The present application is a National Stage of International Application No. PCT/JP20171033019 (filed Sep. 13, 2017), and claims priority from Japanese Patent Application No. 2016-179287 (filed on Sep. 14, 2016), the contents of which are hereby incorporated in their entirety by reference into this specification. The present invention relates to a system analysis support device, a system analysis support method, and a program, and in particular to a system analysis support device, a system analysis support method, and a program, that support analysis operations for system fault analysis and predictive detection of malfunctions.
Patent Literature (PTL) 1 discloses an operations management device in which it is possible to predictively detect malfunctions of a management device configuring a system. Specifically, the operations management device includes: a correlation model generation part 123 that, with a performance item or management device as an element, derives a correlation function of at least first performance series information indicating a time series change of performance information related to a first element, and second performance series information indicating a time series change of performance information related to a second element, generates a correlation model based on the correlation function, and obtains this correlation model for a combination among the respective elements; and a correlation change analysis part 124 that analyzes change of correlation model based on performance information newly detected and obtained from the management device.
Patent Literature 2 discloses an abnormality detection system that predictively detects a system abnormality using a data analysis method known as clustering. Patent Literature 3 discloses an abnormality detection method in which data are obtained from a plurality of sensors, and based on degree of similarity among the data, in a case of data in which the degree of similarity among the data is low, by performing addition or removal of data with regard to learning data, using existence or nonexistence of an abnormality in the data, the learning data is generated/updated, and, based on deviation of individual data items included in the learning data, and newly obtained observed data, abnormality in observed data is detected.
The following analysis is given according to the present invention. In the abnormality determination method according to accumulation of abnormality degree as represented in Patent Literature 1, according to prediction error being large or duration thereof or a combination thereof, the abnormality degree is calculated and an abnormal invariant or sensor is identified (for example, refer to FIG. 14 in Patent Literature 1).
In a case of identifying an abnormality cause having only a high or low score with regard to the abovementioned abnormality degree, events that occurred incidentally, events that occurred as a result (of a true cause), or events indicating an abnormality even in a normal period may affect abnormality degree score. Since it is difficult for these to be automatically determined, it has been necessary to individually determine and screen out sensor values related to these events or predictive models using these.
It is an object of the present invention to provide a system analysis support device, a system analysis support method, and a program, that can perform system abnormality detection, abnormality monitoring, and abnormality cause identification with good accuracy, without performing a process of screening out sensor values or predictive models in system invariant analysis as represented in the abovementioned Patent Literature 1.
According to a first aspect, a system analysis support device is provided that includes a data acquisition part that obtains time series data (items) measured in a system that is to be analyzed. The analysis support device further includes an overall abnormality degree calculation part that calculates transition of overall abnormality degree of the system to be analyzed, using a predictive model generated so that, with 2 or more time series data (items) as input, values representing a relationship between the 2 or more time series data items are outputted, and the time series data (items). The analysis support device further includes an overall operation representative value extraction part that selects and presents time series data (items) indicating change similar to transition of overall abnormality degree of the system that is to be analyzed, from among the time series data (items).
According to a second aspect, a system analysis support method is provided for a computer comprising a storage part that stores a predictive model generated so that, with 2 or more time series data (items) as input, values representing a relationship between the 2 or more time series data (items) are outputted, and a calculation part that calculates transition of abnormality degree representing overall abnormality degree of a prescribed system that is to be analyzed, using the predictive model and the time series data (items), wherein the computer obtains time series data (items) measured in a system that is to be analyzed, calculates transition of abnormality degree representing overall abnormality degree of the prescribed system that is to be analyzed, and selects and presents time series data (items) indicating change similar to transition of overall abnormality degree of the system that is to be analyzed, from among the time series data (items). The present method is associated with a particular apparatus, referred to as a computer that has the abovementioned storage part and calculation part (processor).
According a third aspect, a program is provided that causes execution of a computer including a storage part that stores a predictive model generated so that, with 2 or more time series data (items) as input, values representing a relationship between the 2 or more time series data (items) are outputted, and a calculation part that calculates transition of abnormality degree representing overall abnormality degree of a prescribed system that is to be analyzed, using the predictive model and the time series data (items), wherein the computer executes: a process of obtaining time series data (items) measured in a system that is to be analyzed, a process of calculating transition of abnormality degree representing overall abnormality degree of the prescribed system that is to be analyzed, and a process of selecting and presenting time series data (items) indicating change similar to transition of overall abnormality degree of the system that is to be analyzed, from among the time series data (items), to a user. It is to be noted that this program may be recorded on a computer-readable (non-transient) storage medium. That is, the present invention may be embodied as a computer program product.
According to the present invention, facilitation and accuracy improvement are realized for abnormality detection, abnormality monitoring, and abnormality cause identification for a system. That is, the present invention, converts a system analysis support device disclosed in background art to a system analysis support device having dramatically improved performance.
First, a description is given of an outline of an exemplary embodiment of the present invention, making reference to the drawings. It is to be noted that reference symbols in the drawings attached to this outline are added to respective elements for convenience, as examples in order to aid understanding, and are not intended to limit the present invention to modes illustrated in the drawings. Connection lines between blocks in the diagrams referred to in the following description include both unidirectional and bidirectional. Unidirectional arrows schematically show flow of main signals (data), but do not exclude bidirectionality.
The present invention, in an exemplary embodiment thereof as shown in
More specifically, the representative index selection part 14 selects and presents time series data indicating change similar to a transition of overall abnormality degree of the system that is to be analyzed, from among time series data. For example, a calculation result indicating change of overall abnormality degree as in (a) of
For this type of transition of overall abnormality degree, there is increase trend at time t for both sensor data 1 and sensor data N, but change does not necessarily resemble overall abnormality degree transition, such as sensor data N value dropping rapidly at time t0. Meanwhile, sensor data 1 indicates that for all intervals, similarity degree is high, such as increase and decrease trends matching overall abnormality degree. In this case, the representative index selection part 14 selects sensor data 1 as a representative index, and presents this to a user. It is to be noted that the similarly degree, on creating a prediction formula among the sensor data, can perform calculation according to error between predicted value and actually measured value. For example, in a case of calculating similarity degree between sensor data 1 and overall abnormality degree, a prediction formula is created to predict overall abnormality degree from the sensor data 1, and calculation is performed according to error between the predicted value and the actually measured value. Clearly, it is also possible to use various types of other methods as the method of calculating waveform similarity degree.
A user who obtains this type of result can thereafter focus on sensor data 1, to consider a mechanism for performing detection of system abnormality, or perform abnormality monitoring by the sensor data 1, or perform identification of abnormality cause using sensor data 1.
In particular in a large scale system as described at the outset of this document, for countless sensors, effects are due not only to human causes but also to weather and seasonal changes, so that various events occur. By using the present invention, it is possible to sift out sensor data of values not worth being focused on, and to present important sensor data to the user. Clearly, the user can focus not only on presented sensor values, but can create a predictive model using a sensor other than the sensor in question, and can perform evaluation of an existing predictive model using the presented sensor data.
Next, a detailed description is given concerning a first exemplary embodiment of the present invention, making reference to the drawings.
The data receiving part 101 receives time series data from various types of sensor included in the system and accumulates them in the time series data storage part 102. It is to be noted that embodiments in which the data receiving part 101 collects data from various types of sensor, include an embodiment in which the data receiving part 101 directly receives time series data from a sensor or an IoT (Internet of Things) terminal or the like connected to a sensor, and an embodiment in which the data receiving part 101 obtains necessary time series data from a server or the like disposed in a cloud system or the like.
The time series data storage part 102 is configured by a database or the like that holds data collected by the data receiving part 101 as respective time series data.
The predictive model generation part 103 generates a predictive model (correlation model) using time series data of a learning period in question (learning interval), among the time series data accumulated in the time series data storage part 102. Specifically, the predictive model generation part 103 derives a correlation function among 2 or more time series data items of the learning interval, generates a predictive model (correlation model) based on the correlation function, and obtains the predictive model (correlation model) with regard to a combination among respective sensor data items. For example, as shown in
The overall abnormality degree calculation part 104 uses prediction error, which is the difference between actual time series data and values obtained by inputting time series data for an arbitrary period to a predictive model (correlation model), or a continuous period thereof, or a numerical formula of a combination thereof, to calculate time series change of overall abnormality degree of a system that is to be analyzed.
The representative index selection part 105 selects 1 or more items of sensor data indicating movement similar to time series change of overall abnormality degree calculated by the overall abnormality degree calculation part 104, from among the time series data accumulated in the time series data storage part 102. It is to be noted that the time series data “indicating similar movement” may be obtained by regression analysis or multivariate analysis, with modeling by an ARX (Auto-Regressive eXogeneous) model having overall abnormality degree as object variable, modeling by regression analysis such as Lasso, modeling by principal component analysis or the like.
The result output part 106 outputs 1 or more sensors selected by the representative index selection part 105. For example in a case where the similarity degree shown in the lower part of
Next, a detailed description is given concerning operations of the present exemplary embodiment, making reference to the drawings.
Next, the analysis support device 100 generates a predictive model of respective sensors in a learning period (step S002). It is to be noted that in step S002, all sensors need not be targets as time series data used in predictive model generation. For example, a sensor the user considers to be clearly unnecessary may be excluded (refer to the fourth exemplary embodiment).
Next, the analysis support device 100 calculates overall abnormality from respective predictive models and respective time series data (actually measured values) (step S003). Here, in a case where the time series data (actually measured values) after a certain time deviates largely from respective predicted values obtained from the respective predictive models and this state continues for a long time, for the overall abnormality degree also, a change is indicated where the abnormality degree increases from the relevant time.
Next, the analysis support device 100 calculates similarity of overall abnormality degree and respective time series data, and extracts (selects) representative sensor (group) (step S004).
The upper part of
Finally, the analysis support device 100 outputs the extracted (selected) representative sensor (group) (step S006). In this stage, a sensor with low similarity degree with transition of overall abnormality degree, that is, a sensor with low correlation with overall abnormality degree is filtered in step S004 and excluded from output target. By using the outputted sensor data as representative index, it is possible to perform system abnormality detection, abnormality monitoring and abnormality cause identification with good accuracy.
It is to be noted that in the above description, as representative index, 1 or more sensors were described as being outputted, but as a representative index a modified embodiment is also possible where a predictive model (invariant) is outputted.
The upper part of
According to this type of modified exemplary embodiment, it is possible to present to the user a predictive model (invariant) with a stronger correlation than individual sensor values.
In the abovementioned first exemplary embodiment no particular limit was provided with regard to learning interval, but it is also effective to narrow down the learning interval. Continuing, a description is given concerning a second exemplary embodiment in which a representative index is selected that limits the learning interval to a fixed period up to where an abnormality occurs.
The learning period specification receiving part 107 receives the specification of the learning period from a user and indicates the learning interval to the representative index selection part 105. Various embodiments for receiving the specification of the learning period may be considered; for example, it is possible to use a method in which a graph indicating change of overall abnormality degree as shown in
The upper part of
As described above, according to the second exemplary embodiment in which the learning interval is narrowed down, in addition to the effect of the first exemplary embodiment, it is possible to narrow down the predictive model and sensor indicating useful movement clarifying abnormality occurrence mechanism.
In the abovementioned first exemplary embodiment, a description was given in which a predictive model and sensor indicating movement similar to movement of overall abnormality degree are extracted, but here a description is given concerning a third exemplary embodiment in which a sensor is identified in which change appears in advance of an abnormality in the overall abnormality degree.
The representative index selection part 105a of the present exemplary embodiment, with regard to calculating the similarity degree of time series data accumulated in a time series data storage part 102 and time series change of the overall abnormality degree, shifts the time axis of the time series data by a prescribed time k1, k2, . . . , kn, and calculates similarity degree of n patterns. Here the prescribed time k1, k2, . . . , kn is increased in a prescribed step from a prescribed judged lower limit k1 (a negative value is also possible) up to a prescribed judged upper limit.
The upper part of
In such a case, the representative index selection part 105a of the present exemplary embodiment selects a sensor giving priority to whether or not it is in advance with regard to the overall abnormality degree, rather than similarity degree high or low. For example, in
As described above, according to the third exemplary embodiment in which the time axis of the time series data is moved and similarity degree calculated, and a sensor that is in advance of overall abnormality degree is selected, in addition to the effects of the first exemplary embodiment it is possible to identify and show to the user a sensor having signs of an abnormality.
In the present exemplary embodiment, the result output part 106 may not only output a sensor indicating movement having similarity and being in advance, but also similarity degree of respective sensors, and time shift among respective sensors performing calculation (relative time from overall abnormality degree).
It is to be noted that, similar to the modified exemplary embodiment of the first exemplary embodiment, in the present exemplary embodiment it is possible to have a modified exemplary embodiment that calculates the similarity degree of a predictive model (invariant) and outputs the predictive model (invariant).
It is to be noted that display form or sorting order of respective sensors and models in the sensor and model lists shown in
In calculation of overall abnormality degree in the abovementioned first to third exemplary embodiments, clearly unnecessary sensors are preferably excluded. Below a description is given concerning a fourth exemplary embodiment provided with a user interface to select sensor(s) to be excluded in calculating overall abnormality.
The excluded data selection part 108 presents a sensor selection list to the user, and in calculating the overall abnormality degree, selection of sensors and predictive models not to be used is received from the user. It is to be noted that the excluded data selection part 108 may present a list of sensors currently selected (not excluded) and overall abnormality degree using these to the user, and nay receive the excluded data sensors using interactive processing. In this case, a method may be used that does not simply request recalculation of overall abnormality degree with respect to the overall abnormality degree calculation part 104c, but corrects the overall abnormality degree by removing contribution degree amount of predictive models and sensors selected for exclusion from the overall abnormality degree selected the previous time, and may reduce the number of calculations.
Along with the list shown in
An overall abnormality degree calculation part 104c uses time series data for which an exclusion flag is not set, among time series data contained in the time series data storage part 102c, to calculate overall abnormality degree.
According to the present exemplary embodiment, since the overall abnormality degree is calculated more delicately, selection of sensors by a representative index selection part 105 is also more delicate. It is to be noted that in the example of
In accordance with the scale of a system, it can be expected that the number of sensors and predictive models after extraction as used in the abovementioned first to fourth exemplary embodiments will become very large. Below, a description is given concerning a fifth exemplary embodiment provided with a user interface that organizes the sensors and predictive models after extraction.
The exclusion index selection part 109 receives a selection of unnecessary (not displayed) sensors or predictive models from the user. It is to be noted that a result selected by the exclusion index selection part 109 is sent to a result output part 106, and a display by the result output part 106 is updated.
The result output part 106 updates a sensor list to be outputted as representative index, based on a selection result by the exclusion index selection part 109.
It is to be noted that in the example of
It is to be noted that the respective parts (processing means) of the analysis support device control device shown in the abovementioned respective diagrams may be implemented by a computer program that executes the abovementioned respective processing on a computer configuring the analysis support device, using hardware thereof.
A description has been given above of respective exemplary embodiments of the present invention, but the present invention is not limited to the abovementioned exemplary embodiments, and modifications, substitutions and adjustments may be added within a scope that does not depart from fundamental technical concepts of the invention. Network configurations, respective element configurations and forms for representing message shown in the respective drawings are examples for the purpose of aiding understanding of the invention, and are not intended to limit the invention to configurations illustrated in the drawings.
Finally, preferred modes of the present invention are summarized.
[First Mode]
(Refer to the system analysis support device according to the first aspect described above.)
[Second Mode]
The representative index selection part of the abovementioned analysis support device may have a configuration that selects time series data (items) indicating change similar to transition of overall abnormality degree of the system that is to be analyzed, in a period until transition of the overall abnormality degree of the system that is to be analyzed exceeds a prescribed threshold.
[Third Mode]
The representative index selection part of the abovementioned analysis support device may have a configuration that selects time series data (items) wherein transition of overall abnormality degree of the system that is to be analyzed is similar to transition of overall abnormality degree of the system that is to be analyzed, and changes in advance thereof.
[Fourth Mode]
The representative index selection part of the abovementioned analysis support device may have a configuration that selects time series data (items) in which change similar to transition of overall abnormality degree of the system that is to be analyzed appears in advance by a prescribed time k.
[Fifth Mode]
The abovementioned analysis support device may use, as the time series data (items), time series data (items) obtained by inputting time series data (items) obtained from a plurality of sensors, to a prescribed predictive model.
[Sixth Mode]
The analysis support device is further preferably provided with a user interface that displays a list of time series data (items) along with degree of similarity with transition of overall abnormality degree of the system to be analyzed, from among the series data (items), and preferably receives from a user a selection of time series data (items) indicating change similar to transition of overall abnormality degree of the system that is to be analyzed.
[Seventh Mode]
The user interface displaying the list of time series data preferably displays, with regard to individual time series data items (items), information of degree of similarity with transition of overall abnormality degree of the system that is to be analyzed, and of advance time thereof.
[Eighth Mode]
The abovementioned analysis support device may be provided with an excluded data selection part that receives a selection of time series data (items) to be excluded in calculation of transition of the abnormality degree representing overall abnormality degree of the system that is to be analyzed in the overall abnormality degree calculation part.
[Ninth Mode]
(Refer to the system analysis support method according to the second aspect described above.)
[Tenth Mode]
(Refer to the program according to a third aspect described above.)
It is to be noted that the ninth and tenth modes described above may be expanded with regard to the second to eighth modes, similar to the first mode.
It is to be noted that the various disclosures of the abovementioned Patent Literature are incorporated herein by reference thereto. Modifications and adjustments of exemplary embodiments and examples may be made within the bounds of the entire disclosure (including the scope of the claims) of the present invention, and also based on fundamental technological concepts thereof. Various combinations and selections of various disclosed elements (including respective elements of the respective claims, respective elements of the respective exemplary embodiments and examples, respective elements of the respective drawings and the like) are possible within the scope of the disclosure of the present invention. That is, the present invention clearly includes every type of transformation and modification that a person skilled in the art can realize according to the entire disclosure including the scope of the claims and to technological concepts thereof. In particular, with regard to numerical ranges described in the present specification, arbitrary numerical values and small ranges included in the relevant ranges should be interpreted to be specifically described even where there is no particular description thereof.
Number | Date | Country | Kind |
---|---|---|---|
JP2016-179287 | Sep 2016 | JP | national |
Filing Document | Filing Date | Country | Kind |
---|---|---|---|
PCT/JP2017/033019 | 9/13/2017 | WO |
Publishing Document | Publishing Date | Country | Kind |
---|---|---|---|
WO2018/052015 | 3/22/2018 | WO | A |
Number | Name | Date | Kind |
---|---|---|---|
6532440 | Boppana | Mar 2003 | B1 |
6834256 | House | Dec 2004 | B2 |
7315856 | Iulo | Jan 2008 | B2 |
7975186 | Kato | Jul 2011 | B2 |
8190949 | Kato | May 2012 | B2 |
8286035 | Kato | Oct 2012 | B2 |
9189485 | Suzuki et al. | Nov 2015 | B2 |
10025653 | Goldstein | Jul 2018 | B2 |
10521193 | Tasaki | Dec 2019 | B2 |
11222077 | Martin | Jan 2022 | B1 |
20090217099 | Kato | Aug 2009 | A1 |
20110225462 | Kato | Sep 2011 | A1 |
20120192014 | Kato | Jul 2012 | A1 |
20130097128 | Suzuki et al. | Apr 2013 | A1 |
20130132000 | Tamaki | May 2013 | A1 |
20140157061 | Bird | Jun 2014 | A1 |
20140157235 | Bird | Jun 2014 | A1 |
20150205825 | Sengupta | Jul 2015 | A1 |
20150205827 | Sengupta | Jul 2015 | A1 |
20150206055 | Sengupta | Jul 2015 | A1 |
20160154690 | Horrell | Jun 2016 | A1 |
20160342392 | Tasaki | Nov 2016 | A1 |
20180231969 | Noda | Aug 2018 | A1 |
20180239345 | Noda | Aug 2018 | A1 |
20180293502 | Sengupta | Oct 2018 | A1 |
20210191726 | Tarango | Jun 2021 | A1 |
20210255613 | Mizobuchi | Aug 2021 | A1 |
20210263483 | Masuzaki | Aug 2021 | A1 |
Number | Date | Country |
---|---|---|
2004-240642 | Aug 2004 | JP |
2007-156653 | Jun 2007 | JP |
2009-199533 | Sep 2009 | JP |
2010-170178 | Aug 2010 | JP |
2013-218725 | Oct 2013 | JP |
2015-088078 | May 2015 | JP |
2015-152933 | Aug 2015 | JP |
2011135606 | Nov 2011 | WO |
2014142152 | Sep 2014 | WO |
WO-2019073512 | Apr 2019 | WO |
Entry |
---|
International Search Report of PCT/JP2017/033019 dated Dec. 5, 2017. |
Japanese Office Action for JP Application No. 2018-539744 dated Feb. 22, 2022 with English Translation. |
Number | Date | Country | |
---|---|---|---|
20190266039 A1 | Aug 2019 | US |