Patent Application
20160231738
The present invention relates to an information processing apparatus and an analysis method.
Analysis apparatuses which compare time series of measurement values (metric values) of sensors or the like in a system to be analyzed (a target system) to analyze the status of the target system are known.
As such an analysis apparatus, for example, PTL 1 describes an analysis apparatus (operation management apparatus) which employs a correlation model of a target system. The operation management apparatus described in PTL 1 determines, based on time series of measurement values of a plurality of sensors or the like of a target system, a correlation function representing the correlation between the sensors in a normal state by system identification method to generate a correlation model of the target system. The operation management apparatus then detects destruction (correlation destruction) of the correlation by using the generated correlation model to determine a fault cause of the target system.
[PTL 1] Japanese Patent Publication No. 4872944
When the above-described analysis apparatus is used to analyze states of target systems, a large time lag (time difference) as compared with a measurement interval of measurement value may be generated between measurement values of different sensors in some of the target systems. For example, a case in which the target system is a bridge and the sensor is a vibration sensor that is placed on the bridge is assumed. In this case, between measurement values of different sensors, there is a small time lag generated when a vibration of a vehicle passing through the bridge propagates through steel constituting the bridge and a large time lag generated when a moving vehicle approaches each sensor successively. In a case that the target system is an air-conditioner and the sensors are a current sensor and a temperature sensor for surroundings of the air-conditioner, the temperature changes in delay relative to a change of the state of the air-conditioner. There is thus a large time lag between measurement values of the current sensor and the temperature sensor. Further, in a case that the target system is a vehicle and the sensors are a sensor for a drive system such as an engine of the vehicle and a sensor for an operation system for a driver, there is a delay until a driver can take an action after recognizing a change in the state of the driving system. There are thus also a large time lag between measurement values of the drive system sensor and the operation system sensor.
When there is a large time lag between measurement values of sensors as described above, analysis accuracy may decrease.
For example, the operation management apparatus of PTL 1 determines a correlation function such as the following equation Math. 1 between sensors when a target system is analyzed.
Y(t)=A1Y(t−1)+A2Y(t−2)+ . . . +ANY(t−N)+B1X(t)+B2X(t−1)+ . . . +BMX(t−(M−1)) [Math. 1]
In the equation Math. 1, X(t) and Y(t) are measurement values of the sensors at time t which are an input and an output of the correlation function, respectively. The operation management apparatus determines the coefficients A1, A2, . . . , AN, B1, B2, . . . , BM in the equation Math. 1 for each pair of sensors in a plurality of sensors. Here, values of N and M are input in advance by, for example, a user. The operation management apparatus then detects correlation destruction between the sensors by using the determined correlation function.
When a time lag between sensors is larger than the value of M in the equation Math. 1, the correlation cannot be sufficiently approximated by the correlation function of the equation Math. 1, whereby prediction accuracy by the correlation function decreases. In this case, correlation destruction between the sensors cannot be accurately detected, whereby analysis accuracy decreases.
An object of the present invention is to resolve the above-described issue and to provide an information processing apparatus and an analysis method that prevent a decrease in analysis accuracy of the target system even when there is a large time lag between metrics of a target system.
An information processing apparatus according to an exemplary aspect of the invention includes: a processing means for performing comparative analysis of values of a first metric and a second metric in a system to be analyzed; and a pre-processing means for identifying, with respect to the comparative analysis of the first metric and the second metric, a temporal correspondence relation between respective pieces of data used as the first metric and the second metric.
An analysis method according to an exemplary aspect of the invention includes: identifying, with respect to a comparative analysis of a first metric and a second metric in a system to be analyzed, a temporal correspondence relation between respective pieces of data used as the first metric and the second metric; and performing the comparative analysis of values of the first metric and the second metric.
A computer readable storage medium according to an exemplary aspect of the invention records thereon a program, causing a computer to perform a method including: identifying, with respect to a comparative analysis of a first metric and a second metric in a system to be analyzed, a temporal correspondence relation between respective pieces of data used as the first metric and the second metric; and performing the comparative analysis of values of the first metric and the second metric.
An advantageous effect of the present invention is that a decrease in analysis accuracy of a target system can be prevented even when there is a large time lag between metrics of the target system.
A first exemplary embodiment of the present invention will be described.
First, a configuration of an analysis system of the first exemplary embodiment of the present invention will be described.
Referring to
The analysis apparatus 100 is one exemplary embodiment of an information processing apparatus of the present invention.
Here, the target system 200 is one of a variety of systems in which sensors for monitoring the state of systems are arranged. For example, the target system 200 is a structure such as a building or a bridge on which vibration sensors for diagnosing deterioration, temperature sensors, and/or the like are arranged. The target system 200 may be a plant such as a production plant or a power plant in which temperature sensors, flow sensors, and/or the like for monitoring an operation state are arranged. The target system 200 may be a mobile body such as a vehicle, a ship, or an aircraft in which measuring instruments such as a variety of sensors and/or sequencers for monitoring a driving state are incorporated. The target system 200 is not restricted to a system using physical sensors as described above, and may be a computer system which measures performance information for operation management. The target system 200 may be a computer system in which environment sensors for collecting power, temperature, or the like at the same time as the performance information are further arranged.
Here, each sensor and each performance item measured in the target system 200 are referred to as a “metric”. The term “metric” herein corresponds to the term “element” which is in PTL 1 referred to as a target for generating a correlation model.
The analysis apparatus 100 analyzes the state of the target system 200 based on a time series of measurement values of metrics collected from the target system 200.
The analysis apparatus 100 includes a data collection unit 101, a data storage unit 111, a pre-processing unit 130, a processing unit 140, a dialogue unit 106, and a handling execution unit 107.
The data collection unit 101 collects measurement values of each metric such as measurement values detected by each sensor, from the target system 200 at a predetermined collection interval.
The data storage unit 111 stores a time series of the measurement values of the metrics collected by the data collection unit 101 as time series data 121.
The pre-processing unit 130 identifies a temporal correspondence relation between metric values. In the first exemplary embodiment of the present invention, the pre-processing unit 130 identifies, as the temporal correspondence relation, a time lag of a time-series change of one metric relative to a time-series change of another metric.
The pre-processing unit 130 includes a time lag detection unit 102 and a time lag storage unit 112.
The time lag detection unit 102 generates time lag information 122 based on the time series data 121 stored in the data storage unit 111.
The time lag information 122 represents a time lag for each pair of metrics in a plurality of metrics. In the exemplary embodiment of the present invention, the time lag in a pair of metrics is a time difference between time-series changes of the pair of metrics in a case that similar changes in time-series occur in one of the pair of metrics (preceding metric) and the other of the pair of metrics (succeeding metric). The time lag may be a roughly estimated value of the time difference. In the exemplary embodiment of the present invention, the time lag is detected by comparing a series of a patterns of a time-series change of a metric for every predetermined length of time period (time period consisting of a plurality of monitoring intervals), as described below. For this reason, a value obtained by multiplying the length of time period for extracting a pattern of the time-series change by an integer is used for the value of a time lag.
The time lag storage unit 112 stores the time lag information 122 generated by the time lag detection unit 102.
The processing unit 140 performs a comparative analysis between metric values in a target system. In the first exemplary embodiment of the present invention, the processing unit 140 performs, as the comparative analysis, an analysis based on a correlation between metrics (generation of a correlation model 123 and detection of correlation destruction).
The processing unit 140 includes a correlation model generation unit 103, a correlation model storage unit 113, a correlation destruction detection unit 104, and a data extraction unit 105.
The correlation model generation unit 103 generates the correlation model 123 of the target system 200 based on the time series data 121 stored in the data storage unit 111.
The correlation model 123 includes a correlation function (or a conversion function) representing a correlation of each pair of metrics in a plurality of metrics. For example, the correlation function is represented by the above-described equation Math. 1. In other words, the correlation function is a function that predicts a value of one metric (output metric) of a pair of metrics at time t from a measurement value of the other metric (input metric) at or before time t and a measurement value of the one metric (output metric) before time t. The correlation model generation unit 103 determines coefficients of a correlation function in a similar manner to the operation management apparatus of PTL 1. In other words, the correlation model generation unit 103 determines coefficients A1, A2, . . . , AN, B1, B2, . . . , BM of the correlation function of the equation Math. 1 for each pair of metrics by system identification processing based on the time series data 121 of a predetermined modeling time period.
On the other hand, in the exemplary embodiment of the present invention, a shifted (delayed) time series is used. The shifted time series is obtained by shifting (delaying) the above-described time series of a preceding metric by a time lag, with reference to the time lag information 122 stored in the time lag storage unit 112. The correlation model generation unit 103 determines the correlation function using a time series obtained by shifting (delaying) the above-described time series of a preceding metric by a time lag as a time series of an input metric, and using the above-described time series of a succeeding metric as a time series of an output metric.
In a similar manner to the operation management apparatus of PTL 1, the correlation model generation unit 103 may calculate a weight for each pair of metrics based on a conversion error of a correlation function to obtain, as the correlation model 123, a set of correlation functions (effective correlation functions) in which the weight is equal to or larger than a predetermined value.
The correlation model storage unit 113 stores the correlation model 123 generated by the correlation model generation unit 103.
In a similar manner to the operation management apparatus of PTL 1, the correlation destruction detection unit 104 determines whether a correlation included in the correlation model 123 is maintained or destroyed by using newly collected measurement values of the metrics.
Here, the correlation destruction detection unit 104 uses newly collected measurement values of the metrics which are extracted by the data extraction unit 105. The correlation destruction detection unit 104 calculates, for each pair of the metrics, a difference (prediction error) between a measurement value of an output metric at time t and a prediction value of an output metric at time t calculated by using the correlation function. The correlation destruction detection unit 104 determines that the correlation is destroyed when the calculated difference is equal to or more than a predetermined value.
The data extraction unit 105 extracts newly collected measurement values of the metrics which are needed for detection of correlation destruction from the time series data 121 stored in the data storage unit 111, and outputs them to the correlation destruction detection unit 104. The data extraction unit 105 outputs measurement values of the input metric a time lag earlier with reference to the time lag information 122 stored in the time lag storage unit 112.
The dialogue unit 106 presents a detection result of correlation destruction to a user or the like.
The dialogue unit 106 may also instructs the handling execution unit 107 to execute an operation for handling the correlation destruction in accordance with an operation of a user or the like. In this case, the handling execution unit 107 executes an instructed handling operation on the target system 200.
The analysis apparatus 100 may be a computer which includes a CPU (Central Processing Unit) and a storage medium storing a program, and operates by control based on the program. The data storage unit 111, the correlation model storage unit 113, and the time lag storage unit 112 may be individual storage media, or may be configured as one storage medium.
Referring to
Next, an operation of the analysis apparatus 100 in the first exemplary embodiment of the present invention will be described.
First, the data collection unit 101 collects measurement values for each metric from the target system 200 at a predetermined collection interval (step S101). The data collection unit 101 stores a time series of the collected measurement values for each metric as the time series data 121 in the data storage unit 111.
For example, the data collection unit 101 stores time series data 121 as illustrated in
The time lag detection unit 102 of the pre-processing unit 130 generates the time lag information 122 based on the time series data 121 (step S102). The time lag detection unit 102 stores the generated time lag information 122 in the time lag storage unit 112.
Here, the time lag detection unit 102 detects a time lag by, for example, comparing, for each pair of metrics, series of time-series change patterns of the metrics for each predetermined length of time period.
The time lag detection unit 102 gives, in accordance with the change patterns as illustrated in
For example, in
The time lag detection unit 102 then stores the time lag information 122 as illustrated in
When a plurality of candidates for a time lag for a pair of metrics are detected in the step S102, the dialogue unit 106 may present the candidates for a time lag to a user or the like and receive an input for selection of a time lag from a user.
In this case, the dialogue unit 106 displays, for example, the display screen 501 of
The correlation model generation unit 103 of the processing unit 140 generates the correlation model 123 for each pair of metrics contained in the time lag information 122 (step S103). Here, the correlation model generation unit 103 uses a time series obtained by shifting (delaying) a time series of the preceding metric by the time lag as the time series of the input metric, and uses a time series of the succeeding metric as the time series of the output metric. The correlation model generation unit 103 stores the generated correlation model 123 in the correlation model storage unit 113.
For example, it is assumed that measurement values of the sensor 1 and the sensor 3 are S1(t) and S3(t), respectively. In this case, the correlation model generation unit 103 determines, with respect to a pair of the sensor 1 and the sensor 3 in the time lag information 122 of
As a result, the correlation model generation unit 103 determines a correlation function, for example, as illustrated in
The data collection unit 101 collects a new measurement value of each metric from the target system 200 (step S104). The data collection unit 101 stores time series of the collected new measurement values of the metrics as the time series data 121 in the data storage unit 111.
The data extraction unit 105 extracts, with respect to each correlation function included in the correlation model 123, measurement values needed for detection of correlation destruction from the new measurement values of the metrics, and outputs them to the correlation destruction detection unit 104 (step S105). Here, the data extraction unit 105 extracts a measurement value of the input metric at or before time t and a measurement value of the output metric before time t needed for calculation of a prediction value of the output metric of the correlation function at time t. It is noted that the data extraction unit 105 extracts a measurement value of the input metric that goes back by the length of the time, lag.
For example, the data extraction unit 105 outputs, with respect to a pair of the sensor 1 and the sensor 3 in the correlation model 123 of
The correlation destruction detection unit 104 detects correlation destruction with respect to each correlation function included in the correlation model 123 by using the new measurement value of each metric extracted by the data extraction unit 105 (step S106). Here, the correlation destruction detection unit 104 applies the measurement value of the input metric at or before time t and the measurement value of the output metric before time t extracted by the data extraction unit 105 to the correlation function and calculates a prediction value of the output metric at time t. The correlation destruction detection unit 104 then detects correlation destruction based on the difference between the measurement value and the prediction value of the output metric at time t.
For example, with respect to the pair of the sensor 1 and the sensor 3 in the correlation model 123 of
The dialogue unit 106 presents a detection result of correlation destruction to a user or the like (step S107).
The dialogue unit 106 displays, for example, the display screen 501 of
The dialogue unit 106 may further display a time series graph corrected for the length of the time lag so that time-series changes between metrics are easily compared with each other.
The dialogue unit 106 displays, for example, the display screen 501 of
Thereafter, the dialogue unit 106 instructs the handling execution unit 107 to handle the correlation destruction in accordance with an operation from a user or the like. The handling execution unit 107 executes instructed handling on the target system 200.
This completes the operation of the first exemplary embodiment of the present invention.
In the first exemplary embodiment of the present invention, an analysis based on correlation is performed as a comparative analysis between metric values in a target system. The analysis, however, is not limited thereto, and analyses other than the analysis based on correlation may be performed as long as the analysis is a comparative analysis affected by a time lag between metrics.
In the first exemplary embodiment of the present invention, the equation Math. 1 is used for the correlation function. The equation is, however, not limited thereto, and other correlation functions may be used as long as the correlation function represents a correlation between a pair of metrics.
In the first exemplary embodiment of the present invention, the time lag of each pair of metrics is detected by comparing series of time-series change patterns with one another. The detection method is, however, not limited thereto, and the time lag may be detected by other methods such as a method using a time difference between times at which metric values are maximum or minimum or a method using a phase difference between time series of metrics, as long as the time lag can be detected based on change in the metric value.
In the first exemplary embodiment of the present invention, a change pattern as illustrated in
In the first exemplary embodiment of the present invention, a time difference is used as the time lag, in a case that a change pattern series of a succeeding metric which is similar (same in the increase-decrease trend) to a change pattern series of a preceding metric occurs. The time difference is, however, not limited thereto, and a time difference may also be used as the time lag, even in a case that the change pattern series of the succeeding metric is opposite to the change pattern series of the preceding metric in the increase-decrease trend.
Next, a characteristic configuration of the first exemplary embodiment of the present invention will be described.
The analysis apparatus (information processing apparatus) 100 includes the processing unit 140 and the pre-processing unit 130. The processing unit 140 performs comparative analysis of values of a first metric and a second metric in a system to be analyzed. The pre-processing unit 130 identifies, with respect to the comparative analysis of the first metric and the second metric, a temporal correspondence relation between respective pieces of data used as the first metric and the second metric.
Next, an advantageous effect of the first exemplary embodiment of the present invention will be described.
According to the first exemplary embodiment of the present invention, a decrease in analysis accuracy of a target system can be prevented even when there is a large time lag between metrics of the target system. This is because the pre-processing unit 130 identifies, with respect to the comparative analysis of the first metric and the second metric, a temporal correspondence relation between respective pieces of data used as the first metric and the second metric.
It has been difficult to set the size of a time lag for a correlation as knowledge in advance since the size varies depending on a target system to which sensors are arranged or operating conditions of the target system.
According to the first exemplary embodiment of the present invention, a time lag between metrics depending on a target system can be easily identified. This is because the time lag detection unit 102 detects a time lag by comparing a series of time-series change patterns of a preceding metric, each time-series change pattern being obtained for a predetermined length of time period, with a series of time-series change patterns of a succeeding metric, each time-series change pattern being obtained for the predetermined length of time period.
For the correlation destruction detection unit 104, for example, an analysis engine that uses a dedicated hardware or program may have been used since correlation destruction needs to be detected in real time for newly collected large number of measurement values.
According to the first exemplary embodiment of the present invention, the correlation destruction detection unit 104 (analysis engine) that cannot handle a large time lag can be utilized as it is even when there is a large time lag between metrics of a target system. This is because the data extraction unit 105 extracts a measurement value of the input metric a time lag earlier, and the correlation destruction detection unit 104 detects correlation destruction by using the measurement value extracted by the data extraction unit 105.
Next, a second exemplary embodiment of the present invention will be described.
The second exemplary embodiment of the present invention is different from the first exemplary embodiment of the present invention in that generation of the time lag information 122 and generation of the correlation model 123 are performed in an apparatus which is different from the analysis apparatus 100. In the second exemplary embodiment of the present invention, a component to which the same reference sign is assigned is the same as the component in the first exemplary embodiment of the present invention unless otherwise specified.
Referring to
The analysis apparatus 100 includes the data collection unit 101, the data storage unit 111, the time lag storage unit 112, the correlation model storage unit 113, the correlation destruction detection unit 104, the data extraction unit 105, and the handling execution unit 107, similarly to the analysis apparatus 100 (
The monitoring apparatus 150 includes the time lag detection unit 102, the correlation model generation unit 103, and the dialogue unit 106, similarly to the analysis apparatus 100 (
The time lag detection unit 102 of the monitoring apparatus 150 generates the time lag information 122 based on time series data 121 stored in the data storage unit 111 of the analysis apparatus 100.
The correlation model generation unit 103 of the monitoring apparatus 150 generates the correlation model 123 based on the time series data 121 stored in the data storage unit 111 and the time lag information 122 stored in the time lag storage unit 112 of the analysis apparatus 100.
The handling determination unit 108 of the analysis apparatus 100 determines a handling process to be executed in a predetermined condition depending on a detection result of correlation destruction obtained by the correlation destruction detection unit 104, and instructs the execution of handling to the handling execution unit 107. The handling determination unit 108 then presents the execution result of handling to a user or the like through the dialogue unit 106 of the monitoring apparatus 150. The handling determination unit 108 may instructs the handling execution unit 107 to execute an operation for handling in accordance with an operation of a user or the like which is input via the dialogue unit 106, similarly to the first exemplary embodiment of the present invention.
The time lag detection unit 102 and the correlation model generation unit 103 may be included in a still another apparatus which is different from the monitoring apparatus 150.
Next, an advantageous effect of the second exemplary embodiment of the present invention will be described.
According to the second exemplary embodiment of the present invention, a real time abnormality analysis of the target system 200 by detection of correlation destruction can be performed more rapidly compared with the first exemplary embodiment of the present invention. This is because generation of the time lag information 122 by the time lag detection unit 102 and generation of the correlation model 123 by the correlation model generation unit 103 which become a relatively high processing load are executed by an apparatus which is different from the analysis apparatus 100 which performs the abnormality analysis.
While the invention has been particularly shown and described with reference to exemplary embodiments thereof, the invention is not limited to these embodiments. It will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the spirit and scope of the present invention as defined by the claims.
This application is based upon and claims the benefit of priority from Japanese Patent Application No. 2013-199943, filed on Sep. 26, 2013, the disclosure of which is incorporated herein in its entirety by reference.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2013-199943 | Sep 2013 | JP | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/JP2014/004707 | 9/11/2014 | WO | 00 |
