Analyzing device, analyzing method and analyzing program

Abstract

To analyze, with a low load and high accuracy, multivariable time series data.SOLUTION: An analyzing device includes: a data acquisition unit configured to acquire time-series data of a plurality of types of parameters detected from an analyzed object; a clustering unit configured to calculate amounts of feature of the plurality of types of the time-series data and to group the time-series data by optimization calculation, based on the amounts of feature of the time-series data; and an analyzing unit configured to calculate a time lag of each group, based on the grouped time-series data.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to an analysis device, an analysis method, and an analysis program. [Background technology]

[0002] There is a known technology for detecting multiple types of parameters and analyzing a detection target. For example, Patent Document 1 describes a technology for extracting abnormal sensors having abnormal values ​​from multiple sensors and clustering the abnormal sensors into multiple groups. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2019-145160 A Summary of the Invention [Problem to be solved by the invention]

[0004] However, when the parameters to be detected are time series data from sensors installed in a facility such as a plant, there is a risk of a time lag in the detection timing between the parameters. If the analysis is performed without taking this time lag into account, the accuracy of the analysis may decrease. On the other hand, if the analysis is performed while taking the time lag into account, the calculation load increases and a large amount of time is required for the analysis. Therefore, there is a demand for a low-load, high-accuracy analysis of multivariate time series data.

[0005] An object of the present disclosure is to provide an analysis device, an analysis method, and an analysis program capable of analyzing multivariate time-series data with high accuracy and low load. [Means for solving the problem]

[0006] The analysis device according to the present disclosure includes a data acquisition unit that acquires time series data of multiple types of parameters detected from an analysis target, a clustering processing unit that calculates features for the multiple types of time series data and groups the time series data through optimization calculation based on the features of the time series data, and an analysis unit that calculates a time shift for each group based on the grouped time series data.

[0007] The analysis method according to the present disclosure includes the steps of acquiring time series data of multiple types of parameters detected from an analysis target, calculating feature values ​​for the multiple types of time series data, and grouping the time series data by optimization calculation based on the feature values ​​of the time series data, and calculating a time shift for each group based on the grouped time series data.

[0008] The program according to the present disclosure causes a computer to execute the steps of acquiring time series data of multiple types of parameters detected from an analysis target, calculating feature values ​​for the multiple types of time series data, and grouping the time series data through optimization calculation based on the feature values ​​of the time series data, and calculating a time shift for each group based on the grouped time series data. Effect of the Invention

[0009] According to the present disclosure, multivariate time series data can be analyzed with high accuracy and low load. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a block diagram showing an analysis device according to the present embodiment. [Diagram 2] FIG. 2 is a flowchart illustrating the preprocessing. [Diagram 3] FIG. 3 is a graph illustrating an example of calculation of a feature amount. [Figure 4]FIG. 4 is a graph illustrating an example of calculation of a feature amount. [Diagram 5] FIG. 5 is a flowchart illustrating the process flow for calculating the time lag. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Preferred embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings. Note that the present disclosure is not limited to these embodiments, and when there are multiple embodiments, the present disclosure also includes combinations of the embodiments.

[0012] (Analysis equipment) FIG. 1 is a block diagram showing an analysis device according to the present embodiment. The analysis device 10 according to the present embodiment analyzes time series data of multiple types of parameters related to an analysis target. The analysis target may be any object, and may be, for example, a device installed in a plant, such as a boiler or a gas turbine. The time series data here is data detected by multiple sensors installed in the analysis target. In the present embodiment, in an analysis system including the analysis device 10 and a sensor, it can be said that the time series data detected by the sensor is analyzed by the analysis device 10. For example, if the sensor is a pressure gauge, the type of parameter is pressure, and the parameter value of the time series data detected by the sensor is the pressure value detected for each time series. Also, if the sensor is a thermometer, the type of parameter is temperature, and the parameter value of the time series data detected by the sensor is the temperature detected for each time series.

[0013] The analysis device 10 according to the present embodiment is, for example, a computer, and has an input unit 12, an output unit 14, a communication unit 16, a storage unit 18, and a control unit 20, as shown in FIG. 1. The input unit 12 is a device that accepts user operations, and may be, for example, a mouse, a keyboard, a touch panel, or the like. The output unit 14 is a device that outputs information, and may be, for example, a display that displays an image, or the like. The communication unit 16 is a module that communicates with an external device, and may include, for example, an antenna, or the like. In this embodiment, the communication method by the communication unit 16 is wireless communication, but the communication method may be any communication method. The analysis device 10 may not have the input unit 12, the output unit 14, and the communication unit 16. The analysis device 10 may be configured as a single device, may be configured integrally with other devices, or may be configured as a system combining various devices such as a calculation device and a data server, and is not particularly limited.

[0014] The storage unit 18 is a memory that stores various information such as the contents of calculations and programs of the control unit 20, and includes at least one of a main storage device such as a RAM (Random Access Memory), a ROM (Read Only Memory), and an external storage device such as a HDD (Hard Disk Drive). The program for the control unit 20 saved in the storage unit 18 may be stored in a recording medium readable by the analysis device 10.

[0015] The control unit 20 is a calculation device and includes a calculation circuit such as a CPU (Central Processing Unit). The control unit 20 includes a data acquisition unit 30, a preprocessing unit 32, a clustering processing unit 34, an analysis unit 36, and an output control unit 38. The control unit 20 realizes the data acquisition unit 30, the preprocessing unit 32, the clustering processing unit 34, the analysis unit 36, and the output control unit 38 by reading and executing a program (software) from the storage unit 18, and executes the processes. The control unit 20 may execute these processes using one CPU, or may include multiple CPUs and execute the processes using the multiple CPUs. At least a part of the data acquisition unit 30, the preprocessing unit 32, the clustering processing unit 34, the analysis unit 36, and the output control unit 38 may be realized by hardware.

[0016] The processing contents of the analysis device 10 will be described below.

[0017] (Acquisition of time series data) The data acquisition unit 30 acquires multiple types of time series data detected from the analysis target. The data acquisition unit 30 acquires, as the time series data, the time at which the time series data was detected and a parameter value (e.g., pressure value) of the time series data. The data acquisition unit 30 acquires time series data of each parameter detected in the time period to be analyzed. That is, the data acquisition unit 30 sets a global time period as a reference variable and acquires time series data detected in a predetermined time period in the global time period. The data acquisition unit 30 may acquire the time series data by any method, for example, may acquire the time series data from a sensor via the communication unit 16, or may acquire the time series data by inputting the time series data to the input unit 12 by a user.

[0018] (Preprocessing section) The preprocessing unit 32 performs preprocessing on the time series data of each parameter acquired by the data acquiring unit 30. The preprocessing here refers to a process of correcting the time series data in order to calculate the time shift. Specifically, the preprocessing unit 32 standardizes the time series data of each parameter. The standardization here refers to a process of making the dimensions of parameters measured in different units dimensionless.

[0019] The processing contents by the pre-processing unit 32 may be any processing contents including standardization processing, but a specific example of the processing contents will be described below.

[0020] 2 is a flowchart illustrating the pre-processing. The pre-processing unit 32 performs a periodic variation detection process on the acquired time series data (step S12). The pre-processing unit 32 may perform noise removal before the periodic variation detection process. As the periodic variation detection process, the pre-processing unit 32 detects the periodic variation by, for example, performing a chi-square test on the periodogram of the time series data.

[0021] After performing the periodic variation detection process, the pre-processing unit 32 determines whether or not there is a periodic variation (step S14). If the pre-processing unit 32 determines that there is a periodic variation (Yes in step S14), it removes the periodic variation component (step S16) and proceeds to step S18. The pre-processing unit 32 calculates a moving average at a predetermined interval from the time series data, and removes the moving average from the time series data to remove the periodic variation component. If the pre-processing unit 32 determines that there is no periodic variation (No in step S14), it proceeds to step S18.

[0022] When the result of the determination in step S14 is No, or after the process of step S16 is executed, the preprocessing unit 32 performs a normal distribution detection process (step S18). For example, the Kolmogorov-Smirnov test is used for detecting the normal distribution.

[0023] After performing the normal distribution detection process, the preprocessing unit 32 determines whether the normal distribution is within an allowable range (step S20). Here, a preset value may be used as the allowable range. The allowable range may be a boundary between the normal distribution and whether correction is required for analysis.

[0024] If the pre-processing unit 32 determines that the normal distribution is within the acceptable range (Yes in step S20), the process proceeds to step S24. If the pre-processing unit 32 determines that the normal distribution is not within the acceptable range (No in step S20), the process corrects the distribution of the time series data (step S22) and proceeds to step S24. The pre-processing unit 32 performs correction so that the time series data approaches a normal distribution. For example, Johnson transformation is used for the correction. By correcting the time series data, the pre-processing unit 32 makes the data have normal distribution within the acceptable range.

[0025] When the result of the judgment in step S20 is Yes, or after the correction process in step S22 is performed, the preprocessing unit 32 performs an outlier detection process (step S24). It is determined whether or not an outlier is included in the time series data. For example, the Smirnoff-Grubbs test is used to judge whether or not an outlier is included.

[0026] As a result of the outlier detection process, the preprocessing unit 32 determines whether or not there is an outlier (step S26). If the analysis device 10 determines that there is no outlier (No in step S26), it ends this process. If the analysis device 10 determines that there is an outlier (Yes in step S26), it removes the outlier (step S28) and ends the preprocessing. The preprocessing unit 32 performs a standardization process on the time series data that has been processed as described above, and creates non-dimensional time series data (time series data after preprocessing). The preprocessing unit 32 performs the same process on each of the time series data for each parameter.

[0027] (Determining whether the data is linear) The preprocessing unit 32 judges whether the time series data is linear data based on the parameter values ​​of the time series data and the time when the time series data is detected. In the present embodiment, the preprocessing unit 32 judges whether the time series data after preprocessing by the preprocessing unit 32 is linear data, but is not limited thereto, and may judge whether the time series data before preprocessing is linear data.

[0028] The preprocessing unit 32 determines that the time series data is linear data when the degree of change in the parameter value of the time series data in one direction (increase or decrease) in the time series is equal to or greater than a predetermined threshold. Any method may be used to determine whether the data is linear, and the preprocessing unit 32 may determine whether the data is linear using, for example, Pearson's product-moment correlation. That is, for example, the preprocessing unit 32 may calculate the Pearson's product-moment correlation coefficient for the time series data for each time, and determine that the data is linear when the Pearson's product-moment correlation coefficient is equal to or greater than a positive predetermined value or equal to or less than a negative predetermined value. The data acquiring unit 30 determines whether the time series data of each parameter is linear data.

[0029] For time series data determined to be linear data, the preprocessing unit 32 calculates the time shift of the time series data based on the result of the Pearson product-moment correlation calculation. The time shift here means the amount of shift from the reference time to the time when the time series data is detected.

[0030] On the other hand, the analysis device 10 determines that the time series data that is not determined to be linear data is nonlinear data, and calculates the time shift for the time series data determined to be nonlinear data by the method described below. However, the data is not limited to being divided into nonlinear data and linear data in this manner, and the time shift may be calculated for all time series data by the method described below.

[0031] (Clustering process) The clustering processor 34 calculates feature quantities such as maximum and minimum values ​​for time series data including multiple types of analysis targets, and groups each piece of time series data by optimization calculation based on the feature quantities of the time series data. In this embodiment, the clustering processor 34 calculates feature quantities for time series data that is nonlinear data, and groups the data by optimization calculation.

[0032] (Calculation of feature values) 3 and 4 are graphs for explaining an example of calculation of the feature amount. When calculating the feature amount, the clustering processing unit 34 sets the length of the unit time slot W and divides the time series data for each time into each unit time slot W (unit window). That is, the clustering processing unit 34 sets the length of the unit time slot W to a predetermined setting value and divides the time series data for each time into each unit time slot W (unit window). The clustering processing unit 34 calculates the unit time slot W of the time series data of each parameter by optimization calculation. That is, the clustering processing unit 34 uses the set unit time slot W to calculate the feature amount P described below, and repeats the process of grouping each time series data based on the feature amount P while changing the unit time slot W and the grouping method, thereby setting the optimized unit time slot W and grouping. However, the unit time slot W is not limited to being the target of the optimization calculation, and the unit time slot W may be set to a predetermined fixed value and the optimization calculation of grouping may be performed.

[0033] FIG. 3 shows an example of time series data for each unit time slot. Line D in FIG. 3 shows an example of time series data for a certain parameter. The horizontal axis in FIG. 3 represents time, and the vertical axis represents the parameter value of the time series parameter. In the example of FIG. 3, the time series data is divided into six unit time slots W1, W2, W3, W4, W5, and W6. However, the length of the unit time slot W and the number of unit time slots W may be set arbitrarily.

[0034] The clustering processing unit 34 calculates the feature amount P for each unit time slot W based on the time series data in the unit time slot W. The clustering processing unit 34 calculates the feature amount P for the unit time slot W based on the time series data in the unit time slot W. That is, the feature amount P can be said to be a reference value of the time series data in the unit time slot W, which is calculated based on each time series data in the unit time slot W. Any method may be used to calculate the feature amount P based on the time series data in the unit time slot W. For example, the clustering processing unit 34 may set the maximum value of the parameter values ​​of the time series data in the unit time slot W as the feature amount P in the unit time slot W. For another example, the clustering processing unit 34 may set the extreme value (maximum value or minimum value) of the parameter values ​​of the time series data in the unit time slot W as the feature amount P in the unit time slot W. For another example, the clustering processing unit 34 may set the average value of the parameter values ​​of the time series data in the unit time slot W as the feature amount P in the unit time slot W.

[0035] FIG. 4 shows an example of a feature calculated for each unit time slot W. In the example of FIG. 3 and FIG. 4, extreme values ​​are calculated as feature values. That is, in the example of FIG. 3 and FIG. 4, the extreme value of the parameter value in unit time slot W1 is the feature value P1 in unit time slot W1, the extreme value of the parameter value in unit time slot W2 is the feature value P2 in unit time slot W2, the extreme value of the parameter value in unit time slot W3 is the feature value P3 in unit time slot W3, the extreme value of the parameter value in unit time slot W4 is the feature value P4 in unit time slot W4, the extreme value of the parameter value in unit time slot W5 is the feature value P5 in unit time slot W5, and the extreme value of the parameter value in unit time slot W6 is the feature value P6 in unit time slot W6. In this example, there are both a minimum value and a maximum value in unit time slot W5, and the maximum value is the feature value P5, but this is not limited to this.

[0036] The clustering processing unit 34 preferably performs clustering processing on the calculated feature quantities. The clustering processing is processing for dividing the time series data for each parameter into a plurality of groups. In other words, the clustering processing is a grouping processing that is performed preliminarily before grouping by the optimization calculation described later. By performing the clustering processing in advance by a method that does not use the optimization calculation, the amount of calculation in the optimization calculation can be reduced. Here, examples of the clustering processing include K-means, Ward hierarchical clustering, and mixed Gaussian.

[0037] (Optimization calculation) The clustering processing unit 34 groups each of the time series data by optimization calculation based on the feature amount of each unit time slot W of the time series data. That is, the clustering processing unit 34 classifies each of the time series data for each parameter into any group (cluster) by optimization calculation. That is, a group refers to a group of time series data classified into the same group by the clustering processing unit 34. Note that, when the above-mentioned clustering processing is performed, the time series data of the clustered group may be further grouped.

[0038] More specifically, the clustering processing unit 34 performs grouping so that the evaluation index is optimized by optimization calculation. The evaluation index is an index indicating that each group is appropriately separated from the others, and the more similar the waveforms of the time series data included in a group are and the more dissimilar the waveforms of different groups are, the higher the evaluation (the more optimized it is). The evaluation index can also be said to be an objective function with the feature amount for each unit time period W as a constant and the method of grouping as a variable. The method of grouping means which group each time series data is classified into. Any method of evaluation index may be used, and examples of such evaluation index include silhouette coefficient, CH (Calinski Harabaz) score, DB (Davis Bouldin) score, KL (Krzanowski Lai) score, and Gap Statistics. Furthermore, it is preferable to use a silhouette coefficient indicating both the inter-cluster distance and the intra-cluster distance as the evaluation index. The higher the value of the silhouette coefficient, the greater the distance between clusters is compared to the distance within the cluster. In this case, the clustering processing unit 34 calculates grouping so that the silhouette coefficient approaches 1 as closely as possible (so that the evaluation index is optimized).

[0039] In this embodiment, the clustering processing unit 34 calculates how to divide the data into groups when optimizing the evaluation index, with a constraint that the number of groups is a predetermined setting value. That is, the clustering processing unit 34 sets the total number of groups to be classified in advance as a constraint, and performs grouping so that the evaluation index is optimized while satisfying the constraint (total number of groups).

[0040] In the optimization calculation, the clustering processing unit 34 judges whether the evaluation index, which is the objective function, has converged. The criteria for judging whether the evaluation function has converged may be arbitrary, but for example, if a series of calculations, including setting the unit time slot W, calculating the feature amount, and grouping, is executed a predetermined number of times and the evaluation index is not updated (for example, if the difference between the values ​​of the evaluation index in each calculation continues to be within a predetermined value for a predetermined number of times), it is judged that the evaluation function has converged. The clustering processing unit 34 treats the grouping method when the evaluation index has converged as the result of the optimization calculation. On the other hand, if the evaluation index has not converged, the optimization calculation conditions are changed and the minimization calculation is repeated until the evaluation index converges. The optimization calculation conditions to be changed include the length of the unit time slot W and the number of groups. That is, the clustering processing unit 34 changes the length of the unit time slot W and the number of groups, and repeats the optimization calculation until the evaluation index converges. By using the length of the unit time slot W and the number of groups as coefficients for adjustment in the optimization calculation, the grouping can be performed appropriately.

[0041] (Calculation of time delay) The analysis unit 36 ​​calculates the time shift of each group based on the group (grouped time series data). The method of calculating the time shift based on the group may be arbitrary, but for example, the analysis unit 36 ​​may identify representative time series data from the group and calculate the time shift of each group based on the representative time series data of each group. That is, the analysis unit 36 ​​extracts time series data to be the representative time series data from the time series data included in the group. The method of extracting the representative time series data may be arbitrary, and for example, time series data close to the average value of the time series data included in the group may be selected, or may be selected randomly. The analysis unit 36 ​​performs the same process for each group to identify the representative time series data for each group.

[0042] The analysis unit 36 ​​calculates the time shift of each group based on the representative time series data of each group. For example, the analysis unit 36 ​​may calculate the difference in the time at which the representative time series data was detected as the time shift of the group.

[0043] In this way, by understanding the time lag status for each time series data, it is possible to understand the correlation between fluctuations at each location caused by the same cause. This makes it possible to infer the cause from the fluctuation status. Furthermore, by being able to understand the time lag of the time series data, it is also possible to understand locations where fluctuations that should occur are not occurring. This makes it possible to accurately understand the status of the analysis target, taking into account the time lag.

[0044] (Result output) The output control unit 38 outputs the calculation result of the time shift by the analysis unit 36. Any method of output by the output control unit 38 may be used. For example, the output control unit 38 may output the calculation result of the time shift to the output unit 14 (for example, may display it on a display). Also, for example, the output control unit 38 may transmit (output) the calculation result of the time shift to another device via the communication unit 16.

[0045] (Processing flow) The above-described process flow of calculating the time lag will be described. FIG. 5 is a flowchart illustrating the process flow of calculating the time lag. As shown in FIG. 5, the analysis device 10 acquires time series data of a plurality of types of parameters by the data acquisition unit 30 (step S30), and the preprocessing unit 32 performs preprocessing on each of the time series data (step S32). The analysis device 10 determines whether the time series data is linear data based on the preprocessed time series data (step S34), and if it is linear data (step S34; Yes), estimates the time lag of the time series data (step S36). On the other hand, if the time series data is not linear data (step S34; No), the analysis device 10 sets a unit time slot W by the clustering processing unit 34, divides the time series data into unit time slots W, and calculates a feature amount for each unit time slot W (step S38). That is, the clustering processing unit 34 calculates a feature amount for each unit time slot W for each of the time series data that is not linear data (nonlinear data). Then, the clustering processing unit 34 sets the number of groups, groups the time series data so that the number of groups is the set number (step S40), and calculates an evaluation index when the grouping is applied based on the feature for each unit time slot W (step S42). If the calculated evaluation index has not converged (S44; No), the process returns to step S38, resets the unit time slot W, and continues the optimization calculation for the grouping. On the other hand, if the evaluation index has converged (step S44; Yes), the set grouping is treated as the calculation result of the optimization calculation, and the analysis unit 36 ​​estimates the time lag of each group (step S46).

[0046] In this way, the analysis device 10 according to the present embodiment groups the time series data by optimization calculation based on the feature quantities for the multiple types of time series data, and calculates the time lag for each group. According to the present embodiment, the time series data of multiple variables is grouped based on the feature quantities, and representative data of the groups is compared, thereby suppressing a decrease in the calculation accuracy of the time lag, while reducing the number of data to be compared, thereby reducing the possibility that the calculation will diverge and not converge, and the amount of calculation can also be reduced. Therefore, according to the present embodiment, the time series data of multiple variables can be analyzed with high accuracy and low load. Furthermore, since the time series data of multiple variables each have different characteristics, there is a risk that the conditions for grouping cannot be set appropriately, and grouping cannot be performed appropriately. In contrast, according to the present embodiment, grouping can be performed appropriately by performing grouping by optimization calculation. In particular, grouping can be performed more appropriately by performing optimization calculation while adjusting the length of the unit time zone W and the number of groups.

[0047] For example, in a boiler or gas turbine plant, when an abnormality occurs at a certain location due to equipment failure, the effect propagates to downstream of the piping. Here, multivariate time series values ​​are measured as operation data and are used for analysis to improve performance and to analyze the causes of trouble. In utilizing multivariate time series values, time lag correction, that is, time alignment of the time series is performed for analysis. Time alignment of time series values ​​can be performed manually by drawing a trend graph of the sensor value and comparing it with the design knowledge of the plant. However, as the number of sensors increases with the digitalization of the plant, the amount of work increases exponentially with the number of sensors, making manual time alignment difficult. In addition, when automatically performing time alignment of a series, a method that can calculate time alignment of time series values ​​including nonlinear relationships in a realistic time cannot be established because it requires a large amount of calculation resources. As a result, analysis for improving performance and analysis of the causes of trouble may be insufficient, and in software sensors and control systems, there is a possibility that the accuracy of the system may decrease or trouble may occur during system operation. In response to this, the analysis device 10 of this embodiment performs the above processing to automatically adjust the time of time-series values ​​such as sensor indication values ​​and actuator command values ​​installed at various locations in the plant in accordance with the installation locations of sensors and actuators such as valves, upstream, midstream, downstream, etc., and synchronize the timing, thereby making it possible to increase the accuracy of detection of the occurrence of abnormal events, estimation of their causes, and analysis.

[0048] (effect) As described above, the analysis device 10 according to the first aspect of the present disclosure includes a data acquisition unit 30 that acquires time series data of multiple types of parameters detected from an analysis target, a clustering processing unit 34 that calculates feature amounts for the multiple types of time series data and groups each of the time series data by optimization calculation based on the feature amounts of the time series data, and an analysis unit 36 ​​that identifies representative time series data from the grouped time series data and calculates a time shift for each group based on the representative time series data of each group. According to the present disclosure, by grouping multivariate time series data based on the feature amounts and comparing the representative data of the groups, it is possible to perform analysis with high accuracy and low load.

[0049] The analysis device 10 according to the second aspect of the present disclosure is the analysis device 10 according to the first aspect, in which the clustering processing unit 34 performs grouping so as to optimize an evaluation index showing that the groups are appropriately separated from each other. According to the present disclosure, by performing grouping so as to optimize the evaluation index, it is possible to perform grouping more appropriately.

[0050] The analysis device 10 according to a third aspect of the present disclosure is the analysis device 10 according to the first or second aspect, and the clustering processing unit 34 executes optimization calculations under the constraint that the number of groups is a predetermined set value. According to the present disclosure, by setting the number of groups and executing optimization calculations, it is possible to perform grouping more appropriately.

[0051] An analysis device 10 according to a fourth aspect of the present disclosure is the analysis device 10 according to any of the first to third aspects, in which the clustering processing unit 34 divides the time series data for each time into unit time slots W, and calculates a feature amount for each unit time slot W based on the time series data in the unit time slot W. According to the present disclosure, setting the feature amount in this manner allows for more appropriate grouping.

[0052] An analysis device 10 according to a fifth aspect of the present disclosure is the analysis device 10 according to the fourth aspect, in which the clustering processing unit 34 calculates feature quantities using a predetermined set value for the unit time slot W. According to the present disclosure, by setting the unit time slot W and performing optimization calculations, grouping can be performed more appropriately.

[0053] The analysis device 10 according to a sixth aspect of the present disclosure is the analysis device 10 according to any one of the first to fifth aspects, further including a preprocessing unit 32 that standardizes the time series data of the types, and a clustering processing unit 34 that calculates features from the standardized time series data. By performing the preprocessing, the grouping can be performed more appropriately.

[0054] The analysis method according to the seventh aspect of the present disclosure includes the steps of: acquiring time series data of multiple types of parameters detected from an analysis target; calculating feature values ​​for the multiple types of time series data, and classifying the time series data into groups by optimization calculation based on the feature values ​​of the time series data; identifying representative time series data from the grouped time series data, and calculating a time shift for each group based on the representative time series data for each group. According to the present disclosure, analysis can be performed with low load and high accuracy.

[0055] A program according to an eighth aspect of the present disclosure causes a computer to execute the steps of: acquiring time series data of multiple types of parameters detected from an analysis target; calculating feature values ​​for the multiple types of time series data, and grouping each of the time series data by optimization calculation based on the feature values ​​of the time series data; and identifying representative time series data from the grouped time series data, and calculating a time shift for each group based on the representative time series data for each group. According to the present disclosure, analysis can be performed with high accuracy and low load.

[0056] Although the embodiment of the present disclosure has been described above, the embodiment is not limited by the contents of this embodiment. In addition, the above-mentioned components include those that a person skilled in the art can easily imagine, those that are substantially the same, and those that are within the so-called equivalent range. Furthermore, the above-mentioned components can be appropriately combined. Furthermore, various omissions, substitutions, or modifications of the components can be made without departing from the gist of the above-mentioned embodiment. [Explanation of symbols]

[0057] 10 Analysis device 30 Data Acquisition Section 32 Pre-processing section 34 Clustering Processing Section 36 Analysis Department 38 Output control section

Claims

1. a data acquisition unit for acquiring time series data of a plurality of types of parameters detected from an analysis target; a clustering processing unit that calculates feature amounts for the plurality of types of time series data and groups the time series data by optimization calculation based on the feature amounts of the time series data; an analysis unit that calculates a time lag for each group based on the grouped time series data; Including, Analysis device.

2. The analysis device according to claim 1 , wherein the clustering processing unit performs grouping so that an evaluation index indicating that the groups are appropriately separated from each other is optimized.

3. The analysis device according to claim 1 , wherein the clustering processing unit executes optimization calculations under a constraint that the number of groups is set to a predetermined set value.

4. 3. The analysis device according to claim 1, wherein the clustering processing unit divides the time series data for each time into unit time periods and calculates the feature amount for each unit time period based on the time series data for each unit time period.

5. The analysis device according to claim 4 , wherein the clustering processing unit calculates the feature amount by using the unit time period as a predetermined setting value.

6. Further comprising a pre-processing unit that standardizes the plurality of types of time series data; The analysis device according to claim 1 , wherein the clustering processing unit calculates the feature amount from the standardized time-series data.

7. acquiring time series data of a plurality of types of parameters detected from an analysis target; calculating feature amounts for the plurality of types of time series data, and classifying the time series data into groups by optimization calculation based on the feature amounts of the time series data; identifying representative time series data from the grouped time series data, and calculating a time shift for each group based on the representative time series data for each group; Including, Analysis method.

8. acquiring time series data of a plurality of types of parameters detected from an analysis target; calculating feature amounts for the plurality of types of time series data, and classifying the time series data into groups by optimization calculation based on the feature amounts of the time series data; calculating a time lag for each group based on the grouped time series data; to cause a computer to execute Analysis program.

Images