WO2015145500A1 - システム分析装置、分析モデル生成方法、システム分析方法およびシステム分析プログラム - Google Patents
システム分析装置、分析モデル生成方法、システム分析方法およびシステム分析プログラム Download PDFInfo
- Publication number
- WO2015145500A1 WO2015145500A1 PCT/JP2014/005336 JP2014005336W WO2015145500A1 WO 2015145500 A1 WO2015145500 A1 WO 2015145500A1 JP 2014005336 W JP2014005336 W JP 2014005336W WO 2015145500 A1 WO2015145500 A1 WO 2015145500A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- model
- data item
- regression equation
- group
- correlation model
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B17/00—Systems involving the use of models or simulators of said systems
- G05B17/02—Systems involving the use of models or simulators of said systems electric
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0221—Preprocessing measurements, e.g. data collection rate adjustment; Standardization of measurements; Time series or signal analysis, e.g. frequency analysis or wavelets; Trustworthiness of measurements; Indexes therefor; Measurements using easily measured parameters to estimate parameters difficult to measure; Virtual sensor creation; De-noising; Sensor fusion; Unconventional preprocessing inherently present in specific fault detection methods like PCA-based methods
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0243—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/31—From computer integrated manufacturing till monitoring
- G05B2219/31357—Observer based fault detection, use model
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Definitions
- the present invention relates to a system analysis apparatus, an analysis model generation method, a system analysis method, and a system analysis program for analyzing a system state.
- processing for analyzing the state of the system is performed based on sensor values obtained from system components.
- the system is, for example, a unity or mechanism composed of elements that affect each other, such as an ICT (Information and Communication Technology) system, a chemical plant, a power plant, and power equipment.
- ICT Information and Communication Technology
- the sensor values are various values obtained from system components. For example, the valve opening, the liquid level, the temperature, the flow rate, and the pressure obtained through sensors provided in the system components. Measured values of current, voltage, etc., predicted values calculated using such measured values, values of control signals issued by information processing devices used to change the system to a desired operating state, etc. .
- various values obtained from system components are simply referred to as sensor values without distinguishing their types.
- a method for determining whether the state of the device is normal or abnormal is widely used.
- the system analysis method may further include a method for identifying a malfunctioning facility when a predetermined number of sensor values exceed a threshold value.
- a method for determining whether the state of the system is normal or abnormal based on such a sensor value itself, or specifying a failure location is referred to as a first method.
- the first method has a problem that a system state change due to an internal factor or an external factor such as an air temperature, a system load, or a set value is erroneously detected as a system abnormality.
- the threshold is set to a low level to avoid such false detections, system abnormalities that should be detected, such as equipment failures and operation mistakes, cannot be detected, and serious damage may be caused to the system and its surroundings. There was a problem that caused sex.
- a regression analysis is performed on sensor values obtained from system components, and the results are based on the difference between the predicted values of other sensor values and the actual sensor values.
- a method of determining whether the state of the system is normal or abnormal by using regression analysis for such sensor values is referred to as a second method.
- the predicted value can be calculated in accordance with the state change of the system (how it has changed), so whether the system state is normal or abnormal based on the sensor value itself.
- the sensitivity of abnormality detection can be increased while suppressing erroneous detection.
- the multicollinearity phenomenon is a phenomenon that causes an incomputable problem and insufficient accuracy when a plurality of linked data items are included in an explanatory variable in regression analysis.
- the “data item” is associated with various types of sensor values obtained from the system components, and represents a set of sensor values corresponding to the corresponding item, or the corresponding item. It is used to represent identification information that identifies the corresponding sensor value. Elements such as devices constituting the system operate in conjunction with other elements in order to achieve system purposes such as information processing, production of chemical products, and energy conversion. Therefore, it is conceivable that the sensor values obtained from the components of the system also change in conjunction with other sensor values. When regression analysis is used for such sensor values, the above-described multicollinearity phenomenon is likely to occur.
- Patent Document 1 uses PLS (Partial Least Square) as an example of a method for constructing a regression equation for predicting manufacturing quality.
- PLS Partial Least Square
- Patent Document 2 describes a method for determining parameters of a regression equation for detecting plant abnormality using PLS as in Patent Document 1.
- Patent Document 3 as an example of a method for constructing a regression equation for detecting an abnormality in a heat pump heat exchanger, a data item having a strong correlation is excluded from explanatory variables included in one regression equation. Describes a method for constructing a regression equation by avoiding the multicollinearity phenomenon.
- Patent Document 4 data items are classified into three types: objective variables, explanatory variables and independence, or explanatory variables and collinearity, and one regression equation is classified into explanatory variables and collinearity.
- a method is described in which a regression equation is constructed by avoiding the multicollinearity phenomenon by allowing only one data item to be entered.
- the second method for determining whether the system state is normal or abnormal using regression analysis compared to the first method for determining whether the system state is normal or abnormal based on the sensor value itself, There is a high possibility that the sensitivity of abnormality detection can be increased while suppressing false detection.
- each sensor value obtained from the system components often changes in conjunction with other sensor values, and there is a high possibility that a multicollinearity phenomenon occurs. Therefore, in the case of using the second method, a change in the sensor value due to a system state change that can normally occur and a change in the sensor value unrelated to such a change in the sensor value can be well separated. It is necessary to construct a regression equation that can avoid the multicollinearity phenomenon while using a plurality of data items corresponding to the types of sensor values that are highly related to each other.
- Patent Documents 1 and 2 use a plurality of data items as explanatory variables, and construct a regression equation using latent variables and objective variables obtained by synthesizing the explanatory variables.
- simply using a plurality of data items as explanatory variables may include data items that hardly contribute to the predicted value of the target variable calculated by the regression equation.
- an abnormality that affects a small number of data items (more specifically, a change in sensor value associated with the abnormality) cannot be detected with high sensitivity.
- Patent Document 3 has a problem that an abnormality appearing in the excluded data item cannot be detected because the data item having a strong correlation is excluded.
- Patent Document 4 since the method described in Patent Document 4 divides data items into objective variables and explanatory variables in advance, depending on which data item is selected for the objective variable, whether the system state is normal or abnormal There was a problem that the judgment results were different. For example, even if the influence of an abnormality appears as a difference in sensor values between data items that are target variables or between data items that are explanatory variables, between the data items that are target variables and the data items that are explanatory variables If the sensor value does not appear as a difference in sensor value, the abnormality cannot be detected, and the system state is determined to be normal even though some abnormality has occurred in the system.
- the collinearity between explanatory variables is due to changes in sensor values that are normally caused by system state changes, and changes in sensor values due to abnormalities appear only between explanatory variables that are collinear between explanatory variables. In such a case, such a change in sensor value may not be detected as abnormal.
- the explanatory variable of the regression equation includes only one data item classified as explanatory variable and collinear, the sensor accompanying the system state change that can normally occur from the actual change of the sensor value. This is because a change in sensor value that is unrelated to a change in the state of the system (a change in sensor value that may be abnormal) cannot be extracted by removing the change in value.
- the explanatory variable And at least two data items classified as collinear.
- Patent Documents 1 to 4 have a problem that it becomes difficult to select appropriate objective variables and explanatory variables when the system becomes large and complicated.
- changes in individual internal or external factors such as changes in temperature, system load, and changes in setting values affect changes in sensor values that indicate the state of the system. This is because it is difficult to grasp the range, and it is difficult to narrow down the data items to be set in the objective variable and the explanatory variable.
- the present invention provides a system analysis apparatus, an analysis model generation method, a system analysis method, and a system analysis apparatus that can accurately analyze the state of the target system even when the target system is complex or multicollinearity exists between data items.
- the purpose is to provide a system analysis program.
- being able to analyze the state of the system means that it is possible to determine at least whether the state of the system is abnormal or normal.
- a system analysis apparatus includes an analysis model generation unit that generates an analysis model for analyzing a state of a target system, using state information that is a collection of information on a plurality of types of data items of the target system.
- the model includes at least one multi-body correlation model, which is a correlation model having at least a regression equation configured using three or more data items and an allowable range of prediction error of the regression equation, and an analysis model generating means Is a data item classification means for classifying the data item group included in the state information into one or more groups, and for each group classified by the data item classification means, at least one of the data items included in the group Select two representative data items, and any two of the data items included in the group excluding the selected representative data item For all combinations for the eye, a regression equation is constructed using the two data items and the representative data item, and the allowable range of prediction error of the regression equation and the superiority of the regression equation are set.
- a many-body correlation model generating means for generating a many-body correlation model having at least a regression equation calculated and constructed, and an allowable range of a prediction error of the regression equation, and a many-body correlation generated by the many-body correlation model generating means
- a model extracting means for extracting a multi-body correlation model satisfying an excellent model condition of a multi-body correlation model for which a regression equation has a predetermined level of excellence from the model group as a multi-body correlation model to be included in the analysis model;
- the data item classifying means includes one data item arbitrarily selected from the data items included in the group in at least one group after classification.
- the first degree of the regression equation constructed using the data item and the second data item which is one of the data items included in the same group as the first data item excluding the first data item, Classifying the data items such that at least one of the goodnesses of the regression equation satisfies a predetermined good model condition when calculated for all possible combinations of the second data items with respect to the data items; To do.
- An analysis model generation method is an analysis model generation method for generating an analysis model for analyzing a state of a target system using state information that is a set of information related to a plurality of types of data items of the target system.
- the data item classification means is a first data item that is one data item arbitrarily selected from the data items included in the group in at least one group after the classification of the data item group included in the state information.
- the first data is the degree of excellence of the regression equation constructed using the item and the second data item that is one of the data items included in the same group as the first data item excluding the first data item.
- At least one of the goodnesses of the regression formula is a predetermined goodness model condition.
- the data is classified into one or more groups so as to satisfy, and the many-body correlation model generation means uses at least one representative data from the data items included in the group for each classified group using the state information.
- An item is selected, and all the combinations for any two data items included in the group excluding the selected representative data item are configured using the two data items and the representative data item.
- the regression equation is calculated, the tolerance of the prediction error of the regression equation and the superiority of the regression equation are calculated, and the constructed regression equation and the tolerance of the prediction error of the regression equation are at least included
- a many-body correlation model is generated, and the model extraction means is a superior model of the many-body correlation model in which the degree of regression is determined in advance from the generated many-body correlation model group
- the many-body correlation model that satisfies matter, and extracts as a multi-body correlation model to be included in the analysis model.
- the data item classifying means includes, in at least one group after classification, the data item group included in the state information that is a set of information related to a plurality of types of data items of the target system.
- a first data item that is one data item arbitrarily selected from the data items included in the group, and one of the data items that are included in the same group as the first data item excluding the first data item.
- the superiority of the regression equation constructed using the second data item is calculated for all combinations of second data items that can be taken for the first data item, at least one of the superiorities of the regression equation is Categorized into one or more groups so as to satisfy a predetermined excellent model condition, and the many-body correlation model generation means uses the state information to classify the classified groups. For each of the data items included in the group, at least one representative data item is selected, and all of the data items included in the group except for the selected representative data item are all selected.
- the model extraction means includes, in the analysis model, a multi-body correlation model that satisfies the excellent model condition of the many-body correlation model for which the degree of superiority of the regression equation is predetermined from the generated many-body correlation model group. It is extracted as a body correlation model, the information of the analysis model including the extracted many-body correlation model group is stored in a predetermined storage device, and the model destruction detection means is newly collected when the state information is newly acquired.
- the predicted value of the objective variable of the regression equation of the correlation model is the correlation model
- the presence or absence of model destruction, which is a phenomenon exceeding the allowable range of the prediction error of the regression equation, is detected, and the abnormality determination means determines that the system status is based on the detection result by the model destruction detection means. And judging whether normal or normal.
- the system analysis program includes, in a computer, a data item group included in state information, which is a set of information related to a plurality of types of data items of the target system, in at least one group after classification.
- a first data item that is one data item arbitrarily selected from the data items to be selected, and second data that is one of the data items included in the same group as the first data item excluding the first data item
- the superiority of the regression equation constructed using the items is calculated for all combinations of the second data items that can be taken for the first data item, at least one of the superiorities of the regression equation is a predetermined value. For each group that has been classified using the data item classification process and status information that are classified into one or more groups to satisfy the excellent model condition.
- At least one representative data item is selected from the data items included in the group, and all combinations of any two data items included in the group excluding the selected representative data item Construct a regression equation composed of the two data items and a representative data item, and calculate the tolerance of the prediction error of the regression equation and the superiority of the regression equation.
- a multi-body correlation model generation process for generating a multi-body correlation model having at least the regression equation and an allowable range of prediction error of the regression equation, A model extraction process that extracts a many-body correlation model that satisfies the pre-determined excellent model of the many-body correlation model as a many-body correlation model to be included in the analysis model.
- the predicted value of the objective variable of the regression equation of the correlation model is a phenomenon that exceeds the allowable range of the prediction error of the regression equation of the correlation model.
- the state of the target system can be analyzed with high accuracy even if the target system is complex or there is multicollinearity between data items.
- FIG. 1211 It is a block diagram showing an example of composition of system analysis device 100 of a 1st embodiment. It is explanatory drawing which shows an example of the extraction method of the excellent cross correlation model by the data item classification means 1211.
- FIG. It is explanatory drawing which shows an example of the extraction method of the excellent many-body correlation model by the model extraction means.
- It is explanatory drawing which shows the construction example of the graph structure of the excellent cross correlation model group.
- Embodiment 1 FIG.
- the system analysis apparatus of the present invention is applied to abnormality detection of a power plant system.
- FIG. 1 is a block diagram illustrating a configuration example of the system analysis apparatus 100 according to the first embodiment.
- the system analysis device 100 of this embodiment is connected to a monitored system including one or more monitored devices 200.
- the monitored device 200 is a device or subsystem as a component constituting the power plant system, and is, for example, a turbine, a feed water heater, a condenser, or the like.
- the monitored device 200 also includes elements that connect the devices, such as piping and signal lines.
- the entire system is assumed as a monitored system like a power plant system, but the monitored system may be a part of a certain system. That is, you may limit a to-be-monitored system to the component for implement
- Each monitored device 200 measures a sensor value obtained by the monitored device 200 for each predetermined period and transmits the measured sensor value to the system analysis device 100.
- the sensor value is a value obtained from the sensor. Examples of sensor values include measured values measured by measuring devices installed in equipment such as valve opening, liquid level height, temperature, flow rate, pressure, current, and voltage, and predictions calculated from the measured values. Value and control signal value.
- each sensor value is represented by a numerical value such as an integer or a decimal.
- one data item is assigned to each sensor corresponding to the sensor value obtained from each monitored device 200.
- a set of sensor values collected at the same timing from each monitored device 200 is called state information, and a set of data items corresponding to the sensor values included in the state information is called a data item group.
- the items collected at the timing considered to be the same are not limited to those collected by the monitored devices 200 at the same time and transmitted to the system analysis device 100, but the time difference within a predetermined range. Measured and transmitted to the system analysis device 100, and those collected from each monitored device 200 by the system analysis device 100 through a series of collection processes.
- a device that stores sensor values acquired by the monitored device 200 between the monitored device 200 and the system analysis device 100 such as a data server, a DCS (Distributed Control System), or the like.
- a process computer may be provided. In such a case, even if the monitored device 200 acquires the sensor value at an arbitrary timing and stores it in the storage device, the system analysis device 100 reads the sensor value stored in the storage device every predetermined period. Good.
- the system analysis apparatus 100 constructs an analysis model for analyzing the state of the monitored system while acquiring the state information of the monitored system at regular intervals, and uses the constructed analysis model to state the state of the monitored system which includes state information collection means 11, analysis model generation means 12, analysis means 13, state information storage means 14, and analysis model storage means 15.
- Status information collection means 11 collects status information of the monitored system at regular intervals.
- the time series data of the state information may be referred to as state series information.
- the state information storage unit 14 stores the state information collected by the state information collection unit 11 in time series. In other words, the state information storage unit 14 stores the state information collected by the state information collection unit 11 as state series information.
- the state information storage unit 14 may store information indicating the collection time and state information (more specifically, a collection of collected sensor values) in association with each other.
- the state information may be, for example, information in which sensor values to be collected are arranged in a predetermined order.
- the state information storage unit 14 of this embodiment has a storage area for storing state series information for at least a predetermined period.
- the analysis model generation unit 12 generates an analysis model for analyzing the state of the monitored system based on the state series information for a predetermined period stored in the state information storage unit 14.
- the analysis model of the present embodiment includes one or more multi-body correlation models, which are correlation models having a regression equation configured using three or more data items and an allowable range of prediction error of the regression equation. More specifically, the analysis model of the present embodiment is a set of correlation models composed of one or more many-body correlation models.
- a model having at least a regression equation that defines a relationship between data items and an allowable range of a prediction error of the regression equation is called a correlation model, and a regression configured using two of the data items.
- a correlation model including an equation is referred to as a “cross-correlation model”, and a correlation model including a regression equation configured using three or more data items is referred to as a “multi-body correlation model”.
- a correlation model including a regression equation configured using three or more data items is referred to as a “multi-body correlation model”.
- An arbitrary period that does not include a failure is set by the operator as the predetermined period of the state series information used for generating the analysis model. From the viewpoint of increasing the sensitivity of abnormality detection, it is preferable to set the period as short as possible so as not to be affected by the secular change of the monitored system. For example, if the maintenance cycle of the monitored system is one year, the predetermined period may be a period that is sufficiently shorter than that, such as one month, one week, or one day. At this time, it is preferable that the period includes an influence on the sensor value due to the factor that causes the largest change in the sensor values among the factors of the system state change that can normally occur.
- the predetermined period may be a period such as 9 months or 1 year.
- the analysis model storage unit 15 stores analysis model information that is information on the analysis model generated by the analysis model generation unit 12.
- the analysis model information may be, for example, a set of model information for each of correlation models included in the analysis model (one or more multi-body correlation models in the present embodiment).
- Model information includes, for example, an identifier of the correlation model, regression information of the correlation model (information of data items used as objective variables, information of data items used as explanatory variables, information of relational expressions between data items, etc. ) And information indicating an allowable range of the prediction error of the regression equation.
- the model information may further include a degree of quality that is an index representing the quality of the regression equation.
- the analysis unit 13 analyzes the state of the monitored system using the analysis model generated by the analysis model generation unit 12. As shown in FIG. 1, the analysis unit 13 includes a model destruction detection unit 131 and an abnormality determination unit 132.
- the model destruction detection unit 131 predicts the regression equation of the correlation model for each correlation model included in the analysis model indicated by the analysis model information stored in the analysis model storage unit 15. Detects whether model destruction, a phenomenon in which the error exceeds the allowable range, has occurred.
- the occurrence of such model destruction is referred to as a change in the sensor value that accompanies a change in the state of the system that can occur normally (hereinafter referred to simply as a normal change). It is used as an index indicating that a sensor value change unrelated to the change is included.
- the model destruction detection unit 131 uses, for example, the correlation model for each correlation model included in the analysis model using the state series information including the latest state information stored in the state information storage unit 14 and the analysis model. Calculate the predicted value at least the latest time for the objective variable of the regression equation, calculate the prediction error that is the difference between the calculated predicted value of the objective variable and the actual sensor value, and allow the calculated prediction error The presence or absence of model destruction is detected by determining whether or not the range is satisfied. Further, the model destruction detection unit 131 outputs the model destruction detection result as model destruction information.
- the model destruction information may be information indicating the model destruction status of each correlation model included in the analysis model, for example.
- the model destruction information may be information that can identify a correlation model in which model destruction has occurred, or information that can identify a correlation model in which model destruction has not occurred.
- the model destruction information includes information obtained from the correlation model in addition to or in place of information that can specify the correlation model (for example, information on data items included in the regression equation of the model, calculated prediction error, etc. ) May be included.
- the information necessary for model destruction information depends on the alarm condition that is the condition for issuing an alarm.
- the model destruction information is information that can specify the number of correlation models in which model destruction has occurred (for example, each of the models included in the analysis model). The presence or absence of model destruction of the correlation model, a set of identifiers of the correlation model in which the model destruction has occurred, and the like.
- the model destruction information includes information indicating the superiority of the regression equation of the correlation model in which model destruction has occurred. It is only necessary to include information (such as a set of identifiers of correlation models in which model destruction has occurred) that can specify the superiority of the regression equation of the correlation model in which model destruction has occurred.
- the abnormality determination unit 132 determines whether the model destruction status of the analysis model satisfies the alarm condition based on the model destruction information obtained from the model destruction detection unit 131. Moreover, if the alarm condition is satisfied as a result of the determination, the abnormality determination unit 132 determines that the state of the monitored system is abnormal, and notifies the operator and the monitored system of the determination result.
- equation (1) can be used as the alarm condition.
- Na is the number of correlation models in which model destruction has occurred among the correlation models included in the analysis model.
- the analysis model generation means 12 of this embodiment includes an analysis model candidate generation means 121 and a model extraction means 122 as shown in FIG.
- the analysis model candidate generation unit 121 includes a data item classification unit 1211 and a many-body correlation model generation unit 1212.
- the analysis model candidate generation unit 121 generates correlation model candidates used (included) in the analysis model.
- the data item classification means 1211 classifies the data item group into one or more groups, and the many-body correlation model generation means 1212 uses one or more many-body correlation models based on the classified groups. Is generated.
- the data item classification means 1211 classifies the data item group included in the state information into one or more groups. More specifically, the data item classification means 1211 includes, in at least one group after classification, a first data item that is one data item arbitrarily selected from the data items included in the group, and the first data item The degree of excellence of the regression equation configured using the second data item that is one of the data items excluding the first data item among the data items included in the same group as the data item was arbitrarily selected When the calculation is performed for all combinations of the second data items with respect to the first data item, the data items are classified so that at least one of the superiorities of the regression equations satisfies a predetermined excellent model condition.
- the data item classification unit 1211 first uses the state series information for a predetermined period, and first sets the first provisional cross-correlation model for all combinations of two arbitrary data items in the data item group. And a second provisional cross-correlation model.
- each of the first tentative cross-correlation model and the second tentative cross-correlation model has at least a regression equation configured using the two selected data items, and a superiority of the regression equation. It is a model including.
- the objective variable data item (described later) and the explanatory variable data item (described later) of the regression formula are replaced with those of the first temporary cross-correlation model.
- t represents an index of state information.
- the state information index t is given a sequential integer whose value increases from the oldest to the newest.
- N, K, and M are arbitrary integers.
- Y (t) represents the value of the data item y when the index of the state information is t. Therefore, y (t ⁇ N) represents a past value corresponding to N sensor value collection intervals compared to y (t).
- “...” In the expression (2) indicates that a term obtained by increasing the value subtracted from t by 1 from the left to the right is omitted.
- y (t) with a hat on the left side of the first equal sign corresponds to the objective variable.
- the objective variable is a predicted value of the data item y when the status information index is t.
- y (t-1) to y (tN) and u (tK) to u (tKM) on the right side of the second equal sign correspond to explanatory variables.
- the u sensor value is used.
- the data item y used for the objective variable is defined as “object variable data item”
- the data item u used only for the explanatory variable is defined as “explanatory variable data item”.
- f (u, y) between the first equal sign and the second equal sign indicates that the regression equation is a function using two data items u and y.
- Equation (2) The parameters a 1 to a N , b 0 to b M , c, K, N, and M of Equation (2) are determined so that the degree of superiority F of the regression equation is maximized.
- the prediction accuracy of the regression equation as shown in the following equation (3) can be used.
- Equation (3) y with a bar represents the average value of the objective variable during a predetermined period (between the state information index is 1 to N1) included in the state series information.
- the superiority of the regression equation has two viewpoints: high prediction accuracy and low generalization error.
- high prediction accuracy is an example of the superiority determined from the viewpoint of high prediction accuracy
- an information criterion may be used. Note that the superiority of the regression equation may be a combination of these.
- the parameters that maximize the degree of superiority F are determined as a 1 to a N , b 0 to b M , and c so that the degree of superiority F is maximized for the set of K, N, and M. It is determined by selecting a set of K, N, and M that has the highest degree of superiority F. Specifically, the data item classification means 1211 first sets the maximum values of K, N, and M, respectively, formulates a regression equation for each combination of K, N, and M values. On the other hand, the parameters a 1 to a N , b 0 to b M , and c are determined using the least square method so that the numerator of the second term of the above formula (3) is minimized.
- the data item classifying means 1211 calculates the superiority F for each regression equation, and the regression equation parameters a 1 to a N , b 0 to b M , c, K, having the maximum superiority F are obtained. Select N, M. In this way, a desired regression equation can be obtained.
- the operator may set an arbitrary value as the maximum value of K, N, and M.
- the data item classification unit 1211 includes a first temporary cross-correlation model group that is a set of the first temporary cross-correlation models generated as described above and a second temporary cross-correlation model set that is the second temporary cross-correlation model.
- the data item classifying means 1211 determines, for each set of two data items, the one having the higher F of the regression equation among the first temporary cross-correlation model and the second temporary cross-correlation model. Select as a correlation model and obtain a set of cross-correlation models as a set of all sets of cross-correlation models for any two data items so selected.
- the data item classification unit 1211 selects a cross-correlation model that satisfies a predetermined good model condition in the regression formula of each cross-correlation model from the cross-correlation model group thus obtained. And obtain a group of excellent cross-correlation models that is a set of the extracted excellent cross-correlation models.
- the operator can set an arbitrary value for the threshold value Fth for the superiority F of the regression equation.
- the threshold value Fth is preferably set low.
- the threshold F th is preferably set high.
- the threshold value F th is preferably set high.
- the threshold value F th is preferably set high. .
- the threshold value F th is preferably a value from 0.7 to 1 in order to reduce false detections or accurately detect signs of failure. A value of .8 to 1 is more preferred.
- the data item classification means 1211 constructs a graph structure of the extracted excellent cross-correlation model group.
- the graph structure is represented as a network diagram with the data items included in the regression equation of the cross-correlation model as nodes and the regression equation as a line.
- a graph structure in which three or more nodes are connected via a line is called a cluster.
- the data item classification unit 1211 classifies the data items based on the graph structure of the excellent cross-correlation model group so that each of the clusters in the graph structure corresponds to a data item group that is a group of one data item. .
- nodal points in the graph structure correspond to data items, and therefore, data items corresponding to nodal points in each cluster become data items included in each data item group.
- data items not included in the cluster are excluded from the analysis model construction target.
- the first data item that is one data item arbitrarily selected from the data items included in the classified data item group, and the same data item group as the first data item
- the combinations with all the second data items are calculated, the data items are classified so that at least one of the superiorities of the regression equation satisfies the above-described excellent model condition.
- the many-body correlation model generation unit 1212 selects at least one representative data item from the data items included in the group for each group of data items classified by the data item classification unit 1211 and is included in the group. For all combinations of any two data items excluding the representative data item among the data items, a regression equation including the two data items and the representative data item is constructed. In addition, the many-body correlation model generation unit 1212 calculates the tolerance of the prediction error of the constructed regression equation and the superiority of the regression equation, and calculates the constructed regression equation and the tolerance of the prediction error of the regression equation. A multi-body correlation model having at least is generated. Note that the many-body correlation model generation unit 1212 may generate a many-body correlation model including a regression equation, an allowable range of prediction error of the regression equation, and a superiority of the regression equation.
- the many-body correlation model generation unit 1212 uses, for each data item group classified by the data item classification unit 1211, a regression equation configured by using a set of two arbitrary data items in the data item group. Representative data items are selected based on the superiority of the regression equation calculated for all. For example, for each data item group, the many-body correlation model generation unit 1212 calculates the superiority of the regression formula calculated for all of the regression formulas that are configured using a set of two arbitrary data items in the data item group. Then, a statistical value for each data item in the data item group may be calculated, and a representative data item in the data item group may be selected based on the calculated statistical value for each data item.
- the many-body correlation model generation unit 1212 may select, for example, the one having the highest statistical value as the representative data item. Note that the number of representative data items is not necessarily one. In the case of selecting a plurality of representative data items, the many-body correlation model generation unit 1212 may select representative data items from a higher statistical value.
- the many-body correlation model generation unit 1212 accumulates, for each data item, the degree of superiority of the regression equation of each cross-correlation model included in the cross-correlation model group generated by the data item classification unit 1211. For each of the data items belonging to the data item group, the data item having the highest cumulative value of the regression equation may be used as the representative data item.
- the superiority of the regression equation of each cross-correlation model included in the cross-correlation model group is associated with the data item used for the objective variable and the data item used for the explanatory variable.
- the degree of excellence of the regression equation is added to both data items.
- the cumulative value of the superiority of the regression equation corresponds to the statistical value.
- the regression equation shown in the following equation (5) can be used as the regression equation of the many-body correlation model.
- K, N, M, L, P, Q, and S are arbitrary integers.
- y (t) with a hat on the left side of the first equal sign corresponds to the objective variable.
- the objective variable is a predicted value of the data item y when the status information index is t.
- (TLQ) and w (tP) to w (tPS) correspond to explanatory variables.
- the sensor value of u and the index of the data item x when the status information index is tP to tPQ, and the data value of the data item w when the index of the data item x and the status information index is tR to tRS Sensor values are used.
- the data item y used for the objective variable is defined as “object variable data item”
- the data items u, x, w is defined as “an explanatory variable data item”.
- two data items (more specifically, the data item x and the data item w) among the four explanatory variable data items correspond to the “representative data item”.
- the multi-body correlation model generation unit 1212 constructs a regression equation using arbitrary two data items and a representative data item
- the multi-body correlation model generation unit 1212 calculates a regression equation using each of the two data items that are not representative data items as objective variables.
- a multi-body corresponding one regression equation to any two data items excluding the representative data item A correlation model group is generated.
- the parameters a 1 to a N , b 0 to b M , c, d 0 to d Q , e 0 to e S , K, L, N, M, P, Q, and S in Expression (5) are, for example, , PLS regression is used to determine the maximum degree F of the regression equation.
- PLS regression is used to determine the maximum degree F of the regression equation.
- the superiority F of the regression equation for example, a value of the prediction accuracy of the regression equation as shown in the above equation (3) can be used.
- the parameters with the highest degree of superiority F are, for example, a 1 to a N , b 0 to so that the degree of superiority F is maximized with respect to a set of K, L, N, M, P, Q, and S.
- b M , c, d 0 to d Q , e 0 to e S are determined, and then a set of K, L, N, M, P, Q, and S with the highest degree of superiority F is selected. Is done.
- the many-body correlation model generation unit 1212 sets, for example, the maximum values of K, L, N, M, P, Q, and S, and sets K, L, N, M, P, Q, and S, respectively.
- Regression formulas are formulated for each combination of values, and for each regression formula, parameters a 1 to a N , b 0 to b M , c, d 0 to d Q , e 0 to e Determine S.
- the maximum value of the prediction error for the state series information used for generating the analysis model is from decrease to before increase. The number of components may be used.
- the many-body correlation model generation unit 1212 calculates the superiority F for each regression equation, and the regression equation parameters a 1 to a N , b 0 to b M , c, having the maximum superiority F are obtained.
- d 0 to d Q , e 0 to e S , K, L, N, M, P, Q, S are selected.
- the operator may set an arbitrary value as the maximum value of K, L, N, M, P, Q, and S.
- the many-body correlation model group obtained in this way is a candidate for a correlation model used as an analysis model.
- a range in which both the following formulas (6) and (7) are satisfied can be used as the allowable range of the prediction error of the regression equation of the many-body correlation model.
- T 1 is the upper threshold for the prediction error.
- T 2 is a lower threshold for the prediction error.
- R i is a prediction error at the i-th time of the objective variable calculated using the regression equation of the multi-body correlation model.
- i represents a time identifier in the state series information. More specifically, i is a sequence number when status information from a certain reference time to the corresponding time is arranged in ascending or descending order. For example, it may represent the i-th time from the start of monitoring.
- the allowable range of the prediction error of the regression equation of the many-body correlation model includes a prediction error calculated over the period of the state series information used to construct the regression equation (hereinafter referred to as a prediction error during the calculation period). .))
- a prediction error during the calculation period a prediction error calculated over the period of the state series information used to construct the regression equation.
- T 1 max ⁇
- T 2 ⁇ T 1 (9)
- max ⁇ is a function that outputs the maximum value from the input numerical values.
- is an operator for extracting the absolute value of x.
- N1 is the number of prediction errors calculated from the state series information using the regression equation of the many-body correlation model.
- the average value of the prediction error is calculated using the standard deviation.
- the value obtained by adding three times the T 1 a value obtained by subtracting three times the standard deviation from the average value of prediction errors may be T 2.
- the model extraction unit 122 extracts a correlation model used for the analysis model from the correlation model candidates generated by the analysis model candidate generation unit 121.
- the model extraction unit 122 selects each correlation model from the correlation model group generated by the analysis model candidate generation unit 121 (in this embodiment, the many-body correlation model group generated by the many-body correlation model generation unit 1212).
- One or more many-body correlation models are extracted based on the superiority of the regression equation.
- the model extraction unit 122 extracts, for example, a multi-body correlation model that satisfies a predetermined excellent model condition as an excellent multi-body correlation model from the many-body correlation model group generated by the many-body correlation model generation unit 1212 and extracted.
- An excellent many-body correlation model group that is a set of excellent many-body correlation models may be used as the analysis model.
- the model extraction unit 122 stores analysis model information including model information of each excellent multi-body correlation model extracted in the analysis model storage unit 15.
- the above equation (4) can be used.
- the operator can set an arbitrary value for the threshold value Fth for the superiority F of the regression equation. From the viewpoint of expanding the target range of abnormality detection, it is preferable to set the threshold value Fth low. From the viewpoint of reducing erroneous detection due to a system state change or the like, the threshold F th is preferably set high.
- the threshold F th is preferably a value from 0.7 to 1. A value of 0.8 to 1 is more preferred.
- the superiority of the same regression equation may be used.
- both the cross-correlation model and the many-body correlation model use the prediction accuracy as the superiority F of the regression equation
- the excellent model condition for the cross-correlation model is “F> 0.6”
- the excellent model condition for the many-body correlation model is For example, “F> 0.7”.
- the index of superiority and the excellentness of the same regression equation are used regardless of the number of representative data items included in the multi-body correlation model group to be extracted.
- the model condition is used, an index of superiority and an excellent model condition of different regression equations may be used according to the number of representative data items included in the many-body correlation model group to be extracted.
- the prediction accuracy is used as the F of the regression equation of the many-body correlation model regardless of the number of representative data items
- the excellent model condition for the many-body correlation model is “F> 0.6 ⁇ (number of representative data items). -1/3 "and the like.
- FIG. 2 is an explanatory view showing an example of a method for extracting an excellent cross-correlation model by the data item classification means 1211.
- reference numeral 701A represents a cross-correlation model group from which the excellent cross-correlation model is extracted.
- Reference numeral 701B represents the extracted excellent cross-correlation model group.
- Reference numeral 701 ⁇ / b> C represents an excellent model condition used for extracting the excellent cross-correlation model of this example.
- the “Item 1” and “Item 2” columns in the upper and lower tables of FIG. 2 represent the data items of the explanatory variables used in the regression equation of the cross correlation model.
- the “Regulation equation” column represents a regression equation of the cross-correlation model.
- each cross-correlation model included in the cross-correlation model group may include an allowable range of the prediction error of the regression equation, but the allowable range of the prediction error of the regression equation is not used for extracting the excellent cross-correlation model. This is omitted in the figure.
- values that are not used for explanation may be omitted from the drawings.
- the data item classification unit 1211 acquires state series information from the state information storage unit 14 and generates a cross-correlation model group 701A.
- the cross-correlation model group 701A of this example is composed of five cross-correlation models.
- the data item group included in the state series information used to generate the cross-correlation model group 701A of this example includes data item A, data item B, data item C, data item D, data item G, and data item H.
- f j () is a function for calculating the predicted value of the objective variable data item from the data items contained in parentheses (where j is an integer that is an identifier for identifying the regression equation).
- the explanatory variable of the function includes not only the value (sensor value) of the data item included in the state series information stored in the state information storage unit 14 but also the data item included in the parenthesis. It is also possible to use a conversion value calculated by using the past value.
- the data item classification unit 1211 extracts a good cross-correlation model that satisfies the good model condition 701C from the cross-correlation model group 701A, and obtains a good cross-correlation model group 701B.
- the excellent model condition 701C of this example is “regression excellence F> 0.6” as shown in FIG. For this reason, the data item classification unit 1211 extracts a cross-correlation model having a regression formula with a superiority F exceeding 0.6 among the five cross-correlation models constituting the cross-correlation model group 701A as an excellent cross-correlation model.
- FIG. 3 is an explanatory diagram showing an example of a method for extracting an excellent many-body correlation model by the model extracting means 122.
- reference numeral 702A represents a many-body correlation model group from which the excellent many-body correlation model is extracted.
- Reference numeral 702B represents the extracted excellent multi-body correlation model group.
- Reference numeral 702C represents an excellent model condition used for extracting the excellent many-body correlation model of this example.
- the “Item 1” and “Item 2” columns in the upper and lower tables of FIG. 3 represent data items excluding the representative data items among the data items of the explanatory variables used in the regression equation of the many-body correlation model. ing.
- the “Representative Item” column in the table represents representative data items used in the regression equation of the many-body correlation model.
- the model extraction unit 122 acquires the many-body correlation model group 702A from the many-body correlation model generation unit 1212.
- the many-body correlation model group 702A of this example is composed of two many-body correlation models.
- the data item group included in the state series information used for generating the many-body correlation model group 702A of this example includes a data item A, a data item B, a data item C, and a data item D.
- the model extraction unit 122 extracts a superior many-body correlation model that satisfies the excellent model condition 702C from the many-body correlation model group 702A, and obtains an excellent many-body correlation model group 702B.
- the excellent model condition 702C of this example is “regression excellence F> 0.6” as shown in FIG.
- the model extracting means 122 uses, as the excellent many-body correlation model, the many-body correlation model in which the degree of superiority F of the regression equation exceeds 0.6 among the two many-body correlation models constituting the many-body correlation model group 702A. Extract.
- the graph structure of the cross-correlation model group and the many-body correlation model group in this embodiment will be described using a specific example.
- the graph structure is constructed as an example of a directed graph in which an arrow is attached from the explanatory variable data item used in the regression equation of the correlation model included in the target model group to the target variable data item.
- the graph structure may be constructed as an undirected graph in which no arrow is attached to the line between the nodes.
- FIG. 4 is an explanatory diagram showing a construction example of the graph structure of the excellent cross-correlation model group 701B shown in FIG.
- the data item classification unit 1211 converts the data items used in the regression equation of each excellent cross-correlation model included in the excellent cross-correlation model group 701 ⁇ / b> B, which is the target model group, to each nodule. Represented by a point.
- the nodes, for each regression equation there is an arrow pointing from the explanatory variable data item (data item used only for the explanatory variable) to the objective variable data item (data item used for the objective variable).
- the graph structure 701D is constructed by connecting with lines.
- the data item classification means 1211 represents the data item A, the data item B, the data item C, and the data item D as nodal points, and an arrow between the nodal points from the data item A to the data item B is provided.
- a line with an arrow from data item A to data item C, a line with an arrow from data item D to data item B, and a line with an arrow from data item D to data item C, and Structure 701D is obtained.
- FIG. 4 shows an example in which one cluster remains, but a plurality of clusters may remain. In this example, since there is no mass that only two nodes are connected via a line, any node is valid. For example, only two nodes are connected via a line. If there is a mass, the mass is not recognized as a cluster, and the data item corresponding to the nodal point included therein is excluded from the analysis model construction target.
- FIG. 5 is an explanatory diagram showing an example of a method for selecting representative data items.
- the many-body correlation model generation unit 1212 generates a regression equation for each data item corresponding to the nodal point in the cluster.
- the cumulative value of the degree of quality F is calculated. Specifically, for each cluster included in the graph structure 701D, for each data item corresponding to the node included in the cluster, the prominence F of the regression equation corresponding to the arrow connected to the node is accumulated. , Get the cumulative value.
- the obtained cumulative value is set as a score 701E of the data item.
- the data item having the maximum score 701E among the data items is set as the representative data item.
- a representative data item may be selected at random.
- the score 701E of each data item is 1.6, 1.7, 1.5 and 1.6 for data item A, data item B, data item C and data item D, respectively. It is. Therefore, data item B is selected as the representative data item.
- the data item B is selected as the representative data item of the data item group including the data item A, the data item B, the data item C, and the data item D.
- This example is an example in the case where there is only one cluster in the graph structure 701D of the excellent many-body correlation model group. However, when there are a plurality of clusters, the representative data item is obtained for each cluster by the same method. Is elected. According to the method of this example, for example, since the score can be calculated using the prediction accuracy as the superiority F of the regression equation as the weight of each regression equation, other data items can be predicted with the highest accuracy. Is selected as the representative data item. From another point of view, data items that can predict other data items with the highest accuracy behave in an average manner in data items that belong to the same cluster, so that other data items in the same cluster can be predicted with the highest accuracy. It is considered possible.
- the predicted value can be obtained by separating the sensor value change accompanying the system state change from the sensor value change unrelated to the system state change. Since it is possible to construct a regression equation that can be calculated, it is possible to detect signs of failure that are buried in a change in the state of the system.
- FIG. 6 is an explanatory diagram showing an example of a graph structure of an excellent cross-correlation model group having two or more clusters.
- reference numeral 703B represents an excellent cross-correlation model group that is a construction target of the graph structure.
- Reference numeral 703D represents the graph structure of the excellent cross-correlation model group 703B.
- Reference numerals 703F1 and 703F2 represent clusters included in the graph structure 703D.
- the data item classification means 1211 represents each data item used in the regression equation of each excellent cross-correlation model included in the excellent cross-correlation model group 703B by a nodal point, and between the nodal points, for each regression equation, A graph structure 703D is obtained by connecting the explanatory variable data item to the target variable data item with a line with an arrow.
- the graph structure 703D of this example includes a cluster 703F1 composed of data item A, data item B, data item C, and data item D, and data item K, data item L, and data item M.
- a cluster 703F2 composed of
- FIG. 7 is an explanatory diagram showing an example of a method for selecting representative data items when there are two or more clusters in the graph structure of the excellent cross-correlation model group.
- the many-body correlation model generation unit 1212 accumulates, for example, the excellentness F of the regression formula for each data item included in the cluster for each cluster. Calculate the value.
- the superiority F of the regression equation corresponding to each arrow connected to the node included in the cluster is used as the data item used in the regression equation.
- the process of accumulating is performed to obtain the accumulated value for each data item.
- the obtained accumulated value is used as the score of the data item.
- the score of each data item constituting the cluster 703F1 is indicated as 703E1
- the score of each data item constituting the cluster 703F2 is indicated as 703E2.
- the many-body correlation model generation unit 1212 sets, for each cluster, the data item having the maximum score among the data items constituting the cluster as the representative data item. When there are a plurality of data items having the maximum score in one cluster, for example, a representative data item may be selected at random from the data items.
- the scores 703E1 of the data items constituting the cluster 703F1 are 1.6, 1.7, 1..., Respectively, for the data item A, the data item B, the data item C, and the data item D. 5 and 1.6.
- data item B is selected as the representative data item of cluster 703F1.
- the data item B is selected as the representative data item of the data item group including the data item A, the data item B, the data item C, and the data item D.
- the scores 703E2 of the data items constituting the cluster 703F2 are 1.4, 0.7, and 0.7 for the data item K, the data item L, and the data item M, respectively. Therefore, the data item K is selected as the representative data item of the cluster 703F2.
- the data item K is selected as the representative data item of the data item group including the data item K, the data item L, and the data item M.
- the system analysis apparatus 100 may be a computer that includes a CPU and a storage medium that stores a program, and that operates under the control of the CPU based on the program.
- the state information collection unit 11, the analysis model generation unit 12, and the analysis unit 13 are realized by a CPU that operates according to a program.
- the state information storage unit 14 and the analysis model storage unit 15 are realized by a storage medium included in the computer. Note that the state information storage unit 14 and the analysis model storage unit 15 may be realized by individual storage media or may be realized by one storage medium.
- FIG. 8 is a flowchart illustrating an example of the operation of the system analysis apparatus 100 according to the present embodiment.
- the state information collection unit 11 of the system analysis apparatus 100 collects state information from the monitored apparatus 200 and stores it in the state information storage unit 14 (step S101).
- the state information collection unit 11 repeats the operation of step S101 until a determination is made to end the operation (Yes in step S108).
- the state information collecting unit 11 collects the state information at every predetermined cycle and stores the state information in the state information storage unit 14 while the operations in steps S102 to S107 are being performed.
- step S102 if the current timing is the timing at which the analysis model is generated (Yes in step S102), the analysis model generation unit 12 acquires state series information for a predetermined period used for generation of the analysis model from the state information storage unit 14. Then, an analysis model is generated using the acquired state series information (step S103). On the other hand, if the current timing is not the timing for generating the analysis model (No in step S102), the process proceeds to step S104. Whether or not it is the timing for generating the analysis model may be determined based on, for example, whether or not a determination to generate the analysis model has been given.
- step S104 the analysis unit 13 returns to step S101 if the determination that the target system is monitored using the current analysis model (Yes in step S104) is not given. On the other hand, if a determination is made to monitor the target system using the current analysis model (Yes in step S104), the process proceeds to step S105.
- step S105 the model destruction detection unit 131 uses the state information newly collected by the state information collection unit 11 to use the model destruction of the analysis model indicated by the analysis model information stored in the analysis model storage unit 15. Detect the situation. More specifically, the model destruction detection unit 131 determines whether or not model destruction has occurred in each of the many-body correlation models included in the analysis model, and generates model destruction information indicating the result.
- the abnormality determination unit 132 determines whether the model destruction status satisfies the alarm condition based on the model destruction information obtained from the model destruction detection unit 131. If the model destruction status satisfies the alarm condition (Yes in step S106), the abnormality determination unit 132 notifies the operator or the monitored system of the model destruction information indicating the determination result or the model destruction status (step S107). ), The process proceeds to step S108. On the other hand, if the state of model destruction does not satisfy the alarm condition (No in step S106), the abnormality determination unit 132 does not particularly do anything because no abnormality is detected in the system, and returns to step S101.
- the system analysis apparatus 100 continues the above operation until a determination is made to end the operation (Yes in step S108).
- FIG. 9 is a flowchart showing an example of the processing flow of the analysis model generation processing (step S103 in FIG. 8) by the analysis model generation means 12.
- the data item classification unit 1211 acquires state series information for a predetermined period used for generation of the analysis model from the state information storage unit 14, and uses the acquired state series information to create a data item.
- a cross-correlation model group is generated for classification (step S201).
- the data item classification means 1211 extracts a good cross-correlation model from the generated cross-correlation model group based on a predetermined good model condition to obtain a good cross-correlation model group (step S202).
- the data item classification means 1211 constructs a graph structure of the obtained excellent cross-correlation model group (step S203).
- step S203 the data item classification unit 1211 classifies the data items based on the constructed graph structure.
- the data item classification unit 1211 sets each cluster included in the graph structure as a data item group.
- the many-body correlation model generation unit 1212 selects a representative data item for each data item group classified by the data item classification unit 1211 (step S204).
- the many-body correlation model generation unit 1212 generates, for each data item group, the two data items and the representative data item for all combinations of any two data items belonging to the data item group excluding the representative data item.
- a multi-body correlation model having at least a regression equation using and a prediction error tolerance of the regression equation is generated to obtain a multi-body correlation model group (step S205).
- the model extracting means 122 extracts an excellent multi-body correlation model from the generated many-body correlation model group based on a predetermined excellent model condition to obtain an excellent cross-correlation model group (step S206).
- the model extraction unit 122 stores the analysis model information in the analysis model storage unit 15 using the obtained excellent many-body correlation model group as an analysis model (step S207).
- the system analysis apparatus 100 generates an analysis model that can detect an abnormality with high sensitivity even if the target system is complex or multi-collinearity exists between data items. Therefore, the system abnormality can be determined with high accuracy.
- the reason is the six features of this embodiment.
- the data item classification means 1211 generates a cross-correlation model group, selects a regression formula with a high degree of superiority from the regression formulas of the cross-correlation models included in the cross-correlation model group, and the regression formula
- the data items are classified for each data item included in the.
- a normal change in the system state such as the temperature, the load on the system, and the set value appears in the value of the data item.
- the index that represents the superiority of the regression equation such as the prediction accuracy of the regression equation that uses it, is high, so the regression necessary to obtain the desired superiority
- the number of data items included in the expression can be reduced. Since the number of data items included in the regression equation is small, it is possible to detect signs of failure that affect a small number of data items with high sensitivity. Furthermore, since the number of data items included in the regression equation is small, an effect that it is easier to identify the location where the abnormality has occurred can be obtained as compared with the case where the regression equation is constructed using many data items as explanatory variables.
- a representative data item is determined, and as a regression equation of the correlation model to be included in the analysis model, the representative data item and the two data excluding the representative data item are included. It is that the regression formula containing the item is constructed.
- a data item for representing an influence accompanying a change in the state of the system is added to the regression equation, an effect similar to that obtained by removing the influence accompanying a change in the state of the system from each data item can be obtained.
- the regression equation constructed by the above method the predicted value can be calculated by separating the change in the value of the data item accompanying the change in the state of the system and the change in the value of the data item unique to the data item.
- the prediction error can be calculated by paying attention to the fluctuations specific to the data item, so the data item due to the sign of failure that is buried in the change in the value of the data item due to the change in the system status only by monitoring the sensor value Even a change in the value of can be accurately detected.
- the third is that at least the data item with the highest cumulative value of superiority is selected as the representative data item.
- the data item with the highest cumulative value of the superiority to other data items is the data item belonging to the data item group. It is thought that the average effect of the state change appeared. Therefore, the representative data item selected by the above method is suitable as a data item for expressing the change of the sensor value accompanying the change in the state of the system, the change in the value of the data item accompanying the change in the state of the system, and the data This is useful for separating the change in the value of an item-specific data item.
- the fifth reason is that a plurality of regression equations may be constructed for one type of objective variable in the analysis model.
- the percentage of one regression equation that contributes to normal or abnormal judgment results for the system is small, so which data item is used as the objective variable. Can solve the problem that the judgment result of normal or abnormal system may differ.
- the number of regression equations constructed for one type of objective variable is not limited to one, so that anomaly detection omission due to the limitation of the objective variable can be prevented.
- the sixth point is that the selection of representative data items, objective variables, and explanatory variables is not specified in advance, and they are automatically determined appropriately based on the behavior of sensor values. Therefore, the logic related to selection of representative data items, objective variables, and explanatory variables is not affected by the complexity of the system. Therefore, it is possible to solve the problem that it becomes difficult to select appropriate objective variables and explanatory variables. In other words, even if the configuration or the like of the target system is complicated, appropriate target variables and explanatory variables can be selected, so that the state of the target system can be analyzed with high accuracy.
- the system analysis apparatus 100 of the present embodiment has a feature that the accuracy of narrowing down data items that are abnormal factors is high.
- the representative data items included in the regression equation are common, so the excellent generated from one data item group in the correlation model included in the analysis model Focusing on the many-body correlation model group, for example, the following determination can be made. For example, when most of the excellent many-body correlation model group is destroyed, it can be determined that there is a high possibility that the representative data item is a data item that is regarded as an abnormal factor. Conversely, when a small number of the excellent many-body correlation model group becomes model destroyed, it can be determined that there is a high possibility that the data items other than the representative data item are data items that are considered to be abnormal factors.
- the representative data item needs to be selected from a data item group that is grouped for each similar data item that is affected by a change in the state of the system that appears in the value of the data item. Furthermore, it is preferable that the data item has an average influence of a change in the state of the system in the group.
- the system analysis apparatus 100 can select the representative data item as such. Specifically, it is only necessary to select the data item having the highest cumulative value of the degree of superiority (prediction accuracy) in the data item group in which the highly related data item groups are collected.
- the explanatory variable includes the past value of the objective variable (for example, Expression (2) or Expression (5)) is shown as an example of the regression equation of the correlation model. You may exclude past values of.
- the value of the data item is used for the objective variable and the explanatory variable.
- a value obtained by numerically converting the value of the data item may be used. Examples of values obtained by numerically converting data item values include a difference, a power, an average value of state series information in a predetermined time width, and the like.
- each parameter determines each parameter so that the value of the prediction precision which is a superiority may become the maximum as an example of the generation method of the regression equation shown in Formula (2) or Formula (5)
- Other degrees of excellence may be used accordingly.
- each parameter is determined by using the reciprocal of the information criterion such as AIC (Akaike's Information Criterion) or BIC (Bayesian Information Criterion) as the superiority of the regression formula, Good.
- the numerator of the second term of the formula (3) is the minimum
- the method of determining the parameters a 1 to a N , b 0 to b M , and c using the least square method is shown, but is used in Lasso (least absolute shrinkage and selection operator), Ridge regression, etc. These loss parameters with regularization parameters may be used to determine these parameters so that the loss function with regularization parameters is minimized.
- the representative data item may be an objective variable.
- the representative data item may be an objective variable.
- a composite value of the plurality of representative data items may be used as the objective variable.
- these parameters may be determined using, for example, Wal's R criterion, Krzanowski's W criterion, Osten's F criterion, and the like.
- parameters a 1 to a N , b 0 to b M , c, d 0 to d Q , e 0 to K, L, N, M, P, Q, and S are fixed.
- PLS regression principal component regression
- the representative data items may be increased until a predetermined number predetermined by the operator is satisfied, or the information amount criterion is predetermined.
- the representative data items may be increased until the threshold is exceeded. A smaller number of representative data items for each data item group is desirable from the viewpoint of abnormality detection sensitivity and abnormality factor data items, and a larger number is desirable from the viewpoint of false alarm frequency.
- one type of index which is the cumulative value of the superiority of the regression equation
- a plurality of types of indices may be used.
- an index called the cumulative value of the superiority of the regression equation and an indicator called the speed of appearance of the change point described later are used to indicate each index.
- the example which the system analysis apparatus 100 monitors the state of a to-be-monitored system using one analysis model (analysis model which consists of a many-body correlation model group produced
- a plurality of analysis models may be created using state series information having different collection periods. In such a case, the state of the monitored system may be monitored while switching a plurality of analysis models.
- system analysis apparatus 100 monitors the state of the monitored system by using all regression equations included in the analysis model for detection of model destruction has been described. However, only a part of the analysis model is shown. It may be used to monitor the state of the monitored system (for example, only a part of the monitored devices 200).
- condition such as “when the number of correlation models in which model destruction has occurred exceeds a predetermined number” has been shown as an example of the alarm condition.
- a condition such as “when the cumulative value of the superiority of the regression equation of the model exceeds a predetermined value” may be used.
- a condition such as “when the alarm threshold is exceeded for a predetermined period set in advance” may be set as the alarm condition.
- all conditions based on the model destruction information can be set as the alarm condition.
- the monitored system is a power plant
- one or more multi-body correlation models can be generated from information indicating the state of the monitored system, and the generated one or more multi-body
- the monitored system may be another system as long as it can be determined whether or not an abnormality has occurred due to model destruction of the correlation model.
- the monitored system may be an IT system, a plant system, a structure, a transportation device, or the like.
- the system analysis apparatus 100 generates an analysis model as a data item of data items included in information indicating the state of the system, and detects model destruction.
- Embodiment 2 a second embodiment of the present invention will be described. This embodiment is the same as the first embodiment except for a method for selecting representative data items. For this reason, the same parts as those in the first embodiment are denoted by the same reference numerals, and description thereof is omitted.
- each data item group a data item whose value changes earliest in time series, that is, a data item in which a change point appears first is selected as a representative data item.
- the representative data items may be selected from the one where the information change point appears earlier.
- the multi-body correlation model generation means 1212 of this embodiment first randomly selects one of the data items included in the data item group classified by the data item classification means 1211. Then, the selected data item is set as a reference data item, and the reference data item having the maximum cross-correlation coefficient between the reference data item and each of the other data items in the data item group Find the amount of time shift in between. The data item having the largest time shift amount toward the past is set as the representative data item of the data item group.
- FIG. 10 is an explanatory diagram illustrating an example of a method for selecting representative data items in the present embodiment. Note that the example shown in FIG. 10 is an example in which representative data items are selected from one data item group generated as a result of grouping based on the graph structure 701D of the excellent cross-correlation model group 701B shown in FIG. It is.
- the multi-body correlation model generation means 1212 first randomly selects one data item from the data items included in the data item group corresponding to each cluster of the generated graph structure.
- the upper part of FIG. 10 shows the time series change of data item A, data item B, data item C, and data item D (schematic diagram of time series data) in the state series information used for generating the analysis model. ing.
- data item A is selected as the reference data item.
- the time shift amounts of the data item A, the data item B, the data item C, and the data item D with respect to the data item A which is the reference data item were 0, 3, 2, and ⁇ 1, respectively.
- FIG. 10 shows the amount of time shift by shifting the other data items in the time direction under the condition that the cross-correlation function between the reference data item A and the other data items has the maximum value.
- the unit of the time shift amount is one measurement time interval.
- the time shift amount takes a positive value because the other data item is moved in the future, that is, on the right side of the page, the cross-correlation between the data item A which is the reference data item and the other data item. Indicates that the function has the maximum value.
- the time shift amount takes a negative value when the other data item is moved in the past, that is, to the left side of the page, and the cross-correlation function between the data item A which is the reference data item and the other data item. Is the maximum value. Therefore, in this example, the data item having the maximum time shift amount is selected as the representative data item. In the example shown in FIG. 10, data item B is selected as the representative data item.
- the time context since the time context is also taken into account, when the data items in one data item group are composed only of items having similar waveforms, it causes a change in the system state.
- Data items can be extracted.
- the change in the data item value due to the change in the system state and the data item value specific to the data item Since the change can be separated from the change, it is possible to detect a change in the value of the data item due to a failure sign that is buried in the change in the value of the data item accompanying a change in the state of the system only by monitoring the sensor value.
- the past equation of the objective variable is not included in the regression equation of the temporary cross-correlation model, and the superiority of the regression equation is not included.
- the data items in the group may be configured using only the data items used in the high regression equation.
- Embodiment 3 a third embodiment of the present invention will be described.
- This embodiment is different from the above-described embodiments in that not only a many-body correlation model but also a cross-correlation model is used as a correlation model of an analysis model.
- a function that uses a cross-correlation model as an analysis model is added to the configuration of the first embodiment will be described as an example, but the function may be added to the second embodiment.
- the same parts as those in the first embodiment are denoted by the same reference numerals, and description thereof is omitted.
- FIG. 11 is a block diagram illustrating a configuration example of the system analysis apparatus 300 according to the present embodiment.
- the system analysis apparatus 300 shown in FIG. 11 is different from the configuration of the system analysis apparatus 100 of the first embodiment shown in FIG. 1 in that the analysis model generation means 12 is an analysis model generation means 32.
- the analysis model generation unit 32 includes an analysis model candidate generation unit 321 and a model extraction unit 322.
- the analysis model candidate generation unit 321 includes a cross-correlation model generation unit 3213, a data item classification unit 1211, and a many-body correlation model generation unit 1212.
- the point that the analysis model candidate generation unit 321 of the analysis model generation unit 32 further includes the cross-correlation model generation unit 3213 is different from the first embodiment.
- the operation of the model extraction unit 322 is also different from that of the first embodiment.
- the cross-correlation model generation unit 3213 generates a cross-correlation model that is a candidate for a correlation model to be included in the analysis model.
- the cross-correlation model generation unit 3213 constructs a regression equation including any two data items for all combinations of the arbitrary two data items, and sets the allowable range of the prediction error of the regression equation. The degree of excellence of the regression equation is calculated, and a cross-correlation model including at least the constructed regression equation and the allowable range of the prediction error of the regression equation is generated.
- the cross-correlation model generating unit 3213 may generate a cross-correlation model including the constructed regression equation, the allowable range of the prediction error of the regression equation, and the superiority of the regression equation.
- the cross-correlation model generated by the cross-correlation model generation unit 3213 may be referred to as an analysis cross-correlation model.
- a set of analysis cross-correlation models may be referred to as an analysis cross-correlation model group.
- the cross-correlation model group generated by the data item classification unit 1211 of the first embodiment may be referred to as a cross-correlation model group for classification.
- the data item classification means 1211 replaces the first temporary cross-correlation model with the objective variable data item and the explanatory variable data item in the first temporary cross-correlation model for any two data items.
- the second provisional cross-correlation model having the higher F of the regression formula a classification cross-correlation model group corresponding to one regression formula for any two data items is obtained.
- the cross-correlation model generation unit 3213 performs the same selection process. In other words, for any two data items, an analysis mutual expression in which one regression equation having a higher superiority among two regression equations in which the objective variable data item and the explanatory variable data item are interchanged is associated. A correlation model group is generated.
- the model extraction unit 322 of the present embodiment has the following function. That is, the model extraction unit 322 extracts a cross-correlation model that satisfies a predetermined excellent model condition from among the cross-correlation model group for analysis generated by the cross-correlation model generation unit 3213, and extracts it as an excellent cross-correlation model for analysis.
- the analyzed excellent cross-correlation model group is used as a correlation model included in the analysis model. Therefore, the analysis model of the present embodiment can include one or more cross-correlation models in addition to the one or more many-body correlation models.
- the model extraction means 322 stores the information of the many-body correlation model and the cross correlation model, which are the correlation models included in the analysis model as a result of extraction, in the analysis model storage means 15 as analysis model information.
- the cross-correlation model for analysis and the cross-correlation model for classification may be the same or different in the regression equation, the index of the superiority of the regression equation, and the excellent model condition.
- the analytical cross-correlation model and the many-body correlation model may be the same or different in the calculation method of the allowable range of the prediction error of the regression equation, the index of the superiority of the regression equation, and the excellent model conditions. Also good. In either case, the one shown in the first embodiment can be used.
- model destruction detection method for the cross-correlation model included in the analysis model is the same as that for the many-body correlation model. Accordingly, when new state information is collected, the model destruction detection unit 131 of the analysis unit 13 performs the first operation on each correlation model included in the analysis model indicated by the analysis model information stored in the analysis model storage unit 15. The presence or absence of model destruction may be detected by the same method as in the first embodiment.
- FIG. 12 is a flowchart showing an example of the operation of the analysis model generation unit 32 of the present embodiment.
- the same operations as those shown in FIG. 9 are denoted by the same reference numerals, and description thereof is omitted.
- the cross correlation model generation unit 3213 stores state series information for a predetermined period used for generation of an analysis model as state information storage.
- An analysis cross-correlation model group is generated using the state series information acquired from the means 14 (step S301).
- the model extraction means 322 extracts an excellent analysis cross-correlation model from the generated analysis cross-correlation model group based on a predetermined excellent model condition, and obtains an excellent analysis cross-correlation model group (step S302).
- the data item classification means 1211 and the many-body correlation model generation means 1212 cooperate to generate a many-body correlation model group (steps S201 to S205) in the same manner as in the first embodiment, and the model extraction means 322 extracts an excellent many-body correlation model from the generated many-body correlation model group based on a predetermined excellent model condition, and obtains an excellent many-body correlation model group (step S206).
- the model extraction means 322 stores the analysis model stored as an analysis model by combining the excellent cross-correlation model group for analysis obtained in step S302 and the excellent multi-body correlation model group obtained in step S206.
- the analysis model information is stored in the means 15 (step S303).
- the cross-correlation model is used not only for grouping data items, but also as a correlation model for the analysis model, so the range of data items to be analyzed. Can be wider.
- the data item classification unit 1211 uses the analysis cross correlation model group generated by the cross correlation model generation unit 3213 as it is.
- the cross-correlation model group for classification may be used.
- the excellent cross-correlation model group can be shared. That is, the analysis excellent cross-correlation model group may be used as it is as the classification excellent cross-correlation model group. In such a case, the processes in steps S201 to S202 in FIG. 12 are omitted.
- Embodiment 4 FIG. Next, a fourth embodiment of the present invention will be described.
- This embodiment is different from each of the above embodiments in that it includes a function of extracting an abnormality factor.
- the said function is added to the structure of 3rd Embodiment is shown as an example, However, The said function may be added to 1st Embodiment or 2nd Embodiment.
- the same parts as those of the third embodiment are denoted by the same reference numerals, and description thereof is omitted.
- FIG. 13 is a block diagram illustrating a configuration example of the system analysis apparatus 400 of the present embodiment.
- the system analysis apparatus 400 shown in FIG. 13 is different from the configuration of the system analysis apparatus 300 of the third embodiment shown in FIG. 11 in that the analysis means 13 is an analysis means 43.
- the analysis unit 43 of this embodiment includes a model destruction detection unit 131, an abnormality determination unit 132, an abnormality factor extraction unit 433, and a model destruction storage unit 434.
- the analysis unit 43 is different from the third embodiment in that the analysis unit 43 further includes an abnormality factor extraction unit 433 and a model destruction storage unit 434.
- the model destruction storage unit 434 stores the model destruction information generated by the model destruction detection unit 131.
- the registration processing of the model destruction information in the model destruction storage unit 434 may be performed, for example, when the abnormality determination unit 132 determines that the system state is abnormal.
- the alarm condition for determining whether the system state is abnormal or normal can be the one shown in the first embodiment.
- the abnormality factor extraction unit 433 calculates the degree of abnormality indicating the degree of abnormality for each data item from the newly added model destruction information. Then, at least one data item is extracted from the higher degree of abnormality, and each of the extracted data items is used as an abnormality factor candidate data item that is a candidate for an abnormality factor, and an abnormality factor candidate that is a set thereof The data item group is notified to the operator and the monitored system.
- the operator can set an arbitrary number of data items to be included in the abnormal factor candidate data item group, that is, the number of abnormal factor candidate data items to be notified, for example, as the number of data items to be confirmed at the time of abnormality.
- the number of abnormal cause candidate data items to be notified increases the possibility of finding the cause of the failure, and is preferable from the viewpoint of understanding the overall situation of the failure, and shortens the investigation time when a false alarm occurs From the point of view, the smaller the number, the better.
- the degree of abnormality of the data item is a value obtained by accumulating the number of data items included in the regression equation of the correlation model in which the model destruction has occurred among the correlation models included in the analysis model, or model destruction. It is possible to use a ratio between a value obtained by accumulating the number of data items included in the regression equation of the correlation model generated and a value obtained by accumulating the number of data items included in the regression equation of the correlation model in which the model destruction has not occurred. .
- the information necessary for the model destruction information in this embodiment depends on the alarm condition and the calculation method of the degree of abnormality of the data item. For example, if the alarm condition relates to the number of correlation models in which model destruction has occurred, information that can identify the number of correlation models in which model destruction has occurred may be included in the model destruction information. In addition, if the method for calculating the degree of abnormality of a data item uses a value obtained by accumulating the number of data items included in the regression equation in which model destruction has occurred for each data item, model destruction occurs in the model destruction information. Information indicating the data item used in the regression equation of the correlation model and information that can identify the correlation model in which the model destruction has occurred may be included.
- the calculation method of the degree of abnormality of the data item is a value related to the number of data items included in the regression equation where the model destruction occurred (cumulative value etc.) and a data item included in the regression equation where the model destruction did not occur If a value using a value (cumulative value, etc.) related to the number of (for example, a ratio of these) is used, the model destruction information includes a regression equation in which model destruction has occurred and a regression equation in which model destruction has not occurred. Information that can be specified may be included.
- FIG. 14 is a flowchart showing an example of the operation of the system analysis apparatus 400 of the present embodiment.
- the same operations as those shown in FIG. 8 are denoted by the same reference numerals, and description thereof is omitted.
- the abnormality determination unit 132 determines that the system state is abnormal in step S107, the fact or model destruction information is notified to the operator or the monitored system. At the same time, the model destruction information is stored in the model destruction storage means 434.
- the abnormal factor extraction unit 433 extracts one or more abnormal factor candidate data items based on the model destruction information newly stored in the model destruction storage unit 434, and shows the extracted abnormal factor candidate data item group Information is notified to the operator or the monitored system (step S401).
- the degree of abnormality of data items is the value obtained by accumulating the number of data items included in the regression equation of the correlation model in which the model destruction occurred among the correlation models included in the analysis model, and no model destruction occurred
- the number of data items included in the regression equation of the correlation model is a ratio with the accumulated value.
- FIG. 15 is an explanatory diagram illustrating an example of a method for extracting an abnormal factor candidate data item group.
- the analysis model used by the system analysis apparatus 400 includes seven correlation models.
- the upper part of FIG. 15 shows a graph structure of a correlation model group included in the analysis model (a structure diagram schematically showing a relationship between data items used in the correlation model group). In the graph structure, nodal points corresponding to common data items are not shown. Further, “common metric” in the figure represents a representative data item.
- the operator or the monitored system can narrow down the cause of the abnormality based on the information notified from the system analysis apparatus 400.
- the data item name is notified as information indicating the abnormal factor candidate data item group.
- the data item to be confirmed is also notified. Is preferred because it is given priority.
- FIG. 16 is a block diagram showing the main part of the system analysis apparatus according to the present invention. As shown in FIG. 16, the system analysis apparatus according to the present invention includes an analysis model generation means 51 as a main configuration.
- the analysis model generation unit 51 analyzes the state of the target system using state information that is a set of information related to a plurality of types of data items of the target system. And a data item classification unit 511, a many-body correlation model generation unit 512, and a model extraction unit 513.
- the analysis model is a multi-body correlation model that is a correlation model having at least a regression equation composed of three or more data items and an allowable range of prediction error of the regression equation. Including one or more.
- the data item classification unit 511 classifies the data item group included in the state information into one or more groups. More specifically, the data item classification means 511 includes, in at least one group after classification, a first data item that is one data item arbitrarily selected from the data items included in the group, and the first data item The superiority of the regression equation constructed using the second data item that is one of the data items included in the same group as the first data item excluding the data item can be taken for the first data item. When the calculation is performed for all the combinations of the second data items, the data items are classified so that at least one of the superiorities of the regression equation satisfies a predetermined excellent model condition.
- the many-body correlation model generation unit 512 (for example, the many-body correlation model generation unit 1212), for each group classified by the data item classification unit 511, at least one representative data item from the data items included in the group. And all combinations of any two data items included in the group excluding the selected representative data item are configured using the two data items and the representative data item.
- a multi-body which constructs a regression equation calculates an allowable range of prediction error of the regression equation and an excellent degree of the regression equation, and has at least the constructed regression equation and an allowable range of prediction error of the regression equation Generate a correlation model.
- the model extraction unit 513 (for example, the model extraction unit 122) is a multi-body correlation model in which the degree of regression is determined in advance from the many-body correlation model group generated by the many-body correlation model generation unit 512.
- a many-body correlation model that satisfies the excellent model condition is extracted as a many-body correlation model included in the analysis model.
- FIG. 17 is a block diagram showing another configuration example of the system analysis apparatus according to the present invention.
- the system analysis apparatus according to the present invention may be configured as shown in FIG. 17, for example.
- the system analysis apparatus shown in FIG. 17 includes an analysis model storage unit 52 and an analysis unit 53 in addition to the analysis model generation unit 51 described above.
- the analysis model storage unit 52 (for example, the analysis model storage unit 15) stores information on the analysis model generated by the analysis model generation unit 51.
- the analysis unit 53 (for example, the analysis unit 13 and the analysis unit 43) is a unit that analyzes the state of the system using the analysis model stored in the analysis model storage unit when new state information is acquired.
- the model destruction detecting unit 531 and the abnormality determining unit 532 are included.
- the model destruction detection unit 531 (for example, the model destruction detection unit 131) is included in the analysis model indicated by the analysis model information stored in the analysis model storage unit using the newly collected state information. For each correlation model, the presence or absence of model destruction, which is a phenomenon in which the predicted value of the objective variable of the regression equation of the correlation model exceeds the allowable range of the prediction error of the regression equation of the correlation model, is detected.
- the abnormality determination unit 532 determines whether the system state is abnormal or normal based on the detection result by the model destruction detection unit 531.
- An analysis model generating unit that generates an analysis model for analyzing the state of the target system by using state information that is a set of information on a plurality of types of data items of the target system.
- One or more multi-body correlation models which are correlation models having at least a regression equation configured using two or more data items and an allowable range of prediction error of the regression equation
- the analysis model generation means includes state information Data item classifying means for classifying the data item group included in one or more groups, and at least one representative data item from among the data items included in the group for each group classified by the data item classifying means All combinations for any two data items included in the group excluding the selected representative data item A regression equation composed of the two data items and the representative data item is calculated, and an allowable range of the prediction error of the regression equation and a superiority of the regression equation are calculated and constructed.
- a multi-body correlation model generating means for generating a multi-body correlation model having at least a regression equation obtained and an allowable range of a prediction error of the regression formula, and a multi-body correlation model group generated by the multi-body correlation model generating means
- a model extracting means for extracting a multi-body correlation model satisfying the pre-requisite condition of the multi-body correlation model for which the degree of excellence of the regression equation is predetermined as a multi-body correlation model to be included in the analysis model, and data items
- the classification means includes a first data item that is one data item arbitrarily selected from the data items included in the group in at least one group after classification, and the first data item For the first data item, the superiority of the regression equation constructed using the second data item that is one of the data items included in the same group as the first data item excluding the data item is taken.
- An analysis model generation device characterized by classifying data items such that at least one of the superiorities of the regression equation satisfies a predetermined excellent model condition when calculated for all possible
- the data item group contained in the status information which is a collection of the information regarding the multiple types of data items of the target system is arbitrarily selected from the data items included in the group in at least one group after classification. Constructed using a first data item that is one selected data item and a second data item that is one of the data items included in the same group as the first data item excluding the first data item.
- the superiority of the regression equation is calculated for all combinations of the second data items that can be taken with respect to the first data item, at least one of the superiorities of the regression equation satisfies the predetermined excellent model condition
- An analysis model generation program for executing a multi-body correlation model generation process for generating a multi-body correlation model having at least an allowable range of prediction error of an expression.
- the present invention can be suitably applied to the detection of system faults and fault signs and the extraction of those factors.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
Description
モデル抽出手段が、生成された多体相関モデル群の中から、回帰式の優良度が予め定められている多体相関モデルの優良モデル条件を満たす多体相関モデルを、分析モデルに含ませる多体相関モデルとして抽出し、抽出された多体相関モデル群を含む分析モデルの情報を所定の記憶装置に記憶し、モデル破壊検出手段が、状態情報が新たに取得されると、新たに収集された状態情報を用いて、所定の記憶装置に記憶されている分析モデルの情報により示される分析モデルに含まれる各相関モデルについて、当該相関モデルの回帰式の目的変数の予測値が、当該相関モデルの回帰式の予測誤差の許容範囲を超える現象であるモデル破壊の発生の有無を検出し、異常判定手段が、モデル破壊検出手段による検出結果に基づいて、システムの状態が異常か正常かを判定することを特徴とする。
以下、本発明の実施形態を図面を参照して説明する。以下の各実施形態では、発電プラントシステムの異常検知に本発明のシステム分析装置を適用した場合を例に用いて説明する。
ri≧T2 ・・・(7)
T2=-T1 ・・・(9)
次に、本発明の第2の実施形態について説明する。本実施形態は、代表データ項目の選出方法を除いて第1の実施形態と同じである。このため、第1の実施形態と同じ部分については同一の符号を付し、説明を省略する。
次に、本発明の第3の実施形態について説明する。本実施形態は、分析モデルの相関モデルとして、多体相関モデルだけでなく、相互相関モデルを用いる点で上記の各実施形態と異なる。以下では、第1の実施形態の構成に、相互相関モデルを分析モデルとして利用する機能を追加する場合を例に示すが、当該機能は、第2の実施形態に追加されてもよい。以下では、第1の実施形態と同じ部分については同一の符号を付し、説明を省略する。
次に、本発明の第4の実施形態について説明する。本実施形態は、異常要因を抽出する機能を含む点で上記の各実施形態と異なる。以下では、第3の実施形態の構成に、当該機能を追加した場合を例に示すが、当該機能は、第1の実施形態や第2の実施形態に追加されてもよい。以下では、第3の実施形態と同じ部分については同一の符号を付し、説明を省略する。
12、32 分析モデル生成手段
121、321 分析モデル候補生成手段
1211 データ項目分類手段
1212 多体相関モデル生成手段
3213 相互相関モデル生成手段
122、322 モデル抽出手段
13、43 分析手段
131 モデル破壊検出手段
132 異常判定手段
433 異常要因抽出手段
434 モデル破壊記憶手段
14 状態情報記憶手段
15 分析モデル記憶手段
100、300、400 システム分析装置
200 被監視装置
51 分析モデル生成手段
511 データ項目分類手段
512 多体相関モデル生成手段
513 モデル抽出手段
52 分析モデル記憶手段
53 分析手段
531 モデル破壊検出手段
532 異常判定手段
701A 相互相関モデル群
701B、703B 優良相互相関モデル群
701C 優良モデル条件
701D、703D グラフ構造
701E、703E1、703E2 スコア
703F1、703F2 クラスタ
702A 多体相関モデル群
702B 優良多体相関モデル群
702C 優良モデル条件
Claims (10)
- 対象システムの複数種のデータ項目に関する情報の集合である状態情報を用いて、前記対象システムの状態を分析するための分析モデルを生成する分析モデル生成手段を備え、
前記分析モデルは、3つ以上のデータ項目を用いて構成される回帰式と、当該回帰式の予測誤差の許容範囲とを少なくとも有する相関モデルである多体相関モデルを1つ以上含み、
前記分析モデル生成手段は、
前記状態情報に含まれるデータ項目群を1つ以上のグループに分類するデータ項目分類手段と、
前記データ項目分類手段によって分類されたグループごとに、該グループに含まれるデータ項目の中から、少なくとも1つの代表データ項目を選出し、選出した代表データ項目を除く該グループに含まれるデータ項目のうちの任意の2つのデータ項目に対する全ての組み合わせについて、当該2つのデータ項目と、前記代表データ項目とを用いて構成される回帰式を構築するとともに、該回帰式の予測誤差の許容範囲と、該回帰式の優良度とを算出し、構築した前記回帰式と、該回帰式の予測誤差の許容範囲とを少なくとも有する多体相関モデルを生成する多体相関モデル生成手段と、
前記多体相関モデル生成手段によって生成された多体相関モデル群の中から、回帰式の優良度が予め定められている多体相関モデルの優良モデル条件を満たす多体相関モデルを、前記分析モデルに含ませる多体相関モデルとして抽出するモデル抽出手段とを含み、
前記データ項目分類手段は、分類後の少なくとも1つのグループにおいて、当該グループに含まれるデータ項目から任意に選択された1つのデータ項目である第1データ項目と、該第1データ項目を除く該第1データ項目と同一のグループに含まれるデータ項目の1つである第2データ項目とを用いて構築される回帰式の優良度を、前記第1データ項目に対して取りうる第2データ項目の全ての組み合わせについて算出したとき、該回帰式の優良度の少なくとも1つが、所定の優良モデル条件を満たすように、データ項目を分類する
ことを特徴とするシステム分析装置。 - 分析モデル生成手段によって生成された分析モデルの情報を記憶する分析モデル記憶手段と、
状態情報が新たに取得されると、前記分析モデル記憶手段に記憶された前記分析モデルを用いて、システムの状態を分析する分析手段とを備え、
前記分析手段は、
新たに収集された前記状態情報を用いて、前記分析モデル記憶手段に記憶されている分析モデルの情報により示される分析モデルに含まれる各相関モデルについて、当該相関モデルの回帰式の目的変数の予測値が、当該相関モデルの回帰式の予測誤差の許容範囲を超える現象であるモデル破壊の発生の有無を検出するモデル破壊検出手段と、
前記モデル破壊検出手段による検出結果に基づいて、システムの状態が異常か正常かを判定する異常判定手段とを含む
請求項1に記載のシステム分析装置。 - 分析手段は、
異常判定手段によって異常と判定された場合に、モデル破壊検出手段による検出結果に基づいて、異常要因の候補とするデータ項目を抽出する異常要因抽出手段を含み、
前記異常要因抽出手段は、前記モデル破壊検出手段による検出の結果示される、データ項目別のモデル破壊の発生状況に基づいて、データ項目別に異常の度合いを示す異常度を算出し、算出したデータ項目別の異常度に基づいて、異常要因の候補とするデータ項目を抽出する
請求項2に記載のシステム分析装置。 - 代表データ項目は、選出対象とされたグループ内の任意の2つのデータ項目を用いて構成される各回帰式の優良度を用いて算出される当該グループに属するデータ項目別の統計値に基づいて、選出される
請求項1から請求項3のうちのいずれか1項に記載のシステム分析装置。 - データ項目別の統計値は、代表データ項目の選出対象とされたグループに属する各データ項目について、当該グループ内の任意の2つのデータ項目を用いて構成される回帰式のうち当該データ項目を用いて構成される回帰式の優良度を用いて算出される、前記優良度の平均値、中央値、最小値、最大値、累積値のいずれかである
請求項4に記載のシステム分析装置。 - 代表データ項目は、所定期間分の状態情報によって示される、選出対象とされたグループ内の各データ項目の情報の変化点の出現の早さに基づいて選出される
請求項1から請求項5のうちのいずれか1項に記載のシステム分析装置。 - 分析モデルは、2つのデータ項目を用いて構成される回帰式と、当該回帰式の予測誤差の許容範囲とを少なくとも有する相関モデルである相互相関モデルをさらに1つ以上含み、
分析モデル生成手段は、
状態情報に含まれるデータ項目群の任意の2つのデータ項目に対する全ての組み合わせについて、当該2つのデータ項目を用いて構成される回帰式を構築するとともに、該回帰式の予測誤差の許容範囲と、該回帰式の優良度とを算出して、構築した前記回帰式と、該回帰式の予測誤差の許容範囲とを少なくとも有する相互相関モデルを生成する相互相関モデル生成手段を含み、
モデル抽出手段は、多体相関モデル生成手段によって生成された多体相関モデル群の中から、回帰式の優良度が予め定められている多体相関モデルの優良モデル条件を満たす多体相関モデルを、前記分析モデルに含ませる多体相関モデルとして抽出するとともに、前記相互相関モデル生成手段によって生成された相互相関モデル群の中から、回帰式の優良度が予め定められている相互相関モデルの優良モデル条件を満たす相互相関モデルを、前記分析モデルに含ませる相互相関モデルとして抽出する
請求項1から請求項6のうちのいずれか1項に記載のシステム分析装置。 - データ項目分類手段が、対象システムの複数種のデータ項目に関する情報の集合である状態情報に含まれるデータ項目群を、分類後の少なくとも1つのグループにおいて、当該グループに含まれるデータ項目から任意に選択された1つのデータ項目である第1データ項目と、該第1データ項目を除く該第1データ項目と同一のグループに含まれるデータ項目の1つである第2データ項目とを用いて構築される回帰式の優良度を、前記第1データ項目に対して取りうる第2データ項目の全ての組み合わせについて算出したとき、該回帰式の優良度の少なくとも1つが、所定の優良モデル条件を満たすように、1つ以上のグループに分類し、
多体相関モデル生成手段が、前記状態情報を用いて、分類された前記グループごとに、該グループに含まれるデータ項目の中から、少なくとも1つの代表データ項目を選出し、選出した代表データ項目を除く該グループに含まれるデータ項目のうちの任意の2つのデータ項目に対する全ての組み合わせについて、当該2つのデータ項目と、前記代表データ項目とを用いて構成される回帰式を構築するとともに、該回帰式の予測誤差の許容範囲と、該回帰式の優良度とを算出して、構築した前記回帰式と、該回帰式の予測誤差の許容範囲とを少なくとも有する多体相関モデルを生成し、
モデル抽出手段が、生成された前記多体相関モデル群の中から、回帰式の優良度が予め定められている多体相関モデルの優良モデル条件を満たす多体相関モデルを、前記対象システムの状態を分析するための分析モデルに含ませる多体相関モデルとして抽出する
ことを特徴とする分析モデル生成方法。 - データ項目分類手段が、対象システムの複数種のデータ項目に関する情報の集合である状態情報に含まれるデータ項目群を、分類後の少なくとも1つのグループにおいて、当該グループに含まれるデータ項目から任意に選択された1つのデータ項目である第1データ項目と、該第1データ項目を除く該第1データ項目と同一のグループに含まれるデータ項目の1つである第2データ項目とを用いて構築される回帰式の優良度を、前記第1データ項目に対して取りうる第2データ項目の全ての組み合わせについて算出したとき、該回帰式の優良度の少なくとも1つが、所定の優良モデル条件を満たすように、1つ以上のグループに分類し、
多体相関モデル生成手段が、前記状態情報を用いて、分類された前記グループごとに、該グループに含まれるデータ項目の中から、少なくとも1つの代表データ項目を選出し、選出した代表データ項目を除く該グループに含まれるデータ項目のうちの任意の2つのデータ項目に対する全ての組み合わせについて、当該2つのデータ項目と、前記代表データ項目とを用いて構成される回帰式を構築するとともに、該回帰式の予測誤差の許容範囲と、該回帰式の優良度とを算出して、構築した前記回帰式と、該回帰式の予測誤差の許容範囲とを少なくとも有する多体相関モデルを生成し、
モデル抽出手段が、生成された前記多体相関モデル群の中から、回帰式の優良度が予め定められている多体相関モデルの優良モデル条件を満たす多体相関モデルを、分析モデルに含ませる多体相関モデルとして抽出し、抽出された前記多体相関モデル群を含む分析モデルの情報を所定の記憶装置に記憶し、
モデル破壊検出手段が、状態情報が新たに取得されると、新たに収集された前記状態情報を用いて、前記所定の記憶装置に記憶されている分析モデルの情報により示される分析モデルに含まれる各相関モデルについて、当該相関モデルの回帰式の目的変数の予測値が、当該相関モデルの回帰式の予測誤差の許容範囲を超える現象であるモデル破壊の発生の有無を検出し、
異常判定手段が、前記モデル破壊検出手段による検出結果に基づいて、システムの状態が異常か正常かを判定する
ことを特徴とするシステム分析方法。 - コンピュータに、
対象システムの複数種のデータ項目に関する情報の集合である状態情報に含まれるデータ項目群を、分類後の少なくとも1つのグループにおいて、当該グループに含まれるデータ項目から任意に選択された1つのデータ項目である第1データ項目と、該第1データ項目を除く該第1データ項目と同一のグループに含まれるデータ項目の1つである第2データ項目とを用いて構築される回帰式の優良度を、前記第1データ項目に対して取りうる第2データ項目の全ての組み合わせについて算出したとき、該回帰式の優良度の少なくとも1つが、所定の優良モデル条件を満たすように、1つ以上のグループに分類するデータ項目分類処理、
前記状態情報を用いて、分類された前記グループごとに、該グループに含まれるデータ項目の中から、少なくとも1つの代表データ項目を選出し、選出した代表データ項目を除く該グループに含まれるデータ項目のうちの任意の2つのデータ項目に対する全ての組み合わせについて、当該2つのデータ項目と、前記代表データ項目とを用いて構成される回帰式を構築するとともに、該回帰式の予測誤差の許容範囲と、該回帰式の優良度とを算出して、構築した前記回帰式と、該回帰式の予測誤差の許容範囲とを少なくとも有する多体相関モデルを生成する多体相関モデル生成処理、
生成された前記多体相関モデル群の中から、回帰式の優良度が予め定められている多体相関モデルの優良モデル条件を満たす多体相関モデルを、分析モデルに含ませる多体相関モデルとして抽出するモデル抽出処理、
抽出された前記多体相関モデル群を含む分析モデルの情報を、所定の記憶装置に記憶する処理、
状態情報が新たに取得されると、新たに収集された前記状態情報を用いて、前記所定の記憶装置に記憶されている分析モデルの情報により示される分析モデルに含まれる各相関モデルについて、当該相関モデルの回帰式の目的変数の予測値が、当該相関モデルの回帰式の予測誤差の許容範囲を超える現象であるモデル破壊の発生の有無を検出するモデル破壊検出処理、および
前記モデル破壊検出処理での検出結果に基づいて、システムの状態が異常か正常かを判定する異常判定処理
を実行させるためのシステム分析プログラム。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2016509613A JP6493389B2 (ja) | 2014-03-27 | 2014-10-21 | システム分析装置、分析モデル生成方法、システム分析方法およびシステム分析プログラム |
US15/129,402 US20170103148A1 (en) | 2014-03-27 | 2014-10-21 | System-analyzing device, analysis-model generation method, system analysis method, and system-analyzing program |
EP14887291.4A EP3125057B1 (en) | 2014-03-27 | 2014-10-21 | System-analyzing device, analysis-model generation method, system analysis method, and system-analyzing program |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2014-065120 | 2014-03-27 | ||
JP2014065120 | 2014-03-27 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2015145500A1 true WO2015145500A1 (ja) | 2015-10-01 |
Family
ID=54194102
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2014/005336 WO2015145500A1 (ja) | 2014-03-27 | 2014-10-21 | システム分析装置、分析モデル生成方法、システム分析方法およびシステム分析プログラム |
Country Status (4)
Country | Link |
---|---|
US (1) | US20170103148A1 (ja) |
EP (1) | EP3125057B1 (ja) |
JP (1) | JP6493389B2 (ja) |
WO (1) | WO2015145500A1 (ja) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2017215832A (ja) * | 2016-06-01 | 2017-12-07 | 株式会社神戸製鋼所 | 回転機械の運転状態を診断する診断装置及び診断方法 |
JPWO2017064855A1 (ja) * | 2015-10-13 | 2018-08-02 | 日本電気株式会社 | 構造物異常検知システム、構造物異常検知方法及び記録した記録媒体 |
JP2018180759A (ja) * | 2017-04-07 | 2018-11-15 | 株式会社日立製作所 | システム分析装置、及びシステム分析方法 |
JP2018538587A (ja) * | 2016-02-01 | 2018-12-27 | ▲騰▼▲訊▼科技(深▲セン▼)有限公司 | リスク評価方法およびシステム |
WO2020105179A1 (ja) * | 2018-11-22 | 2020-05-28 | 日本電気株式会社 | 情報処理装置、制御方法、及びプログラム |
JPWO2019026193A1 (ja) * | 2017-08-02 | 2020-07-02 | 日本電気株式会社 | 情報処理装置、情報処理システム、情報処理方法、及び、プログラム |
US10866311B2 (en) | 2017-09-15 | 2020-12-15 | Kabushiki Kaisha Toshiba | Distance measuring device |
JP2021039565A (ja) * | 2019-09-03 | 2021-03-11 | 東芝情報システム株式会社 | 状態変動検出補助装置、状態変動検出装置、状態変動検出補助用プログラム、及び状態変動検出用プログラム |
JP2021168364A (ja) * | 2020-04-13 | 2021-10-21 | キヤノン株式会社 | 情報処理装置、検出方法、プログラム、基板処理システム、及び物品の製造方法 |
US11216741B2 (en) | 2017-03-13 | 2022-01-04 | Kabushiki Kaisha Toshiba | Analysis apparatus, analysis method, and non-transitory computer readable medium |
TWI838606B (zh) | 2020-04-13 | 2024-04-11 | 日商佳能股份有限公司 | 資訊處理裝置、檢測方法、程式、基板處理系統及物品之製造方法 |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11080613B1 (en) * | 2016-04-29 | 2021-08-03 | EMC IP Holding Company LLC | Process monitoring based on large-scale combination of time series data |
JP6879239B2 (ja) | 2018-03-14 | 2021-06-02 | オムロン株式会社 | 異常検知システム、サポート装置およびモデル生成方法 |
CN108595892A (zh) * | 2018-05-11 | 2018-09-28 | 南京林业大学 | 基于时间差分模型的软测量建模方法 |
JP7124648B2 (ja) * | 2018-11-06 | 2022-08-24 | 株式会社島津製作所 | データ処理装置及びデータ処理プログラム |
CN110414713A (zh) * | 2019-06-27 | 2019-11-05 | 电子科技大学 | 一种基于同步数据流压缩的径流实时预测方法 |
WO2023215903A1 (en) * | 2022-05-06 | 2023-11-09 | Mapped Inc. | Automatic link prediction for devices in commercial and industrial environments |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2003314933A (ja) * | 2002-04-17 | 2003-11-06 | Tokyo Gas Co Ltd | ヒートポンプ熱交換器の異常検出装置 |
JP2006350698A (ja) * | 2005-06-16 | 2006-12-28 | Taiyo Nippon Sanso Corp | 異常診断装置、異常診断方法、異常診断プログラム |
JP2010218187A (ja) * | 2009-03-17 | 2010-09-30 | Fuji Electric Holdings Co Ltd | 製造条件調整装置 |
JP5108116B2 (ja) * | 2009-01-14 | 2012-12-26 | 株式会社日立製作所 | 装置異常監視方法及びシステム |
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 (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP5297272B2 (ja) * | 2009-06-11 | 2013-09-25 | 株式会社日立製作所 | 装置異常監視方法及びシステム |
WO2012086444A1 (ja) * | 2010-12-24 | 2012-06-28 | 日本電気株式会社 | 監視データ分析装置、監視データ分析方法および監視データ分析プログラム |
JP5284433B2 (ja) * | 2011-09-14 | 2013-09-11 | 株式会社東芝 | プロセス監視・診断・支援装置 |
-
2014
- 2014-10-21 JP JP2016509613A patent/JP6493389B2/ja active Active
- 2014-10-21 WO PCT/JP2014/005336 patent/WO2015145500A1/ja active Application Filing
- 2014-10-21 EP EP14887291.4A patent/EP3125057B1/en active Active
- 2014-10-21 US US15/129,402 patent/US20170103148A1/en not_active Abandoned
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2003314933A (ja) * | 2002-04-17 | 2003-11-06 | Tokyo Gas Co Ltd | ヒートポンプ熱交換器の異常検出装置 |
JP2006350698A (ja) * | 2005-06-16 | 2006-12-28 | Taiyo Nippon Sanso Corp | 異常診断装置、異常診断方法、異常診断プログラム |
JP5108116B2 (ja) * | 2009-01-14 | 2012-12-26 | 株式会社日立製作所 | 装置異常監視方法及びシステム |
JP2010218187A (ja) * | 2009-03-17 | 2010-09-30 | Fuji Electric Holdings Co Ltd | 製造条件調整装置 |
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 EP3125057A4 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPWO2017064855A1 (ja) * | 2015-10-13 | 2018-08-02 | 日本電気株式会社 | 構造物異常検知システム、構造物異常検知方法及び記録した記録媒体 |
JP2018538587A (ja) * | 2016-02-01 | 2018-12-27 | ▲騰▼▲訊▼科技(深▲セン▼)有限公司 | リスク評価方法およびシステム |
JP2017215832A (ja) * | 2016-06-01 | 2017-12-07 | 株式会社神戸製鋼所 | 回転機械の運転状態を診断する診断装置及び診断方法 |
US11216741B2 (en) | 2017-03-13 | 2022-01-04 | Kabushiki Kaisha Toshiba | Analysis apparatus, analysis method, and non-transitory computer readable medium |
JP2018180759A (ja) * | 2017-04-07 | 2018-11-15 | 株式会社日立製作所 | システム分析装置、及びシステム分析方法 |
US11941495B2 (en) | 2017-08-02 | 2024-03-26 | Nec Corporation | Information processing device, information processing method, and recording medium |
JPWO2019026193A1 (ja) * | 2017-08-02 | 2020-07-02 | 日本電気株式会社 | 情報処理装置、情報処理システム、情報処理方法、及び、プログラム |
US10866311B2 (en) | 2017-09-15 | 2020-12-15 | Kabushiki Kaisha Toshiba | Distance measuring device |
JPWO2020105179A1 (ja) * | 2018-11-22 | 2021-09-30 | 日本電気株式会社 | 情報処理装置、制御方法、及びプログラム |
JP7211429B2 (ja) | 2018-11-22 | 2023-01-24 | 日本電気株式会社 | 情報処理装置、制御方法、及びプログラム |
WO2020105179A1 (ja) * | 2018-11-22 | 2020-05-28 | 日本電気株式会社 | 情報処理装置、制御方法、及びプログラム |
JP2021039565A (ja) * | 2019-09-03 | 2021-03-11 | 東芝情報システム株式会社 | 状態変動検出補助装置、状態変動検出装置、状態変動検出補助用プログラム、及び状態変動検出用プログラム |
JP2021168364A (ja) * | 2020-04-13 | 2021-10-21 | キヤノン株式会社 | 情報処理装置、検出方法、プログラム、基板処理システム、及び物品の製造方法 |
JP7423396B2 (ja) | 2020-04-13 | 2024-01-29 | キヤノン株式会社 | 情報処理装置、検出方法、プログラム、基板処理システム、及び物品の製造方法 |
TWI838606B (zh) | 2020-04-13 | 2024-04-11 | 日商佳能股份有限公司 | 資訊處理裝置、檢測方法、程式、基板處理系統及物品之製造方法 |
Also Published As
Publication number | Publication date |
---|---|
EP3125057B1 (en) | 2019-07-03 |
JP6493389B2 (ja) | 2019-04-03 |
US20170103148A1 (en) | 2017-04-13 |
EP3125057A4 (en) | 2018-06-06 |
JPWO2015145500A1 (ja) | 2017-04-13 |
EP3125057A1 (en) | 2017-02-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6493389B2 (ja) | システム分析装置、分析モデル生成方法、システム分析方法およびシステム分析プログラム | |
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 | |
KR102011620B1 (ko) | 이상 데이터의 중요도 판정 장치 및 이상 데이터의 중요도 판정 방법 | |
WO2019142331A1 (ja) | 障害予測システムおよび障害予測方法 | |
JP5740459B2 (ja) | 設備状態監視方法 | |
JP5431235B2 (ja) | 設備状態監視方法およびその装置 | |
JP5538597B2 (ja) | 異常検知方法及び異常検知システム | |
JP5753286B1 (ja) | 情報処理装置、診断方法、およびプログラム | |
JP5868216B2 (ja) | クラスタリング装置及びクラスタリングプログラム | |
JP5875726B1 (ja) | 異常予兆診断装置のプリプロセッサ及びその処理方法 | |
WO2010095314A1 (ja) | 異常検知方法及び異常検知システム | |
JP2009536971A (ja) | 異常事象検出(aed)技術のポリマープロセスへの適用 | |
JP6164311B1 (ja) | 情報処理装置、情報処理方法、及び、プログラム | |
KR20170078252A (ko) | 시계열의 데이터를 모니터링 하는 방법 및 그 장치 | |
JP7469991B2 (ja) | 診断装置及びパラメータ調整方法 | |
CN111241683A (zh) | 一种基于动态时间规整的设备工况故障预测方法和系统 | |
JPWO2019073512A1 (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 | |
JP2023518537A (ja) | 製油所の単数または複数の蒸留塔の目詰まりの予測方法、コンピュータプログラムおよび関連する予測システム | |
Hu et al. | A multi-feature-based fault diagnosis method based on the weighted timeliness broad learning system | |
KR20120074630A (ko) | 설비 이상을 예측하기 위한 의사 결정 나무 구축 방법 및 그 시스템 | |
JP7330058B2 (ja) | 性能診断システム及びその応用システム | |
JP7012928B2 (ja) | 状態変動検出装置及び状態変動検出用プログラム | |
Viswanathan et al. | An Application of Sensor and Streaming Analytics to Oil Production. | |
JP2017182450A (ja) | 管理装置、これを備える管理システム、及び、管理方法 |
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: 14887291 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2016509613 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 15129402 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: 2014887291 Country of ref document: EP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2014887291 Country of ref document: EP |