WO2015174063A1 - 情報処理装置、分析方法、及び、記録媒体 - Google Patents
情報処理装置、分析方法、及び、記録媒体 Download PDFInfo
- Publication number
- WO2015174063A1 WO2015174063A1 PCT/JP2015/002365 JP2015002365W WO2015174063A1 WO 2015174063 A1 WO2015174063 A1 WO 2015174063A1 JP 2015002365 W JP2015002365 W JP 2015002365W WO 2015174063 A1 WO2015174063 A1 WO 2015174063A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- correlation
- time series
- model
- learning
- reliability
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/22—Detection or location of defective computer hardware by testing during standby operation or during idle time, e.g. start-up testing
- G06F11/2257—Detection or location of defective computer hardware by testing during standby operation or during idle time, e.g. start-up testing using expert systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
- G06F11/3447—Performance evaluation by modeling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
- G06F11/3409—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment for performance assessment
Definitions
- the present invention relates to an information processing apparatus, an analysis method, and a recording medium, and more particularly, to an information processing apparatus, an analysis method, and a recording medium that perform system analysis using a correlation.
- Patent Document 1 An example of an operation management system that performs system modeling based on the correlation between time series of system performance and determines factors such as system failures and abnormalities using the generated model is disclosed in Patent Document 1. ing.
- the operation management system described in Patent Literature 1 is based on a time series of measured values of a plurality of metrics when the system is normal (learning period), and represents a correlation between each pair of the plurality of metrics. Determine the function. Then, the operation management system generates a correlation model of the system by selecting the correlation according to the weight calculated based on the error of the correlation function. Furthermore, the operation management system detects the destruction of the correlation (correlation destruction) using the generated correlation model, and determines the failure factor of the system based on the correlation destruction.
- the technique for analyzing the state of the system based on correlation destruction is called invariant relation analysis.
- a correlation function that predicts the value of metric y based on the value of metric u is used. Then, using the time series at the time of model generation, the difference between the actually measured value of metric y and the predicted value by the correlation function, that is, the prediction error is calculated. Further, based on the calculated prediction error, a prediction error threshold allowed at the time of monitoring is set. If the prediction error during monitoring exceeds a threshold (correlation destruction is detected), it is determined that an abnormality has occurred in the system.
- Patent Document 2 discloses a facility state monitoring method for detecting a system abnormality using a time series of system performance.
- an operation pattern label is given to a time-series signal output from the equipment at regular intervals, and a normal model is constructed for each label.
- an operation pattern label is assigned to the detection target period, and abnormality detection is performed using a normal model having the same or similar label.
- Patent Document 3 in an operation management system that performs invariant relationship analysis, based on the degree of fitness for performance information for a predetermined period, from a plurality of correlation models generated for the predetermined period, A method for extracting a basic model and a specific model is disclosed.
- An object of the present invention is to provide an information processing apparatus, an analysis method, and a recording medium that can solve the above-described problems, improve anomaly detection capability in invariant relationship analysis, and reduce erroneous anomaly reports. is there.
- a system analysis apparatus includes: a correlation model generation unit that generates a correlation model including a correlation between metrics based on a time series of learning periods of a plurality of metrics in the system; Learning reliability calculation means for calculating the learning reliability of the correlation based on the time-series behavior of the learning period of each of the metrics related to the correlation included.
- An analysis method generates a correlation model including a correlation between metrics based on a time series of learning periods of a plurality of metrics in the system, and relates to the correlation included in the correlation model.
- the learning reliability of the correlation is calculated based on the time-series behavior of the learning period of each metric.
- the computer-readable recording medium generates a correlation model including a correlation between metrics based on a time series of learning periods of a plurality of metrics in the system.
- a program for executing processing for calculating the learning reliability of the correlation is stored based on the time-series behavior of each of the metrics related to the correlation included in the learning period.
- the effect of the present invention is that in the invariant relationship analysis, the abnormality detection capability can be improved and erroneous abnormality reports can be reduced.
- step S200 shows the detail of the correlation change analysis process (step S200) in the 1st Embodiment of this invention. It is a figure which shows the example of the format of the single time series model in the 1st Embodiment of this invention. It is a figure which shows the example of the time series of performance information in the 1st Embodiment of this invention. 10 is a graph showing metrics A and B in the time series of FIG. 9. 10 is a graph showing metrics C and D in the time series of FIG. 9. It is a figure which shows the example of calculation of learning reliability in the 1st Embodiment of this invention. It is a figure which shows the other example of calculation of the learning reliability in the 1st Embodiment of this invention.
- FIG. 2 is a block diagram showing a configuration of the system analysis apparatus 100 according to the first embodiment of the present invention.
- the system analysis apparatus 100 is an embodiment of the information processing apparatus of the present invention.
- the system analysis apparatus 100 is connected to the monitored system 500 via a network or the like.
- the monitored system 500 is a system that provides an information communication service such as a WEB service or a business service, or a system such as a plant or a power generation facility.
- the monitored system 500 outputs a time series of measured values of system performance.
- the monitored system 500 measures the measured values of the performance values of a plurality of types at regular intervals and transmits them to the system analyzer 100.
- the performance value items for example, CPU (Central Processing Unit) usage rate, memory usage rate, disk access frequency, and the like, usage rates and usage amounts of computer resources and network resources are used.
- the performance value item power, voltage, current, temperature, pressure, or the like measured by various sensors may be used.
- the type of performance value is defined as a metric (performance index), and a set of multiple metric values measured at the same time is defined as performance information.
- Metric values are represented by integers or decimals.
- a metric corresponds to an “element” that is a generation target of a correlation model in Patent Document 1.
- the system analysis apparatus 100 generates a correlation model of the monitored system 500 based on the time series of performance information collected from the monitored system 500, and analyzes the state of the monitored system 500 using the generated correlation model. To do.
- the system analysis apparatus 100 includes a performance information collection unit 110, a performance information storage unit 120, a correlation model generation unit 130, a learning reliability calculation unit 140, a correlation model storage unit 150, a correlation change analysis unit 160, an analysis setting storage unit 170, and The failure analysis unit 180 is included.
- the performance information collection unit 110 collects a time series of performance information from the monitored system 500.
- the performance information storage unit 120 stores a time series of performance information collected by the performance information collection unit 110 during the learning period.
- the correlation model generation unit 130 generates a correlation model of the monitored system 500 based on the time series of the performance information in the learning period, as in Patent Document 1.
- the correlation model includes the correlation of each pair of multiple metrics.
- the correlation is represented by a correlation function (or conversion function) between metrics.
- the correlation function is a function that predicts the value of the other metric (output metric) from the value of one metric (input metric) of the pair of metrics.
- the correlation model generation unit 130 calculates the parameters of the correlation function for each metric pair using the time series of performance information in the learning period stored in the performance information collection unit 110.
- the parameters of the correlation function are determined by the system identification process for the time series of metrics, as in Patent Document 1.
- the correlation model generation unit 130 generates a correlation model by repeating these processes for all pairs of metrics.
- the correlation model generation unit 130 may calculate a weight according to the prediction error of the correlation function and give it to the correlation, as in Patent Document 1.
- the correlation model generation unit 130 may select the correlation according to the weight.
- the learning reliability calculation unit 140 calculates the learning reliability of the correlation included in the correlation model using the time series of performance information stored in the performance information collection unit 110 during the learning period.
- the learning reliability indicates whether the correlation is learning the relationship between metrics.
- the correlation when the time series of each learning period of the metric related to the correlation shows a specific behavior (behavior), the correlation sufficiently learns the relationship between the actual metrics. It is assumed that the learning reliability is low.
- the specific behavior includes, for example, a time series of metrics indicating a constant value, indicating one of two values, or indicating a linear change.
- the correlation change analysis when the value of the metric decreases or when the increase / decrease is repeated, there is a high possibility that even a normal behavior is determined to be abnormal (an erroneous abnormality report is output).
- a time series model (single time series model) showing the above specific behavior is generated based on the time series in the learning period of each metric related to the correlation. Then, the learning reliability is calculated based on the suitability of the single time series model with respect to the time series in the learning period of each metric.
- the learning reliability is calculated so as to be low when the fitness is high and to be high when the fitness is low according to the fitness.
- the fitness is calculated using the prediction error of the single time series model with respect to the time series in the learning period of each metric.
- the learning reliability is calculated to be low when the prediction error is small (high fitness) and high when the prediction error is large (low fitness) according to the prediction error of the single time series model. Is done.
- the correlation model storage unit 150 adds the learning reliability calculated by the learning reliability calculation unit 140 to each correlation of the correlation model generated by the correlation model generation unit 130 and stores the correlation.
- Correlation change analysis unit 160 acquires a correlation model to which learning reliability is added from correlation model storage unit 150, and extracts a correlation whose learning reliability is equal to or higher than a predetermined threshold.
- the correlation change analysis unit 160 may further extract the correlation by the weight.
- the correlation change analysis unit 160 calculates a correlation function prediction error for each extracted correlation by using a time series of performance information in a period (monitoring period) to be subjected to correlation change analysis. Detects whether there is destruction.
- the analysis setting storage unit 170 stores an analysis setting indicating a method and conditions for the failure analysis unit 180 to perform failure analysis. For example, in the analysis setting, conditions relating to the number and rate of correlation destruction, etc., for which the failure analysis unit 180 notifies abnormality (issues a warning) are set.
- the failure analysis unit 180 performs failure analysis according to the analysis setting.
- FIG. 3 is a block diagram showing a configuration of the learning reliability calculation unit 140 in the first embodiment of the present invention.
- the learning reliability calculation unit 140 includes a time series model storage unit 141, a time series model generation unit 142, a prediction error calculation unit 143, and a reliability calculation unit 144.
- the time series model storage unit 141 stores the format of a single time series model.
- the single time series model is a time series model for modeling the time series of each metric.
- a time series model is used that exhibits a behavior that is judged to have a low correlation learning reliability.
- FIG. 8 is a diagram showing an example of the format of a single time series model in the first embodiment of the present invention.
- a constant value model, a binary model, and a straight line model are set as the format of the single time series model.
- X (i) is the value of metric X at time i.
- a, b, and c are parameters.
- the time series model generation unit 142 determines parameters of a single time series model of each format stored in the time series model storage unit 141 based on the time series in the learning period of each metric (generates a single time series model). ) The parameters of the single time series model are determined by, for example, the system identification process for the time series of metrics.
- the prediction error calculation unit 143 calculates the prediction error due to the single time series model as the suitability of the single time series model with respect to the time series of the learning period of each metric.
- the prediction error is calculated, for example, by the root mean square of the difference between the predicted value obtained by applying the time series in the learning period to the single time series model and the actual measurement value.
- the reliability calculation unit 144 calculates the learning reliability of the correlation.
- the reliability calculation unit 144 is configured such that the higher the fitness of the single time series model generated for each metric related to the correlation is (the prediction error is smaller), the lower the learning reliability of the correlation is. Determine learning confidence.
- the reliability calculation unit 144 extracts the smallest prediction error among the prediction errors of a plurality of single time series models calculated for each metric related to the correlation. And the reliability calculation part 144 calculates the sum total of the prediction error extracted about each metric which concerns on correlation as the learning reliability of the said correlation.
- the prediction error of the single time series model is used as the suitability of the single time series model, but the degree of fit of the single time series model with respect to the time series of the learning period is expressed. If it is possible, other than the prediction error may be used.
- the single time series model is a constant value model or a binary model
- the number of times that these constant values or values other than the binary value are shown may be used as the fitness.
- a method for calculating the fitness of the single time series model may be given to the format of each single time series model stored in the time series model storage unit 141.
- system analysis apparatus 100 may be a computer that includes a CPU and a storage medium that stores a program and that operates under control based on the program. Further, the performance information storage unit 120, the correlation model storage unit 150, and the analysis setting storage unit 170 may be configured as individual storage media or a single storage medium.
- FIG. 4 is a block diagram showing a configuration of the system analysis apparatus 100 realized by a computer according to the first embodiment of the present invention.
- the system analysis apparatus 100 includes a CPU 101, a storage unit (storage medium) 102 such as a hard disk and a memory, a communication unit 103 that performs data communication with other devices, an input unit 104 such as a keyboard, and an output unit 105 such as a display. Including.
- the CPU 101 executes a computer program for realizing the functions of the performance information collection unit 110, the correlation model generation unit 130, the learning reliability calculation unit 140, the correlation change analysis unit 160, and the failure analysis unit 180.
- the storage unit 102 stores data of the performance information storage unit 120, the correlation model storage unit 150, and the analysis setting storage unit 170.
- the communication unit 103 receives a time series of performance information from the monitored device 500.
- the input unit 104 receives input of various threshold values and analysis settings from a user or the like.
- the output unit 105 outputs the result of failure analysis to the user or the like.
- each component of the system analysis apparatus 100 shown in FIG. 2 may be an independent logic circuit.
- FIG. 5 is a flowchart showing the overall processing of the system analysis apparatus 100 in the first embodiment of the present invention.
- the system analysis apparatus 100 generates a correlation model (step S100). Then, the system analysis device 100 performs correlation change analysis (invariant relationship analysis) using the generated correlation model (step S200).
- FIG. 6 is a flowchart showing details of the correlation model generation process (step S100) in the first embodiment of the present invention.
- Y (i) a * X (i) + b (a and b are parameters and i is a time) is used as a form of a correlation function representing a correlation between metrics X and Y.
- the performance information collection unit 110 collects a time series of performance information during the learning period from the monitored system 500 (step S101).
- the performance information collection unit 110 stores the collected time series of performance information in the performance information storage unit 120.
- FIG. 9 is a diagram showing an example of time series of performance information in the first exemplary embodiment of the present invention.
- the performance information collection unit 110 collects and stores the time series of metrics A, B, C, and D as the time series of performance information as shown in FIG.
- the correlation model generation unit 130 selects one metric pair from the performance information stored in the performance information collection unit 110 (step S102).
- the correlation model generation unit 130 calculates a correlation function for the selected metric pair using the time series of the learning period (step S103).
- FIG. 10 is a graph showing metrics A and B in the time series of FIG.
- the learning reliability calculation unit 140 calculates the learning reliability of the correlation (Step S104).
- the time series model generation unit 142 of the learning reliability calculation unit 140 has a single time series model of each format stored in the time series model storage unit 141 for the time series of the learning period of each metric related to the correlation. Is generated.
- the prediction error calculation unit 143 calculates a prediction error of each single time series model with respect to the time series of the learning period.
- the reliability calculation unit 144 extracts the smallest prediction error among the prediction errors of the single time series model calculated for each metric.
- the reliability calculation part 144 calculates the sum total of the prediction error extracted about each metric as the learning reliability of the said correlation.
- FIG. 12 is a diagram illustrating an example of calculating the learning reliability according to the first embodiment of this invention.
- the learning reliability calculation unit 140 generates a single time series model for each of the metrics A and B in the time series of FIG. 9 as shown in FIG. 12, and totals the prediction errors (minimum values) of the single time series model. Then, the learning reliability “0” of the correlation between the metrics A and B is calculated.
- the learning reliability calculation unit 140 gives the learning reliability to the correlation and stores it in the correlation model storage unit 150 (step S105).
- FIG. 14 is a diagram showing an example of a correlation model in the first embodiment of the present invention.
- the learning reliability calculation unit 140 assigns and stores the learning reliability to the correlation between the metrics A and B as shown in FIG.
- steps S102 to S105 are repeated for all metric pairs (step S106).
- FIG. 11 is a graph showing metrics C and D in the time series of FIG.
- FIG. 13 is a diagram illustrating another calculation example of the learning reliability according to the first embodiment of this invention.
- the learning reliability calculation unit 140 generates a single time series model for each of the metrics C and D in the time series of FIG. 9, as shown in FIG. 13, and the learning reliability “2. 474 "is calculated.
- the learning reliability calculation unit 140 assigns the learning reliability to the correlation between the metrics C and D as shown in FIG.
- FIG. 7 is a flowchart showing details of the correlation change analysis process (step S200) in the first embodiment of the present invention.
- the correlation change analysis unit 160 acquires a correlation model to which learning reliability is added from the correlation model storage unit 150 (step S201).
- the correlation change analysis unit 160 extracts a correlation whose learning reliability is equal to or higher than a predetermined reliability threshold value from the correlation model (step S202).
- the correlation change analysis unit 160 extracts a correlation between metrics C and D having a learning reliability of “2.474” from the correlation model of FIG. To do.
- the correlation between metrics A and B is not extracted.
- Correlation change analysis unit 160 selects one of the correlations extracted in step S202 (step S203).
- the correlation change analysis unit 160 calculates a prediction error with respect to the time series of the performance information monitoring period collected by the performance information collection unit 110 for the selected correlation (step S204).
- the correlation change analysis unit 160 determines that correlation destruction has been detected (step S206), and calculates a correlation destruction abnormality score (step S207).
- the correlation change analysis unit 160 repeats the processing of steps S203 to S207 for all the correlations extracted in step S202 (step S208).
- the correlation change analysis unit 160 detects the presence or absence of correlation destruction using the time series of the monitoring period for the correlation between the metrics C and D extracted in step S202.
- the failure analysis unit 180 performs failure analysis according to the analysis settings stored in the analysis setting storage unit 170 (step S209).
- the failure analysis unit 180 outputs details of the detected correlation destruction and the result of failure analysis to the user or the like.
- FIG. 1 is a block diagram showing a characteristic configuration of the first embodiment of the present invention.
- the system analysis apparatus 100 includes a correlation model generation unit 130 and a learning reliability calculation unit 140.
- the correlation model generation unit 130 generates a correlation model including a correlation between metrics based on a time series of learning periods of a plurality of metrics in the system.
- the learning reliability calculation unit 140 calculates the learning reliability of the correlation based on the time-series behavior of each metric related to the correlation included in the correlation model.
- the learning reliability calculation unit 140 calculates the learning reliability of the correlation based on the time-series behavior of each learning period of the metric related to the correlation included in the correlation model. . Thereby, an invariant relationship analysis can be performed using a correlation with high learning reliability.
- an invariant relationship analysis can be performed by selecting an appropriate correlation from a plurality of correlations included in the correlation model of the system. This is because the learning reliability calculation unit 140 calculates the learning reliability for each correlation included in the correlation model.
- the time series of performance information is divided in the time direction, and the learning reliability is calculated based on the divided time series. Different from form.
- the time series model generation unit 142 of the learning reliability calculation unit 140 divides the time series of each metric related to the correlation in the learning period in the time direction (first division). ), A single time series model is generated for each divided section by the first division.
- the sum of the prediction errors of the single time series model is minimized from a combination of all possible divisions and the single time series model generated for each divided time series. This is done by selecting such a combination. For example, when a straight line model is used as the single time series model, and the time series of the metric shows a monotone increase and a monotone decrease, the time series is an interval showing the monotone increase and the monotone decrease. Divided.
- FIG. 15 is a diagram showing a time-series division example in the second embodiment of the present invention.
- the time series model generation unit 142 converts the time series of the metric C in the time series of FIG. 9 from time 1 to 6 (division interval c1), time 7 to 12 (division interval c2), as shown in FIG. And it divides
- the time series model generation unit 142 converts the time series of the metric D from time 1 to 8 (division interval d1), time 9 to 12 (division interval d2), and time 13 to 20 (division interval d3). Divide into three.
- FIG. 16 is a diagram illustrating a generation example of a single time series model according to the second embodiment of the present invention.
- the time series model generation unit 142 generates a single time series model for each of the metrics C and D for each divided section (c1, c2, c3, d1, d2, d3) as shown in FIG.
- an upper limit may be set for the number of time-series divisions.
- the prediction error calculation unit 143 further divides the divided section by the first division so that the time series of both metrics is divided at the time at which the time series of either metric is divided (the second division). Split).
- the prediction error calculation unit 143 converts the time series of the metrics C and D from time 1 to 6 (division interval cd1), time 7 to 8 (division interval cd2), time 9 to 12 ( It is divided into four divided sections cd3) and times 13 to 20 (divided sections cd4).
- the prediction error calculation unit 143 assigns a combination of single time series models for each divided section by the second division, and calculates a prediction error of the assigned single time series model.
- the reliability calculation part 144 calculates the sum total of the prediction error calculated for every division
- FIG. 17 is a diagram showing a calculation example of learning reliability in the second embodiment of the present invention.
- the prediction error calculation unit 143 allocates a combination of single time series models for each divided section (cd1, cd2, cd3, cd4), and calculates a prediction error of the single time series model.
- the reliability calculation unit 144 calculates the learning reliability “0.165928” of the correlation between the metrics C and D by summing up the prediction errors for each divided section.
- the learning reliability calculation unit 140 gives the learning reliability to the correlation and stores it in the correlation model storage unit 150.
- FIG. 18 is a diagram showing an example of a correlation model in the second embodiment of the present invention.
- the reliability calculation unit 144 assigns a learning reliability to the correlation between the metrics C and D as shown in FIG.
- the second division was performed so that the time series of both metrics were divided. Then, the learning reliability was calculated by summing the prediction errors calculated for each divided section in which the second division was performed.
- the present invention is not limited to this.
- the second division may be omitted, and the learning reliability may be calculated by adding the prediction errors calculated for each divided section in which the first division is performed.
- the learning reliability can be calculated even when the time-series behavior of the metric in the learning period conforms to a single time-series model that varies with time.
- the reason is that the learning reliability calculation unit 140 divides the time series of each learning period of the metric related to the correlation in the time direction, and based on the suitability of the single time series model for each of the divided time series. This is because the learning reliability is calculated.
- the learning reliability is based on the degree of adaptation of the single time series model to the time series of the learning period and the degree of adaptation of the single time series model to the time series of the monitoring period. Is different from the first embodiment of the present invention in that it is calculated.
- FIG. 19 is a block diagram illustrating a configuration of the system analysis apparatus 100 according to the third embodiment of the present invention.
- the learning reliability calculation unit 140 of the system analysis apparatus 100 includes a learning reliability model generation unit 190 and a learning reliability determination unit 195.
- the learning reliability model generation unit 190 generates a learning reliability model of the correlation included in the correlation model, using the time series of performance information stored in the performance information collection unit 110 during the learning period.
- the learning reliability model indicates a method of determining the learning reliability based on a time series of performance information in the monitoring period.
- the correlation is sufficient for the relationship between the metrics in the monitoring period. Assume that learning is performed and the learning reliability of the correlation is high.
- the correlation model storage unit 150 adds the learning reliability model generated by the learning reliability model generation unit 190 to each correlation of the correlation model generated by the correlation model generation unit 130 and stores the correlation.
- the learning reliability determination unit 195 determines the learning reliability using the time series of performance information in the monitoring period and the learning reliability model.
- FIG. 20 is a block diagram showing a configuration of the learning reliability model generation unit 190 in the third embodiment of the present invention.
- the learning reliability model generation unit 190 includes a time series model storage unit 191, a time series model generation unit 192, and a reliability model generation unit 193.
- the time series model storage unit 191 stores the format of a single time series model.
- the time series model generation unit 192 determines the parameters of the single time series model of each format stored in the time series model storage unit 191 based on the time series in the learning period of each metric (generates a single time series model) )
- the reliability model generation unit 193 generates the above-described learning reliability model based on the suitability of the single time series model generated by the time series model generation unit 192 with respect to the time series in the learning period.
- FIG. 21 is a block diagram showing a configuration of the learning reliability determination unit 195 in the third embodiment of the present invention.
- the learning reliability determination unit 195 includes a time series model storage unit 196, a time series model generation unit 197, and a reliability determination unit 198.
- the time series model storage unit 196 stores the format of a single time series model.
- the time series model storage unit 196 may store the same format as the time series model storage unit 191 as a single time series model format, or may store a different format.
- the time series model generation unit 197 determines parameters of a single time series model of each format stored in the time series model storage unit 196 based on the time series in the monitoring period of each metric (generates a single time series model) )
- the time series model generation unit 197 may generate a single time series model by the same method as the time series model generation unit 192, or may generate a single time series model by a different method.
- the reliability determination unit 198 determines the learning reliability using the suitability of the single time series model generated by the time series model generation unit 197 with respect to the time series in the monitoring period and the learning reliability model.
- FIG. 22 is a flowchart showing details of the correlation model generation process (step S100) in the third embodiment of the present invention.
- processing from when the performance information collection unit 110 collects the time series of performance information until the correlation model generation unit 130 calculates the correlation function is the first embodiment of the present invention. The same as (Steps S101 to S103).
- the learning reliability model generation unit 190 generates a learning reliability model for the correlation (step S114).
- the time series model generation unit 192 of the learning reliability model generation unit 190 has a single time series of each format stored in the time series model storage unit 191 for the time series of the learning period of each metric related to the correlation. Generate a model.
- the reliability model generation unit 193 calculates a prediction error for the time series of the learning period of each generated single time series model, and generates a learning reliability model for a single time series model with a small prediction error (high fitness). To do.
- FIG. 24 is a diagram showing an example of a correlation model in the third embodiment of the present invention.
- the learning reliability model generation unit 190 generates a single time series model as shown in FIG. 12 for each of the metrics A and B in the time series of FIG. And the learning reliability model production
- the learning reliability model generation unit 190 assigns a learning reliability model to the correlation and stores it in the correlation model storage unit 150 (step S115).
- the learning reliability model generation unit 190 assigns and stores a learning reliability model to the correlation between metrics A and B as shown in FIG.
- steps S112 to S115 are repeated for all metric pairs (step S116).
- FIG. 23 is a flowchart showing details of the correlation change analysis process (step S200) in the third embodiment of the present invention.
- the correlation change analysis unit 160 acquires a correlation model to which a learning reliability model is added from the correlation model storage unit 150 (step S211).
- the learning reliability determination unit 195 selects one of the correlations included in the correlation model (step S212).
- the learning reliability determination unit 195 determines the learning reliability of the selected correlation (step S213).
- the time series model generation unit 197 of the learning reliability determination unit 195 has a single time series model of each format stored in the time series model storage unit 196 for the time series of the monitoring period of each metric related to the correlation. Is generated.
- the reliability determination unit 198 calculates a prediction error for the time series of the monitoring period of each generated single time series model.
- the reliability determination unit 198 determines the learning reliability using the prediction error of each single time series model and the learning reliability model given to the correlation.
- the reliability determination unit 198 calculates the learning reliability for the correlation between the metrics A and B using the learning reliability model in FIG.
- the reliability determination unit 198 sets “1” to the learning reliability according to the learning reliability model. Set.
- the reliability determination unit 198 sets “0.5” as the learning reliability. In cases other than the above, the reliability determination unit 198 sets “0” as the learning reliability.
- the correlation change analysis unit 160 detects correlation destruction as in the first embodiment (steps S204 to S207) of the present invention. This is performed (steps S215 to S218). Then, the processes in steps S212 to S218 are repeated for all correlations included in the correlation model (step S219).
- the failure analysis unit 180 performs failure analysis (step S220) as in the first embodiment (step S209) of the present invention.
- the learning reliability can be calculated more accurately than in the first embodiment of the present invention.
- the reason for this is that the learning reliability calculation unit 140 determines the degree of adaptation of the single time series model to the time series of the learning period and the degree of adaptation of the single time series model to the time series of the monitoring period for each of the metrics related to the correlation. This is for calculating the learning reliability based on the above.
- the correlation is represented by a correlation function between metrics, but the correlation may be represented by a correlation coefficient between metrics.
- the correlation model generation unit 130 detects a correlation for a pair of metrics whose correlation coefficient in the learning period is equal to or greater than a predetermined threshold.
- the learning reliability calculation unit 140 calculates the learning reliability of the detected correlation. Then, for example, for a correlation whose learning reliability is equal to or higher than a predetermined threshold, the correlation change analysis unit 160 determines that an abnormality has occurred in the system when the correlation coefficient in the monitoring period is less than the predetermined threshold.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- Quality & Reliability (AREA)
- Computer Hardware Design (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Debugging And Monitoring (AREA)
Abstract
Description
本発明の第1の実施の形態について説明する。
次に、本発明の第2の実施の形態について説明する。
次に、本発明の第3の実施の形態について説明する。
101 CPU
102 記憶手段
103 通信手段
104 入力手段
105 出力手段
110 性能情報収集部
120 性能情報記憶部
130 相関モデル生成部
140 学習信頼度算出部
141 時系列モデル記憶部
142 時系列モデル生成部
143 予測誤差算出部
144 信頼度算出部
150 相関モデル記憶部
160 相関変化分析部
170 分析設定記憶部
180 障害分析部
190 学習信頼度モデル生成部
191 時系列モデル記憶部
192 時系列モデル生成部
193 信頼度モデル生成部
195 学習信頼度決定部
196 時系列モデル記憶部
197 時系列モデル生成部
198 信頼度決定部
500 被監視システム
Claims (10)
- システムにおける複数のメトリックの学習期間の時系列をもとに、メトリック間の相関関係を含む相関モデルを生成する、相関モデル生成手段と、
前記相関モデルに含まれる前記相関関係に係るメトリックの各々の前記学習期間の時系列の振る舞いをもとに、当該相関関係の学習信頼度を算出する、学習信頼度算出手段と、
を備えた、システム分析装置。 - 前記学習信頼度算出手段は、前記相関関係に係るメトリックの各々の前記学習期間の時系列に対する所定形式の時系列モデルの適合度合いをもとに、当該相関関係の前記学習信頼度を算出する、
請求項1に記載のシステム分析装置。 - 前記適合度合いは、前記メトリックの前記学習期間の時系列に対して生成された前記所定形式の時系列モデルによる、当該学習期間の時系列に対する予測誤差をもとに算出される、
請求項2に記載のシステム分析装置。 - 前記学習信頼度算出手段は、前記相関関係に係るメトリックの各々について算出された、1以上の前記所定形式の時系列モデルによる予測誤差の内の、最小の予測誤差の合計値を、当該相関関係の前記学習信頼度として算出する、
請求項3に記載のシステム分析装置。 - 前記学習信頼度算出手段は、前記相関関係に係るメトリックの各々の前記学習期間の時系列を時間方向で分割し、分割された時系列の各々に対する前記所定形式の時系列モデルの適合度合いをもとに、前記学習信頼度を算出する、
請求項2に記載のシステム分析装置。 - 前記学習信頼度算出手段は、前記相関関係に係るメトリックの各々の、前記学習期間の時系列に対する前記所定形式の時系列モデルの適合度合いと、監視期間の時系列に対する当該所定形式の時系列モデルの適合度合いとをもとに、前記学習信頼度を算出する、
請求項2に記載のシステム分析装置。 - 前記所定形式の時系列モデルは、時系列が一定値を示すモデル、二つの値のいずれかを示すモデル、及び、直線的な変化を示すモデルの内のいずれかである、
請求項2乃至6のいずれかに記載のシステム分析装置。 - さらに、前記学習信頼度が所定の信頼度閾値以上の前記相関関係を用いて、当該相関関係に係るメトリック間の相関破壊を検出する、相関変化分析手段を備えた、
請求項1乃至7のいずれかに記載のシステム分析装置。 - システムにおける複数のメトリックの学習期間の時系列をもとに、メトリック間の相関関係を含む相関モデルを生成し、
前記相関モデルに含まれる前記相関関係に係るメトリックの各々の前記学習期間の時系列の振る舞いをもとに、当該相関関係の学習信頼度を算出する、
分析方法。 - コンピュータに、
システムにおける複数のメトリックの学習期間の時系列をもとに、メトリック間の相関関係を含む相関モデルを生成し、
前記相関モデルに含まれる前記相関関係に係るメトリックの各々の前記学習期間の時系列の振る舞いをもとに、当該相関関係の学習信頼度を算出する、
処理を実行させるプログラムを格納する、コンピュータが読み取り可能な記録媒体。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2016519107A JPWO2015174063A1 (ja) | 2014-05-16 | 2015-05-11 | 情報処理装置、分析方法、及び、記録媒体 |
EP15792714.6A EP3144815A4 (en) | 2014-05-16 | 2015-05-11 | Information processing device, analysis method, and recording medium |
US15/128,531 US10157113B2 (en) | 2014-05-16 | 2015-05-11 | Information processing device, analysis method, and recording medium |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2014101948 | 2014-05-16 | ||
JP2014-101948 | 2014-05-16 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2015174063A1 true WO2015174063A1 (ja) | 2015-11-19 |
Family
ID=54479613
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2015/002365 WO2015174063A1 (ja) | 2014-05-16 | 2015-05-11 | 情報処理装置、分析方法、及び、記録媒体 |
Country Status (4)
Country | Link |
---|---|
US (1) | US10157113B2 (ja) |
EP (1) | EP3144815A4 (ja) |
JP (1) | JPWO2015174063A1 (ja) |
WO (1) | WO2015174063A1 (ja) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10649874B2 (en) * | 2015-12-30 | 2020-05-12 | Teradata Us, Inc. | Long-duration time series operational analytics |
CN109145595A (zh) * | 2018-07-31 | 2019-01-04 | 顺丰科技有限公司 | 一种用户异常行为检测系统、方法、设备及存储介质 |
FR3098937B1 (fr) * | 2019-07-15 | 2021-10-08 | Bull Sas | Procédé d’analyse de consommation de ressource d’une infrastructure informatique, alerte et dimensionnement |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2011155621A1 (ja) * | 2010-06-07 | 2011-12-15 | 日本電気株式会社 | 障害検出装置、障害検出方法およびプログラム記録媒体 |
WO2013111560A1 (ja) * | 2012-01-23 | 2013-08-01 | 日本電気株式会社 | 運用管理装置、運用管理方法、及びプログラム |
Family Cites Families (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080243912A1 (en) * | 2007-03-28 | 2008-10-02 | British Telecommunctions Public Limited Company | Method of providing business intelligence |
JP4872944B2 (ja) | 2008-02-25 | 2012-02-08 | 日本電気株式会社 | 運用管理装置、運用管理システム、情報処理方法、及び運用管理プログラム |
CN102713862B (zh) * | 2010-02-15 | 2015-12-02 | 日本电气株式会社 | 故障原因提取装置、故障原因提取方法和程序记录介质 |
WO2012029500A1 (ja) | 2010-09-01 | 2012-03-08 | 日本電気株式会社 | 運用管理装置、運用管理方法、及びプログラム |
US10360527B2 (en) * | 2010-11-10 | 2019-07-23 | International Business Machines Corporation | Casual modeling of multi-dimensional hierarchical metric cubes |
JP5473977B2 (ja) * | 2011-04-14 | 2014-04-16 | キヤノン株式会社 | 撮像装置およびカメラシステム |
US9659250B2 (en) | 2011-08-31 | 2017-05-23 | Hitachi Power Solutions Co., Ltd. | Facility state monitoring method and device for same |
US9197511B2 (en) * | 2012-10-12 | 2015-11-24 | Adobe Systems Incorporated | Anomaly detection in network-site metrics using predictive modeling |
US20150046060A1 (en) * | 2013-08-12 | 2015-02-12 | Mitsubishi Electric Research Laboratories, Inc. | Method and System for Adjusting Vehicle Settings |
US20170076209A1 (en) * | 2015-09-14 | 2017-03-16 | Wellaware Holdings, Inc. | Managing Performance of Systems at Industrial Sites |
-
2015
- 2015-05-11 WO PCT/JP2015/002365 patent/WO2015174063A1/ja active Application Filing
- 2015-05-11 EP EP15792714.6A patent/EP3144815A4/en not_active Withdrawn
- 2015-05-11 JP JP2016519107A patent/JPWO2015174063A1/ja active Pending
- 2015-05-11 US US15/128,531 patent/US10157113B2/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2011155621A1 (ja) * | 2010-06-07 | 2011-12-15 | 日本電気株式会社 | 障害検出装置、障害検出方法およびプログラム記録媒体 |
WO2013111560A1 (ja) * | 2012-01-23 | 2013-08-01 | 日本電気株式会社 | 運用管理装置、運用管理方法、及びプログラム |
Non-Patent Citations (1)
Title |
---|
See also references of EP3144815A4 * |
Also Published As
Publication number | Publication date |
---|---|
US20170139794A1 (en) | 2017-05-18 |
EP3144815A4 (en) | 2018-01-17 |
JPWO2015174063A1 (ja) | 2017-04-20 |
US10157113B2 (en) | 2018-12-18 |
EP3144815A1 (en) | 2017-03-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6394726B2 (ja) | 運用管理装置、運用管理方法、及びプログラム | |
JP6354755B2 (ja) | システム分析装置、システム分析方法、及びシステム分析プログラム | |
JP5910727B2 (ja) | 運用管理装置、運用管理方法、及び、プログラム | |
JP6183450B2 (ja) | システム分析装置、及び、システム分析方法 | |
JP6658540B2 (ja) | システム分析装置、システム分析方法およびプログラム | |
JP6183449B2 (ja) | システム分析装置、及び、システム分析方法 | |
JP6708203B2 (ja) | 情報処理装置、情報処理方法、及び、プログラム | |
JP6489235B2 (ja) | システム分析方法、システム分析装置、および、プログラム | |
WO2018073960A1 (ja) | 表示方法、表示装置、および、プログラム | |
EP2958023B1 (en) | System analysis device and system analysis method | |
WO2015174063A1 (ja) | 情報処理装置、分析方法、及び、記録媒体 | |
JP6973445B2 (ja) | 表示方法、表示装置、および、プログラム |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 15792714 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2016519107 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 15128531 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
REEP | Request for entry into the european phase |
Ref document number: 2015792714 Country of ref document: EP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2015792714 Country of ref document: EP |