JP6961312B2 - State change detection auxiliary device, state change detection device, state change detection auxiliary program, and state change detection program - Google Patents

Description

The present invention relates to a state change detection auxiliary device, a state change detection device, a state change detection auxiliary program, and a state change detection assisting program.

For example, when a state change such as an abnormality is detected by measuring the current, voltage, and vibration of a machine, the abnormality is generally determined based on the past experience of a person or the like. That is, the threshold value of the state change is set by a person, and it is almost impossible for a machine having no abnormal data to detect the state change.

Patent Document 1 discloses a system for predicting the arrival status of the radio wave when an abnormal propagation of the radio wave occurs. This system uses the meteorological information importing means for capturing meteorological information and the meteorological information captured by this meteorological information capturing means to at least information on temperature at each altitude, pressure at each altitude, and humidity at each altitude. One is provided with a meteorological data extraction means for extracting the meteorological data. Furthermore, regarding the arrival of radio waves on the day when the weather data was extracted by the weather data extraction means, a prediction function for predicting abnormal propagation of the radio waves calculated based on the measurement data accumulated in advance is accumulated. Meteorological data extracted by the meteorological data extracting means is input to the accumulated prediction function to determine a predicted value which is the probability of occurrence of abnormal propagation of the radio wave.

Patent Document 2 describes an acquisition step of acquiring time-series data of a sensor possessed by the machine-equipment for each predetermined production unit of the product from a machine-equipment capable of producing the product, and a plurality of divisions of the time-series data. In the division step of generating the divided data group divided into data, the first additional information which is additional information indicating that there is no failure is associated with the predetermined normal quality portion included in the divided data group, and the divided data group is associated with the first additional information. A quality abnormality detection method including a setting process for unconditionally associating a second additional information indicating that there is a failure with respect to the latest production amount included in the above is disclosed.

This quality abnormality detection method further corresponds to the input data by a grouping step of forming two groups in which the normal quality component and the latest production component are mixed, and learning using one of the two groups. In the generation step of generating the prediction model for predicting the additional information, in the prediction result of the prediction model obtained by using the other of the two groups as the input data, the additional information is the second additional information. It has a calculation process for calculating the predicted probability, and determines the quality abnormality of the product based on the probability.

Further, Patent Document 3 discloses a data prediction method for predicting the future based on past achievements. In this data prediction method, the prediction model includes a first prediction model that outputs a prediction value for at least one feature quantity on the prediction target date and a prediction value output from this first prediction model as input factors, and is a prediction target. It is composed of a second prediction model that outputs a predicted value for each predetermined time of the day.

Then, in the data prediction method of Patent Document 3, a prediction model construction means for constructing a prediction model using the collected closest actual data and past actual data, and a prediction input data for input to the constructed prediction model for prediction. The prediction execution means for obtaining the prediction value, the prediction error calculation means for calculating the prediction error or the model error from the collected nearest actual data and the prediction value, and the prediction value based on the prediction error or the model error. The correction coefficient or the correction amount to be corrected is calculated, and the axial correction predicted value is obtained.

FIGS. 1 (A) and 1 (B) are explanatory views for showing a method for detecting state fluctuations in a random forest. On the other hand, in a random forest, a state change detection device is known in which a normal case is indicated by "1" and an abnormal case is indicated by "2" by the boundary line of the region as shown in FIG. 1 (A). Has been done. That is, in this state fluctuation detection device, the region where the value in the vertical axis direction is 160 or more and the value in the horizontal axis direction is smaller than 45 is all normal "1", and in the region where the value in the vertical axis is smaller than 160, the horizontal axis If the value in the direction is smaller than 30, it is a normal "1", and if the value in the horizontal axis direction is 30 or more, it is an abnormal "2". Further, the region where the value in the vertical axis direction is 160 or more and the value in the horizontal axis direction is larger than 45 is abnormal "2" when the value in the vertical axis direction is smaller than 260, and the value in the vertical axis direction is 260 or more. If there is, it becomes a normal "1". In FIG. 1 (A), the plain white region is a normal “1”, and the satin-patterned region is an abnormal “2”.

When the above-mentioned areas are divided into abnormal and normal judgments, it is possible to realize a state change detection device by machine learning that makes predictions by branching (for example, in a tree structure) by a classifier. In such a state change detection device, for example, when the analysis target data D as shown in FIG. 1 (B) is generated, according to the state change detection device for the region of FIG. 1 (A), it is in the vertical direction. It corresponds to a region where the value is 160 or more and the value in the horizontal axis direction is smaller than 45, and is classified as a normal "1" as shown in the figure. However, since the value in the vertical axis direction is approximately 10,000, it is far from the normal vertical axis value 400 and the like as shown by the arrow, and it is clearly anomalous "2" for analysis target data. Yes, there is a misclassification.

Further, FIGS. 2A and 2B are explanatory diagrams for showing the problems of the method when the state change detection is performed by the random forest. When the regions shown in FIGS. 2 (A) and 2 (B) are separated by a boundary line, it is assumed that the objective variable is defined by a dummy variable as shown in FIG. 2 (A). That is, when the dummy variable is "1" in the plain white area, the dummy variable is "99" in the satin pattern area, and the dummy variable is "100" in the plain gray area, FIG. 2 (B). ), It is assumed that the data to be analyzed is obtained.

The average and error of the numerical values in each region in the case of FIG. 2 (B) are as described in FIG. 2 (B). That is, in the plain white area, the numerical average is "43.4" and the error is "42.4", and in the pear-skin pattern area, the numerical average is "99.4" and the error is "0.4". In the plain gray area, the numerical average is "100.0" and the error is "0". If there is a difference in the size of the dummy variables in this way, the error will be significantly different. Therefore, there is a problem that appropriate detection cannot be performed even if an attempt is made to capture a state change by comparing errors.

In Patent Document 4, a two-dimensional plane is considered as a space for acquiring signals, and the data used for machine learning and identification is multidimensional information having a three-dimensional structure in total of distribution information and spectral information in the plane. The procedure for machine learning and identification when using this multidimensional information is essentially the same as when using spectral data. Further, in this case, it is possible to acquire a plurality of feature quantities suitable for describing a data pattern and then use them as feature vectors for machine learning and identification processing, as a typical example of feature quantities. Is stated to have volume, curvature, spatial gradient, HLAC (higher-order local autocorrelation), and the like.

Then, it is also possible to select the feature amount used for machine learning in advance. In this case, for example, it is stated that the Mahalanobis distance may be calculated and the feature amount used for identification may be selected. The larger the Mahalanobis distance, the easier it is to identify, so it is possible to preferentially select features that increase the Mahalanobis distance between each group of interest.

That is, the description in Patent Document 4 states that when selecting a feature amount that facilitates machine learning and identification processing, a feature amount that increases the Mahalanobis distance is selected.

Further, in column 0117 of Patent Document 5, "When a decision tree is used as the above-mentioned predetermined algorithm, the degree of relevance (importance or contribution) of each feature amount (explanatory variable) is determined by a random forest method or the like. (It may be called) can be calculated. Specifically, when creating a decision tree, each feature amount (explanatory variable) so as to reduce the impureness indicated by the Gini coefficient, cross-entropy, etc. Therefore, the node (branch) is selected from the above. Therefore, the amount of decrease in purity when each feature amount (explanatory variable) is selected as the node can be used as the importance of the feature amount (explanatory variable). In step S102, the importance may be used to select a feature amount effective for predicting the occurrence of an abnormality from a plurality of types of feature amounts. " Shows selection by importance as described with reference to FIG.

According to Patent Document 6, if the calculated Mahalanobis distance D2i is within "1 + 3 * σ", it is determined that there is no change in the state of the manufacturing apparatus with a risk rate of 0.3%, while "1 + 3". If it is * σ "or more, it is disclosed that it is judged that there is a change in the state of the manufacturing equipment or there is a sign of abnormality in the operation at a risk rate of 0.3%. In this system, the value of D2INI is in the range of 3σ to 3.5σ from the standard deviation σ of the reference space.

Japanese Unexamined Patent Publication No. 2005-315753 Japanese Unexamined Patent Publication No. 2018-147406 Japanese Unexamined Patent Publication No. 2004-94437 Japanese Unexamined Patent Publication No. 2015-52581 JP-A-2018-116545 Japanese Unexamined Patent Publication No. 2000-252179

In this embodiment, even when the number of acquisition sources for acquiring the analysis target data is small, the state change detection assisting device can detect the state change with an accuracy comparable to that when the number of acquisition sources is large. , A state change detection device, a state change detection auxiliary program, and a state change detection program.

The state change detection assisting device according to the present embodiment has the first objective variable data including the first objective variable data and the first explanatory variable data, and the first objective variable data in the first machine learning data. A second machine learning that creates a second machine learning data in which different data is used as the second objective variable data and data different from the first explanatory variable data in the first machine learning data is used as the second explanatory variable data. Data creation means, the objective variable is predicted from the explanatory variables, the first prediction model generated using the teacher data, the objective variable is predicted from the explanatory variables, and the second machine learning data creation from the teacher data. The first analysis target data including the second prediction model generated by using the second teacher data created by the means, the first objective variable data and the first explanatory variable data acquired from a plurality of acquisition sources is described above. The first predicted value acquisition means that applies to one prediction model and obtains the first predicted value data, and the second analysis target data obtained by applying the first analysis target data to the second machine learning data creation means. It is the difference between the second predicted value acquisition means that applies to the second predicted model and obtains the second predicted value data, and the first objective variable data and the first predicted value data in the first analysis target data. A prediction error value calculating means for obtaining a first prediction error value, a second prediction error value which is a difference between the second objective variable data and the second prediction value data in the second analysis target data, and the above. It is characterized by including an output control means for outputting the result calculated by the prediction error value calculation means to the output means.

Explanatory drawing for showing the method in the case of performing state change detection by a random forest. Explanatory drawing to show the problem of the method when the state change detection is performed by a random forest. The block diagram of the computer system which realizes the state change detection auxiliary device 1A and the state change detection device 1B which concerns on embodiment of this invention. The functional block diagram of the state change detection auxiliary device 1A and the state change detection device 1B which concerns on embodiment of this invention. The figure for demonstrating the process by the 2nd machine learning data creation means by the state change detection auxiliary device 1A and the state change detection device 1B which concerns on embodiment of this invention. The figure for demonstrating another processing method by the 2nd machine learning data creation means by the state change detection auxiliary device 1A and the state change detection device 1B which concerns on embodiment of this invention. The figure which shows the result of having performed the error extension by the state change detection auxiliary device 1A and the state change detection device 1B which concerns on embodiment of this invention. The figure which shows the result of having performed the three-dimensional output by the state change detection auxiliary device 1A and the state change detection device 1B which concerns on embodiment of this invention. The process related to the generation of the first prediction model and the second prediction model by the state change detection auxiliary device 1A and the state change detection device 1B and the process of obtaining the first importance and the second importance according to the embodiment of the present invention are shown. flowchart. The flowchart which shows the processing by the state change detection auxiliary apparatus 1A and the state change detection apparatus 1B which concerns on embodiment of this invention.

Hereinafter, embodiments of the state change detection auxiliary device 1A, the state change detection device 1B, the state change detection auxiliary program, and the state change detection program according to the present invention will be described with reference to the accompanying drawings. In each figure, the same components are designated by the same reference numerals, and duplicate description will be omitted. FIG. 3 is a configuration diagram of a computer system that realizes the state change detection auxiliary device 1A and the state change detection device 1B according to the embodiment of the present invention. The state change detection auxiliary device 1A and the state change detection device 1B according to the embodiment of the present invention can be configured by, for example, a personal computer, a workstation, or another computer system as shown in FIG. In this computer system, the CPU 10 controls each part based on the program or data stored in the main memory 11 or read into the main memory 11 and executes necessary processing to execute the state change detection auxiliary device 1A or the state change detection. It operates as a device.

An external storage interface 13, an input interface 14, a display interface 15, and a data input interface 16 are connected to the CPU 10 via a bus 12. A program such as a state change detection program and an external storage device 23 in which necessary data and the like are stored are connected to the external storage interface 13. An input device 24 such as a keyboard as an input device for inputting commands and data and a mouse 22 as a pointing device are connected to the input interface 14.

A display device 25 having a display screen such as an LED or an LCD is connected to the display interface 15. Sensors 26-1, 26-2, ..., 26-m for obtaining measurement data are connected to the data input interface 16. This computer system may be provided with other configurations, and the configuration of FIG. 3 is only an example.

FIG. 4 is a functional block diagram of the state change detection auxiliary device 1A and the state change detection device 1B according to the embodiment of the present invention. In the above, in the CPU 10, each means and the like shown in FIG. 4 are realized by the state change detection assisting program in the external storage device 23. That is, the prediction model creation means 31, the second machine learning data creation means 32, the first prediction value acquisition means 33, the second prediction value acquisition means 34, the prediction error value calculation means 35, the output control means 36, and the error value extension means. 37, the dummy numerical value changing means 38 is realized. Further, the teacher data T is stored in the external storage device 23. The prediction model creating means 31 creates the first prediction model 30A using the teacher data T. Further, in the present embodiment, the prediction model creating means 31 includes the prediction model creating means 31 for creating the first prediction model 30A, but uses the first prediction model 30A created by another device or program. In this case, the prediction model creating means 31 may not be provided.

The first prediction model 30A predicts the objective variable from the explanatory variables by machine learning. Here, as a machine learning algorithm, a random forest of pattern mining can be mentioned, but in addition to this, prediction is made by branching with a classifier (for example, in a tree structure) such as a classification tree or a regression tree. It is possible to adopt the algorithm by machine learning to be performed. Further, the first prediction model 30A may perform machine learning by multiple regression analysis.

The second machine learning data creating means 32 is different from the first objective variable data in the first machine learning data with respect to the first machine learning data including the first objective variable data and the first explanatory variable data. The data is used as the second objective variable data, and the data different from the first explanatory variable data in the first machine learning data is used as the second explanatory variable data to create the second machine learning data.

Further, the second machine learning data creating means 32 is in the teacher data T when the given first machine learning data is the teacher data T including the first objective variable data and the first explanatory variable data. The data different from the first objective variable data is referred to as the second objective variable data, and the data different from the first explanatory variable data in the teacher data T is referred to as the second explanatory variable data. Then, the second machine learning data creating means 32 has the relationship between the second objective variable data and the second objective variable data in which the predicted value corresponding to the second objective variable data can be derived from the second explanatory variable data by the second predicted model. The second machine learning data including the explanatory variable data is created. This second machine learning data is new teacher data. Therefore, it can be said that the second machine learning data creating means 32 creates new second teacher data using the teacher data T as the first teacher data.

Further, the second machine learning data creating means 32 is when the first machine learning data is the first analysis target data including the first objective variable data and the first explanatory variable data acquired from a plurality of acquisition sources. The second objective variable data is data different from the first objective variable data in the first analysis target data, and the second explanatory variable is data different from the first explanatory variable data in the first analysis target data. As the data, new second analysis target data is created as follows. That is, the second machine learning data creating means 32 has the relationship between the second objective variable data and the second objective variable data in which the predicted value corresponding to the second objective variable data can be derived from the second explanatory variable data by the second predicted model. 2 A new second analysis target data including the explanatory variable data is created.

Here, the acquisition source is referred to as an acquisition source when the explanatory variables are acquired / collected using several sensors. Further, when the explanatory variables are acquired / collected using several devices, each device is referred to as an acquisition source. In other words, the acquisition source is used in the sense that it is the source for acquiring the objective variable and the explanatory variable.

A specific example of creating the second analysis target data from the first analysis target data by the second machine learning data creating means 32 will be described. For example, it is assumed that there is an apparatus for manufacturing "something", and "something" is manufactured and completed in 4 steps by Steps 1 to 4 in chronological order. A sensor 1, a sensor 2, a sensor 3, and a sensor 4 are provided in this manufacturing apparatus, and the data values detected at each step are acquired. The data acquired by the sensors 1 to 4 is usually a physical quantity that can be acquired from the manufacturing apparatus at the time of production of "something" such as temperature, humidity, voltage, current, and pressure. Of course, the sensor may be the five senses of a person, and may be sensory information such as hot, lukewarm, or cold.

It is assumed that the data which is the first analysis target data as shown in FIG. 5A is obtained by the measurement by the sensors 1 to 4 as described above. FIG. 5A is a diagram for explaining processing by the second machine learning data creating means by the state change detection auxiliary device 1A and the state change detection device 1B according to the embodiment of the present invention. In FIG. 5A, a plurality of acquired first analysis target data “Step1 to Step4” which are title information indicating the passage of time are arranged in the vertical direction, and “Sensor 1” which is the acquisition source information indicating the acquisition source. ~ 4 ”are arranged in the horizontal direction to form a first matrix in which the acquired value information is arranged at the intersection of each title information and each acquisition source information.

The prediction model creating means 31 creates the second prediction model 30B using the second teacher data created from the teacher data T by the second machine learning data creating means 32. Further, in the present embodiment, the prediction model creating means 31 is provided for creating the second prediction model 30B, but the second prediction model 30B created by another device or program may be used. In this case, the prediction model creating means 31 is unnecessary.

The second prediction model 30B predicts the objective variable from the explanatory variables by machine learning. Here, as a machine learning algorithm, a random forest of pattern mining can be mentioned, but in addition to this, prediction is made by branching with a classifier (for example, in a tree structure) such as a classification tree or a regression tree. It is possible to adopt the algorithm by machine learning to be performed. Further, the second prediction model 30B may perform machine learning by multiple regression analysis.

The second machine learning data creating means 32 transposes the first matrix to create a transposed matrix and uses it as the second analysis target data. FIG. 5B is a diagram for explaining the processing by the second machine learning data creating means by the state change detection auxiliary device 1A and the state change detection device 1B according to the embodiment of the present invention, and is a diagram for explaining the transposed matrix. Is shown. In the first matrix, "Step1 to Step4" are the objective variables, whereas in the transposed matrix, "sensors 1 to 4" are the objective variables.

6 (A) to 6 (C) are for explaining another processing method by the second machine learning data creating means by the state change detection auxiliary device 1A and the state change detection device 1B according to the embodiment of the present invention. It is a figure of. It is assumed that the second machine learning data creating means 32 obtains the first analysis target data as shown in FIG. 6A in addition to creating the transposed matrix as described above. In this case, the second machine learning data creating means 32 considers the data related to each vertical column of "sensors 1 to 4" as one block, for example, the leftmost vertical column of data "sensor 1". The data of "" is used as the objective variable, and the data of each one column in the vertical direction of "sensors 2 to 4" is used as the explanatory variable to create the second analysis target data. That is, one column is taken out from the explanatory variables existing in three columns, and the position is exchanged with the data of one column which is the objective variable. As a result, for example, the second analysis target data as shown in FIG. 6B is obtained.

Further, the position of the explanatory variable may be changed to obtain the second analysis target data as shown in FIG. 6 (B). That is, while FIG. 6 (A) shows a vertical row of the sensor 1 and a vertical row of the sensor 2 alternated, FIG. 6 (C) shows a row of the sensor 3 in the state of FIG. 6 (B). And the position of one row of the sensor 4 are changed. The second analysis target data obtained in this way is suitable for performing machine learning by multiple regression analysis.

The first predicted value acquisition means 33 applies the first analysis target data including the first objective variable data and the first explanatory variable data acquired from a plurality of acquisition sources to the first prediction model 30A, and the first predicted value. It is for getting data. The second prediction value acquisition means 34 applies the second analysis target data obtained by applying the first analysis target data to the second machine learning data creation means 32 to the second prediction model 30B, and second. Predicted value data is obtained. The prediction error value calculating means 35 has a first prediction error value which is a difference between the first objective variable data and the first predicted value data in the first analysis target data, and the above second analysis target data. The second prediction error value, which is the difference between the second objective variable data and the second predicted value data, is obtained.

The output control means 36 outputs the result calculated by the prediction error value calculation means 35 to the output means. Here, the output means can be the display device 25 already described. In addition to the display device 25, a printer or the like may be used. As described above, not only the first prediction error value obtained based on the first analysis target data actually acquired from the acquisition source, but also the second analysis target data created by the second machine learning data creation means 32 is obtained. Since the second prediction error value is obtained, it is expected that the probability of state change detection will increase because the state change detection assistance will be performed by comparable data when the number of acquisition sources is large. Moreover, the second analysis target data includes the second objective variable data and the second explanatory variable data in which the predicted value corresponding to the second objective variable data can be derived from the second explanatory variable data by the prediction model. It can be said that the above expectations are sufficiently high because they are unreasonable.

The error value extending means 37 obtains the first importance of the first prediction model 30A, performs a process of extending the first importance to the first prediction error value at the first importance, and extends the first prediction error value of the second prediction model 30B. The second importance is obtained, and the process of extending the second prediction error value at this second importance is performed. As the method of determining the importance, a known method such as a method used for obtaining an appropriate decision tree when creating a prediction model of a random forest can be used. Further, as the "processing for stretching", a method of multiplying (multiplying) the value of importance can be mentioned, but division can also be performed depending on the value of importance.

FIG. 7 is a diagram showing the results of error expansion by the state change detection auxiliary device 1A and the state change detection device 1B according to the embodiment of the present invention. That is, FIG. 7 (A) shows the result of performing the process of extending the first error value obtained from the first analysis target data shown in FIG. 5 (A), and FIG. 7 (B) shows the result. , The result of performing the process of extending the second error value obtained from the second analysis target data shown in FIG. 5B is shown. In the portion indicated by the ellipse DA in FIG. 7 (A), an error value in which the difference is greatly emphasized is seen as compared with the other portions, which can be considered as the result of elongation. Further, in the portion indicated by the ellipse DB in FIG. 7B, an error value in which the difference is greatly emphasized is seen as compared with the other portions, which can be considered as the result of elongation. Moreover, this is the result of performing a process of extending the second error value obtained from the second analysis target data shown in FIG. 5 (B), and there is a difference as compared with FIG. 7 (A). Largely, "Since the second prediction error value is obtained, it is expected that the probability of state change detection will increase because the state change detection assistance will be performed by comparable data when the number of acquisition sources is large." The effects mentioned above have been proven.

In the first analysis target data, the title information indicating the passage of time is information that is repeated at predetermined time intervals, and the output means has the title information as the first axis and the acquisition source information as the second axis. Three-dimensional output can be performed with time as the third axis. Moreover, the above three axes can be moved to output (display) as seen from various directions. FIG. 8 is a diagram showing the results of three-dimensional output by the state change detection auxiliary device 1A and the state change detection device 1B according to the embodiment of the present invention. In FIG. 8A, the horizontal direction is the time axis, the vertical direction is the axis of the title information (first axis) designated as “step” in the previous example, and the axis extending from the front to the back of the figure is assigned the acquisition source information. It is the second axis. In this display, the part having a larger difference than the other parts (P1 shown by enclosing it in an ellipse) is not identified, but by moving the three axes as shown in FIG. 8 (B), the difference is larger than the other parts. The large portion (P1 shown by enclosing it in an ellipse) becomes clear, which can contribute to the assistance when the operator visually detects the state change.

The dummy numerical value changing means 38 changes the numerical value of the objective variable and / or the numerical value of the explanatory variable to a dummy numerical value, and brings the numerical value of the objective variable and the numerical value of the explanatory variable close to each other. The problem that the error is greatly different when there is a difference in the size of the dummy variable has been described with reference to FIG. The dummy numerical value changing means 38 compares the numerical values of the objective variable and / or the numerical values of the above explanatory variables, and when there is a difference of a predetermined value or more from the others, the numerical values are brought close to each other. For example, when the numerical value of the "step" already described has a difference of, for example, "1" or more from 1, 13, 21, 22, 31, etc., the value is changed to 10001, 10002, 1003, 10004, 10005. And bring them closer. As a result, even if the numerical value of the dummy variable itself is large, when the error value is obtained and the comparison is performed, the error does not jump out and a large value is not taken, which is suitable for the state fluctuation detection.

In FIG. 4, the means described above the broken line is realized by both the state change detection assisting program and the state change detection program, and the means described below the broken line is for state change detection. It is realized only by the program. In the CPU 10, each means and the like shown in FIG. 4 is realized by a state change detection program in the external storage device 23. What is realized only by the state change detection program is the state change determination means 41.

The state change determination means 41 performs statistical processing on the result output from the state change detection auxiliary device 1A output control means, and determines the presence or absence of the state change based on the determination condition information set in advance. As a specific example of "performing statistical processing and determining the presence or absence of a state change based on preset determination condition information", the state change determination means 41 is a unit for a result output from the output control means. An example is a method in which the blur distance is obtained, the average value and the deviation value of the obtained unit blur distance are obtained, and the basic information for determining the presence or absence of a state change is obtained based on the deviation value.

More specifically, the fact that both the deviation value obtained from the first analysis target data and the deviation value obtained from the second analysis target data are equal to or higher than a predetermined value is used as basic information for determining the presence or absence of state fluctuation. Can be done. Then, when the acquisition of the first analysis target data is obtained for each of the above-mentioned "steps", it is assumed that the basic information for one time is obtained, and the data is acquired until the basic information is obtained for a plurality of times. When a plurality of basic informations are obtained in this way, it may be finally determined that there is a state change when the basic information becomes "continuously N times state change".

Since the state change detection auxiliary device 1A and the state change detection device 1B configured by the above means and the like execute the processing operation according to the flowcharts shown in FIGS. 9 and 10, the operation will be described with reference to the flowcharts. .. The flowchart of FIG. 9 shows a process related to the generation of the first prediction model 30A and the second prediction model 30B, and a process of obtaining the first importance and the second importance, and is a process performed by the CPU 10. Using the given teacher data T, the second teacher data is generated by the second machine learning data creating means 32 (S11). The second teacher data generation can be performed by a process of obtaining a transposed matrix. Next, the first prediction model 30A is generated using the teacher data T (S12). In step S12, the numerical value of the objective variable and / or the numerical value of the above explanatory variable by the dummy numerical value changing means 38 can be changed to a dummy numerical value as necessary.

Further, the second prediction model 30B is generated by using the second teacher data generation (S13). Also in this step S13, the numerical value of the objective variable and / or the numerical value of the above explanatory variable by the dummy numerical value changing means 38 can be changed to a dummy numerical value as necessary. Next, the first importance data is obtained using the teacher data and the first prediction model 30A (S14). In addition, the second importance data is obtained using the second teacher data and the second prediction model 30B (S15). In the processing of steps S14 and S15, a known method such as a method used for obtaining an appropriate decision tree when creating a prediction model for a random forest is used as a method for obtaining the importance. Can be done.

The flowchart of FIG. 10 shows the process up to the detection of the state change, and is the process performed by the CPU 10. The first analysis target data including the first objective variable data and the first explanatory variable data is acquired from a plurality of acquisition sources, and the second analysis target data is processed from the first analysis target data by the second machine learning data creation means 32. (S21). The first analysis target data is applied to the first prediction model 30A to obtain the first prediction value (S22), and the second analysis target data is applied to the second prediction model 30B to obtain the second prediction value (S23). In steps S22 and S23, the numerical value of the objective variable and / or the numerical value of the explanatory variable can be changed to a dummy numerical value by the dummy numerical value changing means 38, if necessary.

The first prediction error value is obtained from the difference between the first objective variable data of the first analysis target data and the first predicted value, and the difference between the second objective variable data of the second analysis target data and the second predicted value is used. The second prediction error value is obtained (S24). The first prediction error value is multiplied by the first importance to obtain the first elongation error value, and the second prediction error value is multiplied by the second importance to obtain the second elongation error value (S25). The first unit blur distance is calculated from the first elongation error value, and the second unit blur distance is calculated from the second elongation error value (S26). Here, the Euclidean distance, the Mahalanobis distance, and the like can be used as the blur distance.

The first deviation value is obtained from the average and deviation of the first unit blur distance, and the second deviation value is obtained from the average and deviation of the second unit blur distance (S27). When the standard deviation of the distribution related to the first unit blur distance is σA, the first deviation value is, for example, 3σA or more, and when the standard deviation of the distribution related to the second unit blur distance is σB, the second deviation value is, for example, 3σB. The above measurement target data is set to bx = 1, and the other measurement target data is set to bx = 0 (S28). The bx = 1 and bx = 0 can be used as the above-mentioned basic information. The basic information about the measurement target data acquired in time series (this measurement target data is the data specified by any one "step" and any one "sensor" in FIG. 5) is bx = 1. It is determined that there is a state change when the above is consecutive a predetermined number of times (S29).

The processing and determination after step S27 are only an example, and 3σB or more can be σB or more, 2σB or more, 4σB or more, and so on. Further, even if it does not continue a predetermined number of times, it can be a condition that the frequency of occurrence increases to a predetermined ratio or more.

As described above, in this embodiment, not only the first prediction error value obtained based on the first analysis target data actually acquired from the acquisition source, but also the second analysis target created by the second machine learning data creation means 32. Since the second prediction error value obtained based on the data is obtained, it is expected that the probability of state change detection will increase because the state change detection assistance will be performed by comparable data when the number of acquisition sources is large. Is.

Although a plurality of embodiments according to the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other embodiments, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are also included in the scope of the invention described in the claims and the equivalent scope thereof.

1A ... state change detection auxiliary device, 1B ... state change detection device, 10 ... CPU, 11 ... main memory, 12 ... bus, 13 ... external storage interface, 14 ... input interface, 15 ... display interface, 16 ... data input interface, 22 ... Mouse, 23 ... External storage device, 24 ... Input device, 25 ... Display device, 26 ... Sensor, 30A, 30B ... Prediction model, 31 ... Prediction model creation means, 32 ... Second machine learning data creation means, 33 ... 1st predicted value acquisition means, 34 ... 2nd predicted value acquisition means, 35 ... prediction error value calculation means, 36 ... output control means, 37 ... error value extension means, 38 ... dummy numerical value changing means, 41 ... state change determination means.

Claims (14)

Regarding the first machine learning data including the first objective variable data and the first explanatory variable data, data different from the first objective variable data in the first machine learning data is set as the second objective variable data, and the first 1. A second machine learning data creation means for creating a second machine learning data in which data different from the first explanatory variable data in the machine learning data is used as the second explanatory variable data.
The first prediction model generated using the teacher data, which predicts the objective variable from the explanatory variables,
A second prediction model that predicts the objective variable from the explanatory variables and is generated from the teacher data using the second teacher data created by the second machine learning data creation means.
As a first predicted value acquisition means for obtaining the first predicted value data by applying the first analysis target data including the first objective variable data and the first explanatory variable data acquired from a plurality of acquisition sources to the first predicted model. ,
A second predicted value acquisition means for obtaining second predicted value data by applying the second analysis target data obtained by applying the first analysis target data to the second machine learning data creating means to the second prediction model. When,
The first prediction error value, which is the difference between the first objective variable data and the first predicted value data in the first analysis target data, the second objective variable data in the second analysis target data, and the first. 2 Prediction error value calculation means for obtaining the second prediction error value, which is the difference from the prediction value data, and
A state change detection assisting device including an output control means for outputting the result calculated by the prediction error value calculation means to the output means. The second machine learning data creating means arranges the acquired first analysis target data in a plurality of title information indicating the passage of time in the vertical direction, and arranges a plurality of acquisition source information indicating the acquisition source in the horizontal direction. As the first matrix in which the acquired value information is arranged at the intersection of the title information and each acquisition source information, the first matrix is transposed to create a transposed matrix and used as the second analysis target data. The state change detection assisting device according to claim 1. The first importance of the first prediction model is obtained, and the first importance is used to extend the first prediction error value, and the second importance of the second prediction model is obtained. An error value extending means for performing a process of extending the second prediction error value with a second importance is provided.
The state change detection assisting device according to claim 1 or 2, wherein the output control means outputs the result of the error value extension by the error value extension means. Claims 1 to 1, wherein the numerical value of the objective variable and / or the numerical value of the explanatory variable is changed to a dummy numerical value, and a dummy numerical value changing means for bringing the numerical value of the objective variable and the numerical value of the explanatory variable close to each other is provided. 3. The state change detection assisting device according to any one of 3. In the first analysis target data, the title information indicating the passage of time is information that is repeated at predetermined time intervals.
The claim is characterized in that the output means performs three-dimensional output with the title information as the first axis, the acquisition source information indicating the acquisition source as the second axis, and time as the third axis. Item 2. The state change detection assisting device according to any one of Items 1 to 4. The state change detection assisting device according to any one of claims 1 to 4.
A state change determining means for determining the presence or absence of a state change based on preset determination condition information by performing statistical processing on the result output from the output control means of the state change detecting auxiliary device.
A state change detection device characterized by the present invention. The state change determining means obtains a unit blur distance from the result output from the output control means, obtains an average value and a deviation value of the obtained unit blur distance, and determines the presence or absence of a state change based on the deviation value. The state change detection device according to claim 6. Computer,
Regarding the first machine learning data including the first objective variable data and the first explanatory variable data, data different from the first objective variable data in the first machine learning data is set as the second objective variable data, and the first A second machine learning data creation means for creating a second machine learning data in which data different from the first explanatory variable data in the machine learning data is used as the second explanatory variable data.
First prediction model generated using teacher data, which predicts the objective variable from the explanatory variables,
A second prediction model that predicts the objective variable from the explanatory variables and is generated from the teacher data using the second teacher data created by the second machine learning data creation means.
A first predicted value acquisition means for obtaining first predicted value data by applying first analysis target data including first objective variable data and first explanatory variable data acquired from a plurality of acquisition sources to the first predicted model.
A second predicted value acquisition means for obtaining second predicted value data by applying the second analysis target data obtained by applying the first analysis target data to the second machine learning data creating means to the second prediction model. ,
The first prediction error value, which is the difference between the first objective variable data and the first predicted value data in the first analysis target data, the second objective variable data in the second analysis target data, and the first. 2 Prediction error value calculation means for obtaining the second prediction error value, which is the difference from the prediction value data.
A state fluctuation detection assisting program characterized in that it functions as an output control means for outputting the result calculated by the prediction error value calculation means to the output means. Using the computer as the second machine learning data creation means, a plurality of the acquired first analysis target data are arranged in the vertical direction with title information indicating the passage of time, and a plurality of acquisition source information indicating the acquisition source is arranged in the horizontal direction. As the first matrix in which the acquired value information is arranged at the intersection of each title information and each acquisition source information, the first matrix is transposed to create a transposed matrix and used as the second analysis target data. The state change detection assisting program according to claim 8, wherein the program functions in such a manner. Further, the computer
The first importance of the first prediction model is obtained, and the first importance is used to extend the first prediction error value, and the second importance of the second prediction model is obtained. It is made to function as an error value extension means for performing a process of extending the second prediction error value at the second importance.
The state change detection assisting program according to claim 8 or 9, wherein the computer functions as the output control means so as to output the result of the error value expansion by the error value expansion means. The computer is further characterized by changing the numerical value of the objective variable and / or the numerical value of the explanatory variable to a dummy numerical value, and functioning as a dummy numerical value changing means for bringing the numerical value of the objective variable and the numerical value of the explanatory variable close to each other. The state change detection assisting program according to any one of claims 8 to 10. In the first analysis target data, the title information indicating the passage of time is information that is repeated at predetermined time intervals, the acquisition source information indicating the acquisition source with the computer as the output means and the title information as the first axis. The state change detection assisting program according to any one of claims 8 to 11, wherein the program is made to function so as to perform three-dimensional output with time as the second axis and time as the third axis. Computer,
Regarding the first machine learning data including the first objective variable data and the first explanatory variable data, data different from the first objective variable data in the first machine learning data is set as the second objective variable data, and the first A second machine learning data creation means for creating a second machine learning data in which data different from the first explanatory variable data in the machine learning data is used as the second explanatory variable data.
First prediction model generated using teacher data, which predicts the objective variable from the explanatory variables,
A second prediction model that predicts the objective variable from the explanatory variables and is generated from the teacher data using the second teacher data created by the second machine learning data creation means.
A first predicted value acquisition means for obtaining first predicted value data by applying first analysis target data including first objective variable data and first explanatory variable data acquired from a plurality of acquisition sources to the first predicted model.
A second predicted value acquisition means for obtaining second predicted value data by applying the second analysis target data obtained by applying the first analysis target data to the second machine learning data creating means to the second prediction model. ,
The first prediction error value, which is the difference between the first objective variable data and the first predicted value data in the first analysis target data, the second objective variable data in the second analysis target data, and the first. 2 Prediction error value calculation means for obtaining the second prediction error value, which is the difference from the prediction value data.
An output control means that outputs the result calculated by the prediction error value calculation means to the output means.
A state change determination means that performs statistical processing on the result output from the output control means and determines the presence or absence of a state change based on preset determination condition information.
A program for detecting state fluctuations, which is characterized by functioning as. Using the computer as the state change determining means, the unit blur distance is obtained for the result output from the output control means, the average value and the deviation value of the obtained unit shake distance are obtained, and the presence or absence of the state change is determined based on the deviation value. The state change detection program according to claim 13, wherein the program is made to function as such.

Images