Alternatively, the analysis target determination device 1 may read data from a recording medium on which the data obtained as described above is recorded.

  FIG. 3 is a block diagram illustrating a configuration of the analysis target determination device according to the present embodiment. The analysis target determination device 1 is realized by a computer 10. As shown in FIG. 3, the computer 10 includes a main body 11, a display unit 13, and an input unit 12. The main body 11 includes a CPU 111, a ROM 112, a RAM 113, a hard disk 115, a reading device 114, an input / output interface 116, an image output interface 117, and a communication interface 118. The CPU 111, the ROM 112, the RAM 113, the hard disk 115, the reading device 114, The output interface 116, the image output interface 117, and the communication interface 118 are connected by a bus.

  The CPU 111 can execute a computer program loaded on the RAM 113. When the CPU 111 executes the analysis target determination program 110 which is a computer program for the analysis target determination processing, the computer 10 functions as the analysis target determination device 1.

  The ROM 112 includes a mask ROM, a PROM, an EPROM, an EEPROM, or the like, and stores a computer program executed by the CPU 111, data used for the computer program, and the like.

  The RAM 113 is configured by an SRAM, a DRAM, or the like. The RAM 113 is used to read the analysis target determination program 110 recorded on the hard disk 115. When the CPU 111 executes a computer program, it is used as a work area for the CPU 111.

  The hard disk 115 has installed therein various computer programs to be executed by the CPU 111, such as an operating system and application programs, and data used for executing the computer programs. The analysis target determination program 110 is also installed on the hard disk 115.

  The input / output interface 116 includes, for example, a serial interface such as USB, IEEE1394, or RS-232C, a parallel interface such as SCSI, IDE, or IEEE1284, and an analog interface including a D / A converter and an A / D converter. It is configured. The input unit 12 including a keyboard and a mouse is connected to the input / output interface 116, and the user can input data to the computer 10 by using the input unit 12.

  The image output interface 117 is connected to the display unit 13 composed of an LCD, a CRT, or the like, and outputs a video signal corresponding to image data provided from the CPU 111 to the display unit 13. The display unit 13 displays an image (screen) according to the input video signal.

  The communication interface 118 is for performing data communication with an external device. When receiving data by communication, the communication interface 118 is connected to various sensors or a server that stores data. For example, in the case of a rolled steel plate manufacturing facility, the communication interface 118 is connected to sensors 2, 2,... Such as a vibration sensor, a speed sensor, a tension sensor, a thickness sensor, a pressure sensor, and a temperature sensor. Is output by the communication interface 118.

[Operation of analysis target determination device]
Hereinafter, the operation of the analysis target determination device 1 according to the present embodiment will be described.

  FIG. 3 is a flowchart illustrating a procedure of an analysis target determination process by the analysis target determination device 1 according to the present embodiment.

  When starting the analysis target determination process, the CPU 111 randomly selects some data from the data (step S101).

ここで、Ｎ，Ｍは１より大きい整数であり、ｋは１≦ｋ≦Ｎ，Ｍを満たす整数である。 The data collected by the analysis target determination device 1 is represented as follows.

Here, N and M are integers larger than 1, and k is an integer satisfying 1 ≦ k ≦ N and M.

In the equation (1), Y is a target variable, and X is a variable. The k-th data from the top is (y _k , x _k1 , x _k2 ,..., X _kM ). In the process of step S101, the CPU 111 randomly selects a plurality of k. For example, as shown below, the CPU 111 selects the second, p-th, q-th,... (P and q are integers satisfying 1 ≦ p, q ≦ M).

Next, the CPU 111 applies the stepwise method to the data (Y _si , X _si ) selected in step S101, and data items (variables) suitable for generating the model (linear model) shown in equation (3) Is selected (step S102). Here, i is an integer of 1 or more, and the initial value is 1. In the process of step S102, some variables suitable for model generation are selected from the N variables of _Xsi .

  The processing in step S102 will be described in more detail. In step S102, the CPU 111 applies the stepwise method to the selected data. Regarding the stepwise method, Yutaka Tanaka and Kazumasa Wakimoto, “Multivariate Statistical Analysis”, 9th edition, Gendai Mathematics, Feb. 20, 1994, p. 42-47 (hereinafter referred to as “Reference Document 1”) and the like. Hereinafter, the stepwise method will be described. In the following description of the stepwise method, the notation of parameters used in equations (4) and (5) matches the notation of parameters described in Reference Document 1.

  (Step I) First, a variable having the largest single correlation with the objective variable y is selected. That is, a model (regression equation) is generated by sequentially adopting variables (data items) one by one, and the variable having the largest absolute value or F value (see the following equation) of t for regression coefficient test is selected. Test the hypothesis that the regression coefficient for the selected variable is 0, and do not include any variables in the model unless rejected. If rejected, this variable is taken into the model and the process proceeds to step II.

  (Step II) In addition to the variables already in the model, the remaining variables are adopted one by one in order, and the variable with the maximum partial correlation (the absolute value of t or the F value for the regression coefficient test is the maximum) is determined. Choose. The hypothesis test that the regression coefficient is 0 is performed, and if not rejected, the process ends. If rejected, the selected variables are taken into the model and the process proceeds to step III.

  (Step III) The model is calculated and the regression coefficient is tested for each variable. If the hypothesis is not rejected for the variable with the smallest F value, the variable is excluded from the model.

  (Step IV) If all variables have been taken into the model, the process ends. Otherwise, return to step II.

ここで、ｔ_α（ｎ−ｐ−１）は自由度（ｎ−ｐ−１）のｔ分布の両側１００α％点の限界値（１００×（１−α)％の信頼区間の境界値）を、ｎはステップＳ１０１において選択したデータ個数（すなわち、Ｙ_ｓｉの要素数あるいはＸ_ｓｉの行数）を、ｓ^jjは誤差の分散の不偏推定値を、Ｖ_eはｐ個の変数を含むモデルでの説明変数の分散共分散行列において、その逆行列のｊ番目の対角要素の値をそれぞれ示している。 In the above-mentioned stepwise method, the test of the regression coefficient is performed as follows. The test of the hypothesis H ₀ : a _j = 0 that the regression coefficient for the variable x _j in the model including p variables is 0 is obtained by calculating the t value by the following equation to obtain the t distribution with n-p−1 degrees of freedom. compared with the limit _{value, | t | ≧ t α (} n-p-1) if reject the hypothesis, | t | adopt _{<t α (n-p-} 1) if the hypothesis.

Here, t _α (n-p-1) is the limit value (100 × (1-α)% confidence interval boundary value) at the 100α% point on both sides of the t distribution of the degree of freedom (n-p-1). , N is the number of data selected in step S101 (that is, the number of elements of Y _si or the number of rows of X _si ), s ^jj is an unbiased estimate of error variance, and V _e is a model including p variables. In the variance-covariance matrix of the explanatory variable, the value of the j-th diagonal element of the inverse matrix is shown.

Instead of the above-described regression coefficient test, F in the following equation is obtained from the relationship between the t distribution and the F distribution, and is compared with the limit value of the F distribution having (1, n−p−1) degrees of freedom. If ≧ F ¹ _n−p−1 (α), the hypothesis may be rejected.

  Here, variables suitable for model generation were selected by the stepwise method.However, methods other than the stepwise methods such as the brute force method, forward selection method, backward elimination method, best subset regression, partial least squares method, and principal component regression were used. Variables may be selected by a variable selection method.

ここで、||●||_２はユークリッドノルムを、Ｔは行列又はベクトルの転置を示す。 Next, the CPU 111 determines whether a significant model can be generated, that is, whether a combination of data and data items suitable for generating the model has been successfully selected (step S103). If it is determined that a significant model can be generated (YES in step S103), CPU 111 generates a model using the data item selected in step S102 in the data selected in step S101, and sets the actual _Is compared with the value Y _{mi of the} objective variable obtained from the model, and the error of the model is evaluated (step S104). In step S104, CPU 111 calculates the evaluation value _{J i.} The evaluation value J _i can be, for example, the square root of the sum of squares of the difference between Y _si and Y _mi .

Here, || ● || ₂ indicates the Euclidean norm, and T indicates the transpose of a matrix or vector.

  After the above step S104, the CPU 111 stores the number of the data selected at that time (2, p, q,... In the above example) and the number of the data item selected by the stepwise method on the hard disk 115. (Step S105). After step S105, CPU 111 causes the process to proceed to step S107.

In step S103, if the model generation is determined to not (NO in step S103), CPU 111 sets the evaluation value _{J i} of the error models ∞ (step S106). If you can not express the infinite, it may be set sufficiently large number of J _i. After step S106, CPU 111 proceeds with the process to step S107.

  In step S107, the CPU 111 determines whether or not the processes of S101 to S106 have been repeated a specified number of times. If it has not been repeated the specified number of times (NO in step S107), CPU 111 increments variable i by 1 (step S108), and returns to step S101. Thereby, the processing of steps S101 to S106 is repeated a specified number of times.

When the process of step S101~S106 is repeated predetermined number of times (YES in step S107), CPU 111 selects the top 1% of small evaluation values _{J i,} the evaluation value of the top 1% selected is obtained In this case, the selected data items are totaled (step S109). Incidentally, etc. to 1% rather than the top 5% higher percentage of the evaluation value J _i to be selected can be set as appropriate.

  Among the data items totaled in step S109, the data item whose number of selections is large can be considered as a factor affecting the objective variable Y, that is, a data item suitable for analysis. The CPU 111 determines, for example, among the data items totaled in step S109, data items selected at least a preset number of times as data items suitable for analysis, and stores the numbers of these data items on the hard disk 115. (Step S110).

When it is determined in step S103 that model generation is not possible (that is, when J _i = ∞), the CPU 111 totals the number of times selected for each data (step S111). When certain data is selected and the number of times that it is determined that generation of a model is impossible is large, the data can be considered to be data unsuitable for analysis. On the other hand, data other than data that is not suitable for analysis can be considered to be data that can be analyzed. When it is determined that the model cannot be generated, the CPU 111 determines the data selected a predetermined number of times or more as data that is not suitable for analysis, determines other data as data to be analyzed, and performs analysis. The numbers of the unsuitable data and the data to be analyzed are stored in the hard disk 115 (step S112).

  The CPU 111 causes the display unit 13 to display the processing result screen of the analysis target determination processing as described above (step S113), and ends the analysis target determination processing.

  FIG. 5 is a diagram illustrating an example of the processing result screen. The processing result screen 30 in this example includes the number of data determined to be analyzed and the number of a data item suitable for analysis. Thereby, the user can confirm the data to be analyzed and the data items suitable for the analysis. In addition, a large amount of data can be analyzed by generating a model using data items suitable for analysis in data to be analyzed.

  FIG. 6 is a diagram illustrating another example of the processing result screen. The processing result screen 31 of this example includes a data item number suitable for analysis and a data number not suitable for analysis. Data that is not suitable for analysis may include abnormal data. By outputting data that is not suitable for analysis, data that may be abnormal can be notified to the user. In addition, data other than data that is not suitable for analysis can be set as an analysis target, and a large amount of data can be analyzed by generating a model using data items suitable for analysis in data to be analyzed.

(Performance evaluation test)
The inventor performed a performance evaluation test of the analysis target determination method according to the first embodiment.

  FIG. 7 is a graph showing data used in the performance evaluation test. In FIG. 7, the vertical axis represents the value of the target variable Y, and the horizontal axis represents time. This data is time-series data, and one data (that is, data at one time point) includes 20 data items including time information. The inventor applied the stepwise method to the entire time-series data, but could not select a data item suitable for generating a model.

  Next, the inventor randomly selected 50% of the time-series data. FIG. 8 is a graph for explaining an example of data selection. In FIG. 8, a mark indicates data selected in a certain case, and a mark indicates data not selected in the case.

When the stepwise method was applied to the selected data, eight data items (variables) were selected from the twenty data items. A model was generated using these eight variables, and the value of the objective variable Y was estimated for the selected data. FIG. 9 is a graph showing the estimation result of the objective variable Y. In FIG. 9, in order to clarify the relationship between time and the objective variable, the value of the objective variable is corrected with the values of other explanatory variables, and the corrected value of the objective variable is plotted for each corresponding time. are doing. That is, if the regression model is represented by equation (7), when x ₁ is the time information corresponds to the vertical axis of FIG. 9 is the left side of equation (8).

  As shown in FIG. 9, it can be seen that a highly accurate regression equation is obtained. FIG. 10 is a graph when the present model is applied to the entire time-series data. In FIG. 10, a mark indicates selected data, and a mark indicates data not selected. The data not selected includes data that does not conform to the model, and it can be seen that there is a possibility that the attribute is different from the selected data.

  The above processing was repeated 10,000 times, and the selectivity of each data item when the stepwise method was successful (that is, when the data item could be selected by the stepwise method) was obtained. FIG. 11 is a graph showing the selectivity for each data item when the stepwise method is successful. In FIG. 11, the vertical axis indicates the selectivity, and the horizontal axis indicates the data item number. As shown in FIG. 11, the selectivity of the second data item is remarkably high and is considered to be an important data item. The sixth, eighth, and seventeenth data items also have high selection rates and are considered to be important data items.

  In addition, the inventor evaluated the model error in each case where the stepwise method was successful, and investigated data items that were commonly selected for models with small errors. FIG. 12 is a graph showing the selectivity of data items when the evaluation value of the model error is small. The upper graph in FIG. 12 shows the result when the evaluation value is the root mean square of the error, the middle graph shows the result when the evaluation value is the L1 norm of the error (average absolute value), and the lower graph. Indicates the result when the evaluation value is the ∞ norm (maximum value) of the error. In each graph, the vertical axis indicates the selectivity, and the horizontal axis indicates the number of the data item.

  From FIG. 12, it can be seen that the sixth, sixteenth, eleventh, etc. data items are commonly selected for models with small errors, and are considered important data items.

  In addition, the inventor calculated the selectivity of each data when the stepwise method failed (that is, the data item could not be selected by the stepwise method). FIG. 13 is a graph showing the selectivity for each data when the stepwise method fails. In FIG. 13, the vertical axis indicates the selectivity, and the horizontal axis indicates the data number. As shown in FIG. 13, when the eleventh data is selected, the stepwise method has failed with a probability of about 80%. This data is highly likely to be abnormal data, and it can be seen that this data is data to be excluded from the analysis target.

  As described above, the analysis target determination method according to Embodiment 1 can determine the analysis target data and data items for deriving the relationship with high accuracy even from a data group in which a relationship cannot be found at first glance. .

(Embodiment 2)
The analysis target determination device according to the present embodiment performs processing on the time-series data, and estimates the time when an abnormality has occurred. The configuration of the analysis target determining apparatus according to the present embodiment is the same as the configuration of the analysis target determination apparatus 1 according to the first embodiment, and therefore, the same components are denoted by the same reference numerals and description thereof is omitted.

  The operation of the analysis object determination device according to the present embodiment will be described. FIG. 14 is a flowchart illustrating a procedure of an analysis target determination process performed by the analysis target determination device according to the present embodiment.

  The processing in steps S101 to S112 is the same as that described in the first embodiment, and a description thereof will not be repeated.

  After the process of step S112, the CPU 111 determines that the model can be continuously generated for a predetermined number of times or more until a certain time, based on the determination result of whether or not the model can be generated at each time. The time at which it is determined that the model cannot be generated continuously for the number of times or more is detected (step S213). In the time-series data, when the state changes from a state in which a model can be generated across such a certain time to a state in which a model cannot be generated, it can be considered that an abnormality has occurred at that time. Hereinafter, the time is referred to as “abnormality occurrence time”.

  The CPU 111 causes the display unit 13 to display the processing result screen of the analysis target determination processing as described above (step S214), and ends the analysis target determination processing. FIG. 15 is a diagram illustrating an example of the processing result screen. When the abnormality occurrence time is detected in step S213, the processing result screen 32 displayed in step S214 includes information on the abnormality occurrence time (the abnormality occurrence time and the data number at that time). This allows the user to be notified of the time at which it is estimated that an abnormality has occurred. Further, the user can specify which data of the time-series data has an abnormality.

(Embodiment 3)
The analysis target determining apparatus according to the present embodiment performs grouping of data determined as an analysis target. The configuration of the analysis target determining apparatus according to the present embodiment is the same as the configuration of the analysis target determination apparatus 1 according to the first embodiment, and therefore, the same components are denoted by the same reference numerals and description thereof is omitted.

  The operation of the analysis object determination device according to the present embodiment will be described. FIG. 16 is a flowchart illustrating a procedure of an analysis target determination process performed by the analysis target determination device according to the present embodiment.

  The processing in steps S101 to S112 is the same as that described in the first embodiment, and a description thereof will not be repeated.

  After the process of step S112, when it is determined that the model can be generated a plurality of times, the CPU 111 specifies a plurality of data that are commonly selected when the model is determined to be able to be generated, and A plurality of data are determined as the same group (step S313).

  For example, in the case of data obtained from a plurality of welding robots, differences in the configuration of the welding robot (for example, differences in the size of the robot, differences in the hand torch, whether the robot is stationary on the ground or mounted on the slider), differences in the application ( The tendency of the characteristics of the welding robot differs depending on the difference in the thickness of the welding material, the difference in the joint, the difference in the way of weaving, and the like. For this reason, if data to be determined as to whether or not a model can be generated includes data collected from a plurality of types of welding robots having different configurations, uses, and the like, it is determined that the model cannot be generated. Often. On the other hand, the data selected when it is determined that the model can be generated can be estimated to be collected from welding robots having a common configuration, application, and the like. Therefore, data can be grouped for each tendency by the process of step S313 described above. By analyzing the data of each group, it is possible to grasp the tendency of the data for each group.

  The CPU 111 causes the display unit 13 to display the processing result screen of the analysis target determination processing as described above (step S314), and ends the analysis target determination processing. FIG. 17 is a diagram illustrating an example of the processing result screen. On the processing result screen 33 displayed in step S314, the numbers of the data included in the groups divided in step S313 are displayed for each group. This allows the user to be notified of the grouping result. The user can specify the data included in each group, and can analyze the data for each group.

(Other embodiments)
In the above-described first to third embodiments, the configuration in which the stepwise method is applied to data (Y _si , X _si ) randomly selected from all data has been described, but the present invention is not limited to this. . Based on the result of the determination as to whether or not it is possible to generate a model before the previous time, data that is estimated to be unsuitable for generating a model may be excluded from the target to which the stepwise method is applied. As a result, the data search range can be narrowed, and the amount of calculation in the analysis target determination processing can be reduced.

  Further, in the above-described first to third embodiments, the configuration in which a data item suitable for generating a model is selected from all data items by the stepwise method has been described, but the present invention is not limited to this. Based on the processing results of the stepwise method before the previous time, data items that are estimated to be unsuitable for generating a model may be excluded from the application of the stepwise method. This also makes it possible to narrow the data search range and reduce the amount of calculation in the analysis target determination processing.

  In Embodiments 1 to 3 described above, the analysis target determination apparatus is configured by one computer, but is not limited to this. The analysis target determination apparatus may be configured by a plurality of computers connected to each other so as to be able to perform data communication, and each computer may execute the stepwise method for a different combination of data.

  INDUSTRIAL APPLICABILITY The analysis target determination apparatus and the analysis target determination method of the present invention are useful as an analysis target determination apparatus and an analysis target determination method for determining data to be analyzed from a large amount of various data.

Reference Signs List 1 analysis target determination device 10 computer 11 main body 12 input unit 13 display unit 110 device data processing program 111 CPU
115 hard disk 116 input / output interface 117 image output interface 118 communication interface

Claims (11)

When analyzing data including a plurality of data items, an analysis target determination device that determines an analysis target using a model including at least some data items as variables,
Selecting means for selecting some different data each time from a plurality of data;
A determination unit that determines whether a data item suitable for generating a model is selectable based on the data selected by the selection unit;
After repeating the processing by the selection unit and the determination unit a plurality of times, a determination unit that determines data to be analyzed based on a determination result by the determination unit,
An analysis target determination device comprising: After repeating the processing by the selection unit and the determination unit a plurality of times, further includes a second determination unit that determines a data item used for analysis based on a determination result by the determination unit,
The analysis target determination device according to claim 1. The determining unit is configured to determine a data item used for analysis based on the number of times for each data item selected by the determining unit.
The analysis target determination device according to claim 2. When the data item is determined to be selectable by the determination unit, further includes an evaluation unit that evaluates an error of a model generated using the selectable data item,
The second determining unit is configured to determine a data item used for analysis based on an evaluation result by the evaluating unit after repeating the processing by the selecting unit, the determining unit, and the evaluating unit a plurality of times. Have been
The analysis target determination device according to claim 2. Further comprising an output unit that outputs data of the analysis target determined by the determination unit,
The analysis target determination device according to claim 1. Output means for outputting a data item used for the analysis determined by the second determination means,
The analysis target determination device according to claim 2. The apparatus further includes a second output unit that outputs data different from the data to be analyzed as data that is not suitable for analysis,
The analysis target determination device according to claim 1. The plurality of data is time-series data,
The second output unit, when the determination unit determines that the data item is selectable for a plurality of continuous data, and, when it is determined that the data item is not selectable for the next plurality of continuous data, It is configured to output the leading data of the next plurality of continuous data as abnormal data,
The analysis target determination device according to claim 7. The determining means is configured to determine whether a data item suitable for generating the model can be selected by a stepwise method,
An analysis target determination device according to claim 1. A third determination unit that determines, when the determination unit determines that the data item is selectable a plurality of times, the data commonly selected by the selection unit as the same group;
Third output means for outputting data determined as the same group by the third determination means,
Further comprising,
The analysis target determination device according to claim 1. When analyzing data including a plurality of data items, an analysis target determination method for determining an analysis target using a model including at least some data items as variables,
A step in which the computer selects different data from the plurality of data each time;
The computer determines, based on the selected data, whether a data item suitable for generating a model can be selected, and
The computer, after repeating the selecting step and the determining step a plurality of times, based on the determination result in the determining step, determining data to be analyzed,
An analysis target determination method, comprising:

Images