JP4922265B2 - Plant monitoring apparatus and plant monitoring method - Google Patents

Description

  The present invention relates to a monitoring apparatus and a monitoring method for monitoring changes in the operating state of a plant, and more particularly, to a plant monitoring apparatus and a plant monitoring method with higher versatility for various abnormalities.

  In a large-scale plant such as a nuclear power plant or a thermal power plant, a large number of plant signals are taken into a computer and plant data is displayed on a monitoring screen for monitoring. Since there are thousands of plant signals, it is difficult for operators to constantly monitor all plant data. Therefore, monitoring of main plant data by comparison with preset control thresholds Is done by the computer. Operators are relevant because the plant signals of large plants change in close relation to each other, not to mention examples of feed and condensate systems in nuclear or thermal power plants that heat feed water with steam extracted from turbines. In addition to displaying and monitoring plant data trends, monitoring is performed while checking correlation diagrams and the like. When an abnormality occurs in the plant and the level of the plant signal exceeds the management threshold value, an alarm is issued and an operator can take action.

  However, it is a very difficult task for operators to detect signs that appear in the initial stage of abnormalities from a large number of plant data at an early stage and identify them as abnormal due to the effects of changes in operating conditions. . Even when a large-scale plant such as a power plant is used, and the operation state of a plurality of facilities is remotely monitored centrally, the difficulty of monitoring data of a large number of signals with a small number of personnel is the same.

  For this reason, principal component analysis generally used in the field of statistical process management technology is applied to a large number of plant data to convert it into a small number of statistical data that captures the characteristics of the original plant data. A technique has been devised in which a management threshold value is provided for monitoring a temporal trend of statistical data later or correlation characteristics between statistical data (see, for example, Patent Documents 1 to 3).

  Further, a technique has been devised for monitoring by comparing a pattern of correlation characteristics of a small number of converted statistic data with a normal pattern (see Patent Documents 4 and 5).

  Furthermore, a technique has been devised in which a management threshold is provided for monitoring the result of applying wavelet transform to the small number of converted statistical data (see Patent Document 6).

Separately, principal component analysis is applied to the deviation between the monitored data and the estimated value of the data estimated using a model representing the relationship between plant data, and the obtained statistical data is monitored. A technique for monitoring by providing a threshold has also been devised (see Patent Document 7). Alternatively, a method has been devised in which various management charts used in statistical process management are created for estimated deviations from a model and a management threshold value is provided for monitoring the charts (see Patent Document 8).
Japanese Patent No. 3970483 JP 2002-25981 A JP 2006-135212 A Japanese Patent No. 3495129 Japanese Patent No. 3472729 JP 2008-59270 A JP 2004-303007 A JP-A-5-157449

  Among the techniques for setting and monitoring the management threshold value directly on the statistical data converted by the above-described conventional principal component analysis, the technique described in Patent Document 1 includes three process state quantities related to each other. In order to determine the influence of disturbance from changes, an approximate curve by the least square method is created for the data obtained by rotating the three-dimensional principal component space obtained by performing principal component analysis from the viewpoint of a specific principal component direction. The abnormality is determined by comparing the distance from the curve with the permissible range, but is it possible to remove or emphasize the influence of one state quantity in the two-dimensional plane viewed from a specific principal component direction? Whether or not all depends on the characteristics of the target process, there is a problem that it can not be widely applied to general processes.

In addition, the technique described in Patent Document 2 has a problem of detecting a non-abnormal state as an abnormality when the plant operating conditions are changed, as pointed out in Patent Document 7 described later. In the technique described in Patent Document 3, monitoring is performed using only the Q statistics among the statistics data converted by the principal component analysis. However, various abnormalities that may occur in the plant are detected. Considering that some of them are not detected by the Q statistic but only affect the main component score or the T ² statistic, there is a problem that the versatility is lacking.

  On the other hand, among the techniques for monitoring by comparing the pattern of correlation characteristics of a small number of statistical data converted by the principal component analysis with a normal pattern, the cumulative contribution ratio is a specified value in the method described in Patent Document 4 This method is used to select a principal component that falls within the range of, and when the distance between the judged data calculated on the Cartesian coordinate system using these as parameters and the history data exceeds a threshold, However, there is a problem that it is difficult to grasp a state in which an abnormality gradually develops from a normal state, and it is easily affected by a transient noise.

  Further, the technique described in Patent Document 5 has a problem of lack of versatility because only the main component score is used, contrary to the case of Patent Document 3.

  The techniques described in Patent Documents 6 and 7 that monitor the result of applying the wavelet transform to a small number of statistical data converted by principal component analysis are managed in the same manner as in Patent Document 2 described above. There was a problem that there is a high possibility that the influence of changes in operating conditions is judged as abnormal.

In the technique described in Patent Document 6, a management threshold is provided for the wavelet transform of the T ² statistic or Q statistic to manage the T ² statistic or the Q statistic itself or their contribution. Although thresholds are provided, since both the T ² statistic and the Q statistic are square quantities, even if the ratio of abnormal data included in the data used for setting the management threshold is known, There was a problem that the data dependency of the set management threshold became high.

In contrast to the problem affected by changes in the plant operating conditions pointed out in the prior art described above, the technique described in Patent Document 7 uses a plurality of process data as external variables that directly represent changes in operating conditions and other variables. divided into main variable, estimated removed by partial regression coefficient of influence of the main variables were obtained by least squares method of external variables, by applying the principal component analysis on the resulting estimated differential T ² statistic or Q statistic A method of converting to a quantity and monitoring is disclosed. However, in this method, the monitoring threshold for the converted statistic does not consider the distribution characteristics of the statistic, for example, a value including 99% of the reference data (in other words, a value exceeding 1% of the reference data), It was determined by trial and error. The management threshold value is determined in this way. For example, if a large amount of noise enters the reference data even for a moment, the value exceeding 1% of the reference data (the value set as the management threshold value) becomes a high value. As in the case of Patent Document 6, there is a problem that the data dependence of the detection performance is large, such that the threshold value becomes high.

  As a technique that does not use principal component analysis, Patent Document 8 discloses a control chart in a format that is used in the conventional statistical process management with respect to an observation value of a process variable or an estimated deviation based on an ARIMA (Autogressive Integrated Moving Average) model. For example, a method is disclosed in which monitoring is performed by creating a chart of average values, standard deviations, etc., and comparing the chart with a management threshold value set in a normal state. However, the setting of the control threshold value of the control chart is a trial and error, and there is a problem that the data dependence of the detection performance is large. In addition, for example, when the plant data drifts due to a malfunction of the measuring instrument, the estimated deviation is not affected by the effect of the autoregressive term of the ARIMA model, so that the model is versatile for various abnormalities. There was a problem that was not.

  The present invention has been made in response to such a conventional situation, and even when the operation state of the plant is changed, if it is a normal change, it is not determined to be abnormal, and the management threshold for the monitoring index is not limited. To provide a highly versatile plant monitoring device that is less dependent on the monitoring performance to the reference data used to set the value, and that can be applied to various abnormalities from abnormal plant characteristics to drift in measuring instruments. Objective.

  A plant monitoring apparatus according to the present invention includes plant data input / storage means for inputting measurement data of input variables and output variables of a plant and storing them as time series data, and time series data stored by the plant data input / storage means. A monitoring model generating means for generating a monitoring model based on the monitoring data, and a monitoring processing means for applying the monitoring model to the measurement data to determine normality / abnormality of the plant state and outputting the determination result In the monitoring apparatus, the monitoring model creating means includes an output error model identifying unit that identifies an output error model using the time series data, and the output calculated by applying the output error model to the time series data. An estimated deviation characteristic evaluation unit that evaluates a statistical characteristic of the estimated deviation of the variable, and the monitoring processing means outputs the output to the measurement data. An output estimated deviation calculation unit that calculates an estimated deviation of the output variable by applying an error model, and an estimated deviation test processing unit that tests a variance value change hypothesis of the estimated deviation by applying a sequential probability ratio test to the estimated deviation And a hypothesis adoption rate determination unit that monitors plant abnormality based on the adoption rate of the estimated deviation variance change hypothesis.

  Further, the plant monitoring apparatus according to the present invention includes plant data input / storage means for inputting measurement data of plant input variables and output variables and storing them as time-series data, and when the data is stored by the plant data input / storage means Monitoring model creating means for creating a monitoring model based on series data; and monitoring processing means for applying the monitoring model to the measurement data to determine normality / abnormality of the plant state and outputting the determination result. The monitoring model creating means is configured to apply a principal component analysis to the time series data to construct a load matrix for a plurality of principal components having a large contribution rate, and to calculate the load matrix. An estimated deviation characteristic evaluation unit that evaluates a statistical characteristic of the estimated deviation calculated by applying to the time series data, and the monitoring processing means includes the An output estimated deviation unit that calculates an estimated deviation by applying the load matrix to constant data; an estimated deviation test processing unit that tests a variance value change hypothesis of the estimated deviation by applying a sequential probability ratio test to the estimated deviation; And a hypothesis adoption rate determination unit that monitors plant abnormality based on the adoption rate of the estimated deviation variance change hypothesis.

  Further, the plant monitoring apparatus according to the present invention includes plant data input / storage means for inputting measurement data of plant input variables and output variables and storing them as time-series data, and when the data is stored by the plant data input / storage means Monitoring model creating means for creating a monitoring model based on series data; and monitoring processing means for applying the monitoring model to the measurement data to determine normality / abnormality of the plant state and outputting the determination result. The monitoring model creating means is configured to apply a principal component analysis to the time series data to construct a load matrix for a plurality of principal components having a small contribution rate; and the load matrix A principal component score characteristic evaluation unit that evaluates a statistical characteristic of the principal component score calculated by applying to the time series data, the monitoring processing means, A principal component score calculation unit that calculates the principal component score by applying the load matrix to the measured data, and a principal value test for the variance value change hypothesis of the principal component score by applying a sequential probability ratio test to the principal component score A component score test processing unit; and a hypothesis adoption rate determination unit that monitors plant abnormality based on the adoption rate of the variance change hypothesis of the principal component score.

  Further, the plant monitoring apparatus according to the present invention includes plant data input / storage means for inputting measurement data of plant input variables and output variables and storing them as time-series data, and when the data is stored by the plant data input / storage means Monitoring model creating means for creating a monitoring model based on series data; and monitoring processing means for applying the monitoring model to the measurement data to determine normality / abnormality of the plant state and outputting the determination result. A first principal component analyzer configured to apply a principal component analysis to the time-series data to form a first load matrix for a plurality of principal components having a large contribution ratio. A second principal component analysis unit constituting a second load matrix for a plurality of principal components having a small contribution rate from the result of the principal component analysis, and the first load matrix An estimated deviation characteristic evaluation unit that evaluates a statistical characteristic of an estimated deviation calculated by applying to the time series data, and a statistical characteristic of a principal component score calculated by applying the second load matrix to the time series data A principal component score characteristic evaluation unit that evaluates the estimated deviation calculation unit that calculates an estimated deviation by applying the first load matrix to the measurement data; and A principal component score calculation unit that calculates a principal component score by applying a second load matrix; and an estimated deviation test processing unit that tests a variance value change hypothesis of the estimated deviation by applying a sequential probability ratio test to the estimated deviation A principal component score test processing unit that applies a sequential probability ratio test to the principal component score to test the variance value change hypothesis of the principal component score, and the adoption rate and principal component score of the variance deviation hypothesis of the estimated deviation Adoption of the variance change hypothesis Characterized in that it comprises a a hypothesis adoption rate determination unit for monitoring the abnormality of the plant based on.

  Further, the plant monitoring method according to the present invention includes a plant data input storage step for inputting measurement data of input variables and output variables of a plant and storing them as time series data, and time series data stored in the plant data input storage step. A monitoring model creating step for creating a monitoring model based on the above, and a monitoring processing step for applying the monitoring model to the measurement data to determine normality / abnormality of the plant state and outputting the determination result The monitoring model creation step includes an output error model identification step for identifying an output error model using the time series data, and the output variable calculated by applying the output error model to the time series data. An estimated deviation characteristic evaluation step for evaluating a statistical characteristic of the estimated deviation of the output deviation, The processing step includes an output estimated deviation calculating step of calculating the estimated deviation of the output variable by applying the output error model to the measurement data, and a variance value of the estimated deviation by applying a sequential probability ratio test to the estimated deviation. It comprises an estimated deviation test processing step for testing a change hypothesis, and a hypothesis adoption rate determination step for monitoring a plant abnormality based on the adoption rate of the variance change hypothesis.

  Further, the plant monitoring method according to the present invention includes a plant data input storage step for inputting measurement data of input variables and output variables of a plant and storing them as time series data, and time series data stored in the plant data input storage step. A monitoring model creating step for creating a monitoring model based on the above, and a monitoring processing step for applying the monitoring model to the measurement data to determine normality / abnormality of the plant state and outputting the determination result In the method, the monitoring model creation step includes applying a principal component analysis to the time series data to form a first load matrix for a plurality of principal components having a large contribution rate; Estimated deviation characteristic evaluation for evaluating a statistical characteristic of the estimated deviation calculated by applying the first load matrix to the time series data The output monitoring processing step includes an output estimated deviation step for calculating an estimated deviation by applying the load matrix to the measurement data, and a sequential probability ratio test is applied to the estimated deviation. An estimated deviation test processing step for testing a variance variation change hypothesis of the deviation, and a first hypothesis adoption rate determination step for monitoring a plant abnormality based on the adoption rate of the variance variation hypothesis of the estimated deviation. It is characterized by.

  According to the plant monitoring apparatus and the plant monitoring method according to the present invention, even if the state of the plant is changed due to a change in the operating condition of the plant, it is determined as abnormal if it is a normal change. There is nothing. That is, monitoring can be performed by correctly determining whether the state of the plant is normal or abnormal in consideration of changes in the plant state accompanying changes in the operating conditions of the plant.

  In addition, when setting management thresholds for monitoring indicators, they are set in consideration of statistical characteristics, so the dependency of the monitoring performance on the reference data is reduced, and the reliability of the monitoring results is improved. Can be blistered.

  Furthermore, since it can be applied to various abnormalities from abnormality of plant characteristics to drift of measuring instruments, a plant monitoring apparatus and a plant with higher versatility than the conventional plant monitoring apparatus and plant monitoring method. A monitoring method can be provided.

  Hereinafter, a plant monitoring device and a plant monitoring method according to the present invention will be described with reference to the drawings.

  FIG. 1 is a schematic diagram showing a basic configuration of a plant monitoring apparatus 1 which is an example of a plant monitoring apparatus according to the present invention.

  A plant monitoring apparatus 1 shown in FIG. 1 inputs plant data input for storing a variable that is an input of a change in the operating condition of the plant from a plant 2 to be monitored and a measurement signal of an output variable that changes under the influence of the variable. Storage means 11, monitoring model creation means 12 for creating a monitoring model prior to monitoring execution, and monitoring model creation means for input variable data and output variable data input by plant data input / storage means 11 after the start of monitoring execution And monitoring processing means 13 for detecting an abnormality by monitoring a change in the statistical characteristics of the estimated deviation of the output variable data obtained by applying the monitoring model created in 12 and outputting the processing result to the outside. Configured.

  Further, the plant monitoring apparatus 1 includes an input unit 4 that gives a user request to the plant data input / storage unit 11, the monitoring model creation unit 12, and the monitoring processing unit 13. For example, the reference shown in FIG. It is used when information required for processing execution such as processing request such as data setting (step S1) and setting range information is given to the processing execution means. The input means 4 is composed of, for example, a touch screen, a mouse and a keyboard.

  The plant data input / storage means 11 has a function of inputting, to the plant monitoring apparatus 1, a variable as an input of a change in plant operating conditions from a plant to be monitored and a measurement signal of an output variable that changes under the influence thereof. And a function of storing a signal captured using the input function as a measurement result. In the plant data input / storage means 11, a variable as an input of a change in plant operating conditions and a measurement signal of an output variable that changes under the influence are input continuously or at predetermined intervals from the plant to be monitored. The measurement results (time series data) of each variable are stored as history data. Henceforth, when referring to both an input variable and an output variable, it describes as an input / output variable.

  The monitoring model creating means 12 has a function of creating a monitoring model necessary for executing monitoring of the plant 2 and a function of storing the created monitoring model. In the monitoring model creation means 12, a monitoring model necessary for monitoring execution of the plant 2 is created and saved using the history data (time series data) of the input / output variables input by the plant data input / storage means 11. . The monitoring model creation function includes a function of accepting a user request necessary via the input unit 4 when the monitoring model is created and a function of executing processing necessary for creating the monitoring model according to the user request.

  The monitoring processing unit 13 calculates the estimated deviation of the output variable data obtained by applying the monitoring model stored in the monitoring model creating unit 12 to the input / output variable data input by the plant data input / storage unit 11 after the start of monitoring execution. A function that detects a state out of the normal range by monitoring a change in statistical characteristics, that is, an abnormal state, a function that receives information necessary for executing the monitoring process via the input unit 4, and the input information The function to execute processing according to The monitoring processing unit 13 detects the presence or absence of an abnormality in the plant 2 by applying the monitoring model stored in the monitoring model creating unit 12 to the input / output variable data input by the plant data input / storage unit 11 after the start of monitoring execution. Is done.

  The detection result (monitoring processing result) of whether there is an abnormality in the plant 2 is sent from the monitoring processing means 13 to the display device 3 which is an example of an external device connected to the plant monitoring device 1 and displayed on the display unit of the display device 3. Is done. The user can visually recognize whether there is an abnormality in the plant 2 by looking at the monitoring processing result displayed on the display unit of the display device 3.

  Next, the plant monitoring apparatuses 1A to 1F according to the respective embodiments of the present invention will be described. The plant monitoring apparatuses 1A to 1F described below are characterized by at least one of the monitoring model creating means 12 and the monitoring processing means 13 shown in FIG. 1, and the monitoring model creating means 12 in each of the plant monitoring apparatuses 1A to 1F. And at least one of the monitoring processing means 13 is different, but the other components are common to the respective embodiments. Therefore, in the description of the plant monitoring devices 1A to 1F and the plant monitoring method according to the following embodiments, the monitoring model creating unit 12 and the monitoring processing unit 13 will be mainly described, and the other components will be described as appropriate. Simplified or omitted.

[First Embodiment]
The configuration of the plant monitoring apparatus 1A according to the first embodiment of the present invention is the same as that shown in FIG. 1 except that the plant monitoring apparatus 1, the monitoring model creating means 12 and the monitoring processing means 13 are the plant monitoring apparatus 1A and the monitoring model creating respectively. The configuration is replaced with means 12A and monitoring processing means 13A. That is, the plant monitoring apparatus 1A includes a plant data input / storage unit 11, a monitoring model creation unit 12A, and a monitoring processing unit 13A.

  FIG. 2 is a schematic diagram showing the configuration of the monitoring model creating means 12A provided in the plant monitoring apparatus 1A according to the first embodiment of the present invention.

  As shown in FIG. 2, the monitoring model creation unit 12A includes an output error model identification unit 21 that identifies an output error model (Output error model) necessary for monitoring the plant 2, and the identified output. A first estimated deviation characteristic evaluation unit 22 that evaluates a statistical characteristic of the estimated deviation generated by the error model.

  FIG. 3 is a schematic diagram showing the configuration of the monitoring processing means 13A provided in the plant monitoring apparatus 1A.

  As shown in FIG. 3, the monitoring processing unit 13A applies an output model of the output variable by applying the monitoring model created by the monitoring model creating unit 12A to the input / output variable data (hereinafter, referred to as an output estimated deviation calculating unit). (Referred to as “first output estimated deviation calculation unit”) and the estimated deviation calculated by the first output estimated deviation calculation unit 25 is applied to the probability ratio test to change the variance of the estimated deviation. Estimated deviation test processing unit (hereinafter referred to as “first estimated deviation test processing unit”) 26 for testing the hypothesis of whether or not, and the adoption rate of the variance change hypothesis as a result of the sequential probability ratio test is calculated and adopted A hypothesis adoption rate determination unit 27 that detects an abnormality when the rate exceeds the management threshold.

  The processing contents executed by the plant monitoring apparatus 1A will be briefly described. First, the plant data input / storage means 11 takes in and stores measurement signals of input variables and output variables of the plant 2 to be monitored (plant data input / storage). Step). Here, in the plant data input / storage step, the time series data of the measurement results stored in the plant data input / storage means 11 is the history data.

  Following the plant data input / storing step, the monitoring model creating unit 12A creates and stores a monitoring model necessary for monitoring execution of the plant 2 using the history data stored in the plant data inputting / storing unit 11 (monitoring). Model creation / save step).

  After the monitoring model creating / storing step, the monitoring processing unit 13A applies the monitoring model stored in the monitoring model creating unit 12A to the input / output variable data input by the plant data input / storing unit 11, and the abnormality of the plant 2 is detected. Is detected (monitoring process step). The result of the monitoring processing means 13A detecting whether or not there is an abnormality in the plant 2 is sent from the monitoring processing means 13A to the display device 3 (monitoring processing result output step) and displayed on the display unit of the display device 3 (monitoring). Processing result display step).

  Next, the processing content (monitoring model creation / storing step) executed by the monitoring model creation unit 12A and the processing content (monitoring processing step) executed by the monitoring processing unit 13A will be described in more detail.

  FIG. 4 is a process flow diagram showing the processing contents (monitoring model creation / storing step) executed by the monitoring model creation means 12A.

  As shown in FIG. 4, the monitoring model creation / storing step includes a step of setting reference data (step S1), a step of identifying an output error model (step S2), and an output error model identified in step S2. (Step S3).

  In step S1 immediately after the start of the monitoring model creation / storing step, for example, based on prior knowledge about the plant 2 design information and the dynamic and static characteristics of the plant 2 such as the causal relationship between input variables and output variables, Defines an input variable u that directly represents a change in the operating conditions of the plant and an output variable z of the plant that changes under the influence thereof, sets these sets as reference data, and sets the reference data as an output error model. The data is divided into model creation data used in the identification unit 21 and characteristic evaluation data used in the first estimated deviation characteristic evaluation unit 22. The setting of the reference data is performed by the user inputting and setting from the input unit 4, and the setting information of the reference data is input from the input unit 4 to the output error model identification unit 21 and the first estimated deviation characteristic evaluation unit 22. Given to.

Subsequent to step S1, in step S2, the output error model identification unit 21 identifies an output error model defined by the following equation (1) using the model creation data.

In equation (1), B (q) is a transfer function defined by the following equation (2), e (k) is white noise, and k is time t _k .

In equation (2), q ⁻¹ is a time shift operator defined as q ⁻¹ u (k) ≡u (k−1). At this time, the order _{n b} and delay _{n d} from which to determine the parameters of the model by changing the following formula (3) with the calculated final prediction error (FPE: Final prediction error) the orders and delays to minimize, and Use a set of parameters.

In the above formula (3), N is the number of the reference data used to create the model, V is calculated to give the degree n _b and delay n _d and the parameter _{[b 1, ···, b nb} ] This represents the variance of the estimation error.

  When the output error model is identified in this way, following step S2, in step S3, the estimated deviation characteristic evaluation unit 22 applies the output error model identified in step S2 to the characteristic evaluation data. Then, an estimated value of the output variable z represented by the following equation (4) is calculated from the reference data of the input variable u.

  When the mean and variance values, which are statistical characteristic values of the estimated deviation dz of the output variable, are calculated in step S3, the monitoring model creation / storing step (steps S1 to S3) is completed after completing all the processing steps. Become.

  FIG. 5 is a processing flow diagram showing the processing contents (monitoring processing step) executed by the monitoring processing means 13A.

  As shown in FIG. 5, the monitoring processing step includes a step of setting monitoring data (step S11), a step of calculating an estimated deviation based on the output error model (step S12), and a sequential probability of a variance value of the estimated deviation dz. A step of performing a ratio test (step S13) and a step of monitoring a hypothesis acceptance rate (step S14) are included.

  In step S11 immediately after the start of the monitoring process step, the plant data input by the plant data input / save unit 11 after the start of the monitoring execution is used for monitoring under the same conditions as the reference data set in the monitoring model creating unit 12A. Input / output variable data (hereinafter referred to as “monitoring data”) is set. The setting of the monitoring data is performed by the user inputting and setting from the input unit 4, and the setting information of the monitoring data is given from the input unit 4 to the first estimation processing unit 25.

  When the monitoring data setting (step S11) is completed, in the subsequent step S12, the first estimation processing unit 25 outputs the output error model created by the monitoring model creating means 12A for the monitoring data set in step S11. Is applied to calculate the estimated deviation dz of the output variable.

  When the calculation of the estimated deviation of the output variable (step S12) is completed, in the subsequent step S13, the first estimated deviation test processing unit 26 performs a sequential probability ratio test (Sequential probability test) on the estimated deviation dz calculated in step S12. (ratio test) is applied, and it is determined whether or not the variance value of the estimated deviation has changed by the following method.

More specifically, the hypothesis that the time series of estimated deviations dZ _k = {dz (i), i = 1 to k} has the same characteristics as the characteristics of the normal state obtained by the first estimated deviation characteristic evaluation unit 22. H ₀ , hypothesis H ₁ that the variance value has increased with respect to the normal state, and hypothesis H ₂ that the variance value has decreased with respect to the normal state are defined by the following equation (5): Log-likelihood ratios are calculated as test indices λ _{1, k} and λ _{2, k} , and these test indices are given in advance from a false detection probability (False-alarm rate) α and a detection failure probability (Miss-alarm rate) β. The hypothesis is tested by comparing with threshold values A and B determined by the equation (7) described later.

When the estimated deviation dz is a normally distributed variable independent from each other, and this assumption is almost valid when the monitoring model is valid, λ _{i, k} (i = 1, 2) in the above equation (5) is It is transformed as the sequential formula (6).

When the determination of whether or not the variance value of the estimated deviation dz has changed (step S13) is completed, in step S14, the hypothesis acceptance rate determination unit 27 performs a variance change hypothesis H _i (i = _i) within a predetermined time width. (1) and (2) are calculated, and when the calculated hypothesis selection rate exceeds the management threshold, an abnormality is detected.

  When the estimated deviation dz is a mutually independent normal distribution variable, the adoption rate of the variance change hypothesis within a certain time width in a normal state follows a binomial distribution, and the total number of adoptions and the false detection probability α within the same time width. Determined by. Here, monitoring is performed by setting a normal misjudgment rate determined by considering the variance value of the binomial distribution with respect to the adoption rate of the variance change hypothesis as a management threshold value.

  When the step of monitoring the hypothesis acceptance rate (step S14) is completed, the monitoring processing steps (step S11 to step S14) are completed after completing all the processing steps.

  FIG. 6 is a diagram illustrating an example of a monitoring process result of the first plant monitoring apparatus 1A, and is an explanatory diagram for explaining how the plant 2 is monitored. More specifically, each of FIGS. 6A to 6E is a graph in which the horizontal axis indicates the number of steps, and the vertical axis indicates the output (measured value (solid line) 28 in FIG. 6A. , Estimated value (broken line) 29), input in FIG. 6B, estimated deviation in FIG. 6C, test index in FIG. 6D, and hypothesis acceptance rate in FIG. 6E. It is a thing. FIG. 7 is a partially enlarged view showing the range from 900 steps to 1100 steps in FIG.

  The monitoring result shown in FIG. 6 is an example of the result for the data when the 1st to 999th steps are in a normal state and the transfer gain of the plant is reduced by about 20% after the 1000th step.

The solid line shown in FIGS. 6 (a) and 7 is the measured value 28, and the broken line is the estimated value 29 obtained by the equation (4) by applying the input variable shown in FIG. 6 (b) to the monitoring model. . 6A and 7, no significant change is seen before and after the 1000th step even when the measured value 29 is seen, but as shown in FIG. 6C, the estimated deviation is around 1000 steps. From FIG. 6D, the test index λ _{1, k} for the variance increase hypothesis H ₁ has a threshold A indicated by a broken line of 1000 It is frequently exceeded after the first step. Here, both the false detection probability α and the detection failure probability β are set to 0.1%.

Hypothesis acceptance rate shown in FIG. 6 (e), 400 is a selection rate of the dispersion increases hypothesis H ₁ which is calculated by giving the time width of the step, according to FIG. 6 (e), the in FIG even hypothesis acceptance rate The indicated management threshold value is greatly exceeded after the 1000th point, and an abnormality is detected.

  According to the present embodiment, when the output variable changes because the operating condition of the plant has changed, no change appears in the estimated deviation by the monitoring model, so that it is not erroneously determined to be abnormal. In addition, since an output error model is used for the monitoring model, even if a drift occurs in the measurement signal of the output variable, the estimated deviation changes, so that it can be detected correctly, thereby detecting various abnormalities. It is applicable to.

  Furthermore, since the management threshold for the monitoring index is set in consideration of statistical characteristics, the dependency on the reference data is small. As for the result of the sequential probability ratio test, the variance change hypothesis is adopted at a constant rate in the normal state corresponding to the set false detection rate, but the statistical characteristics are taken into account for the adoption rate within a certain time range. Since the management threshold is set, no erroneous detection due to transient noise occurs, and the reliability of the monitoring result is improved.

[Second Embodiment]
A plant monitoring apparatus and a plant monitoring method according to the second embodiment of the present invention will be described. In addition, the same code | symbol is attached | subjected to the component same as the plant monitoring apparatus 1B, and the overlapping description is abbreviate | omitted.

  The configuration of the plant monitoring apparatus 1B according to the second embodiment of the present invention is the same as that shown in FIG. 1 except that the plant monitoring apparatus 1, the monitoring model creating means 12 and the monitoring processing means 13 are respectively the plant monitoring apparatus 1B and the monitoring model creating. The configuration is replaced with the means 12B and the monitoring processing means 13B. That is, the plant monitoring apparatus 1B includes a plant data input / storage unit 11, a monitoring model creation unit 12B, and a monitoring processing unit 13B.

  FIG. 8 is a schematic diagram showing the configuration of the monitoring model creating means 12B included in the plant monitoring apparatus (hereinafter referred to as “second monitoring model creating means”) 1B according to the second embodiment of the present invention. It is.

  As shown in FIG. 8, the second monitoring model creating means 12B performs principal component analysis and displays up to r principal components whose cumulative contribution rate from the largest eigenvalue is equal to or greater than a preset value. A first principal component analyzing unit 31 that creates a load matrix having selected r eigenvectors in a column as a monitoring model, and an estimated deviation generated by the monitoring model created by the first principal component analyzing unit 31 And a second estimated deviation characteristic evaluation unit 32 for evaluating the characteristic.

  FIG. 9 is a schematic diagram showing the configuration of the monitoring processing means 13B (hereinafter referred to as “second monitoring processing means”) provided in the plant monitoring apparatus 1B.

  The second monitoring processing means 13B applies the monitoring model created by the monitoring model creating means 12B to the input / output variable data to calculate the estimated deviation of the monitoring variable corresponding to the input / output variable. A calculating unit 35; a second estimated deviation test processing unit 36 for applying a sequential probability ratio test to the estimated deviation to test a hypothesis as to whether or not a variance value of the estimated deviation has changed; and the sequential probability ratio A hypothesis adoption rate determination unit 27 that calculates an acceptance rate of the variance change hypothesis as a result of the test and detects an abnormality by determining that the acceptance rate exceeds a management threshold value as an abnormality.

  The processing content executed by the plant monitoring device 1B is the same as the processing content executed by the plant monitoring device 1A, although the processing model created and applied is different, but the essential processing steps are the same. That is, in the plant monitoring apparatus 1B, first, a plant data input / storage step is executed, and subsequently, a monitoring model creation / storage step is executed. In this processing step, a history stored in the plant data input / storage means 11 A load matrix serving as a monitoring model necessary for the monitoring execution of the plant 2 is created and stored by the second monitoring model creating means 12B using the data.

  Following the monitoring model creation / storage step, a monitoring processing step is executed, and in this processing step, a plant to which the monitoring model stored in the second monitoring model creation means 12B is applied by the second monitoring processing means 13B. Two abnormalities are detected. After the monitoring processing step, the monitoring processing result output step and the monitoring processing result display step are executed, and the result of the second monitoring processing means 13B detecting whether there is an abnormality in the plant 2 is displayed on the display unit of the display device 3. Is done.

  Next, the processing contents executed by the second monitoring model creation means 12B (hereinafter referred to as “second monitoring model creation / storage step”) and the processing contents executed by the second monitoring processing means 13B ( The “second monitoring process step” will be described in more detail.

  FIG. 10 is a process flow diagram showing the second monitoring model creation / storing step executed by the second monitoring model creation means 12B.

  As shown in FIG. 10, the second monitoring model creation / storing step includes a step of setting reference data (step S21), a step of performing principal component analysis (step S22), and a load on the principal component having a large contribution rate. A step of forming a matrix (step S23), a step of calculating an estimated deviation when the load matrix (monitoring model) configured in step S23 is applied (step S24), and a characteristic of the estimated deviation calculated in step S24 And a step of performing an evaluation (step S25).

  In the step S21 immediately after the start of the second monitoring model creation / storing step, the change in the operating condition of the plant is directly expressed in the history data based on the prior knowledge regarding the dynamic and static characteristics of the plant 2. Defining a monitoring variable x including a variable and a variable that is affected by the variable, setting these sets as reference data, and data for creating a model for using the reference data in the first principal component analysis unit 31; This is divided into characteristic evaluation data X used in the estimated deviation characteristic evaluation unit 32. The setting of the reference data is performed by the user inputting and setting from the input unit 4, and the setting information of the reference data is output from the input unit 4 to the output first principal component analysis unit 31 and the second estimated deviation characteristic. It is given to the evaluation unit 32.

  Subsequent to step S21, in step S22, the first principal component analysis unit 31 performs principal component analysis using the model creation data. In step S23, the first principal component analysis unit 31 determines that the cumulative contribution ratio from the largest eigenvalue, that is, the ratio of the value obtained by adding eigenvalues in order to the total sum of all eigenvalues is equal to or greater than a preset value. Up to r principal components, and a load matrix A having r corresponding eigenvectors in a column is constructed. The load matrix A constructed in step S23 becomes a monitoring model applied by the plant monitoring apparatus 1B.

  Subsequent to step S24, in step S25, the second estimated deviation characteristic evaluation unit 32 calculates an average value and a variance value, which are statistical characteristic values of the estimated deviation calculated in step S24. These characteristic values are used for the sequential probability ratio test processing of the second monitoring processing means 13B. When the mean and variance values, which are statistical characteristic values of the estimated deviation dX, are calculated in step S25, the second monitoring model creation / storing step (steps S21 to S25) completes all processing steps and ends. Become.

  FIG. 11 is a process flow diagram showing the second monitoring processing step executed by the second monitoring processing means 13B.

  As shown in FIG. 11, the second monitoring processing step includes a step of setting monitoring data (step S31), a step of calculating an estimated deviation based on a load matrix (step S32), and a variance value of the estimated deviation dX. A step of performing a sequential probability ratio test (step S33) and a step of monitoring a hypothesis acceptance rate (step S14).

  In step S31 immediately after the start of the monitoring process step, among the plant data input by the plant data input / save unit 11 after the start of the monitoring execution, the monitoring data for the same monitoring variable for which the reference data is set in the monitoring model creation unit 2 Set. The setting of the monitoring data is performed by the user inputting and setting from the input unit 4, and the setting information of the monitoring data is given from the input unit 4 to the second estimation processing unit 35.

  When the monitoring data setting (step S31) is completed, in the subsequent step S32, the second output estimation calculation unit 35 creates the monitoring model (load matrix) created by the second monitoring model creation means 12B for the monitoring data. ) To calculate the estimated deviation dX of the monitored variable.

  When the calculation of the estimated deviation of the monitoring variable (step S32) is completed, in the subsequent step S33, the second estimated deviation test processing unit 36 applies the sequential probability ratio test to the estimated deviation dX calculated in step S32. Then, it is determined whether or not the variance value of the estimated deviation has changed. The determination method is substantially the same as in step S13.

When the determination of whether or not the variance value of the estimated deviation dX has changed (step S33) is completed, in step S14, the hypothesis acceptance rate determination unit 27 performs a variance change hypothesis H _i (i = _i) within a predetermined time width. (1) and (2) are calculated, and when the calculated hypothesis selection rate exceeds the management threshold, an abnormality is detected. When the step of monitoring the hypothesis acceptance rate (step S14) is completed, the second monitoring processing steps (step S31 to step S14) are completed after completing all the processing steps.

  FIG. 12 is an explanatory diagram for explaining how the second plant monitoring device 1B monitors the plant 2. FIG. 6 shows an example of the monitoring processing result of the first plant monitoring device 1A. It is an equivalent figure.

  More specifically, each of FIGS. 12A to 12E is a graph in which the horizontal axis indicates the number of steps, and the vertical axis indicates the change in operating conditions of the plant 2 in FIG. The variable z corresponding to the output that changes due to the influence of FIG. 12B, the variable u corresponding to the input that directly represents the change in the operating condition of the plant 2 in FIG. 12B, the estimated deviation in FIG. In FIG. 12 (d), it is expressed as a test index, and in FIG. 12 (e), it is expressed as a hypothesis acceptance rate.

In the example shown in FIG.

  In the case of the example shown in FIG. 12, the cumulative contribution ratio of the two main components is 99% or more. 12C to 12E, a large change appears in the estimated deviation dz of the variable z corresponding to the output variable of the plant 2 after the 1000th step when the abnormality occurs. In particular, the test index for the variance increase hypothesis of the estimated deviation dz shown in FIG. 12D shows a very large value after the 1000th step. The hypothesis adoption rate shown in FIG. 12 (e) is the adoption rate of the variance increase hypothesis calculated by giving a time width of 400 steps, and the management threshold indicated by a broken line in the figure is the 1000th step and after. In this case, an abnormality is detected.

  According to the present embodiment, when the output variable changes due to a change in the operating conditions of the plant, the estimation by the monitoring model is performed well, so the estimated deviation does not appear, and thus it is erroneously determined as abnormal. Never do.

  In addition, by constructing the monitoring variables from variables corresponding to plant input / output variables, if the measurement signal of the variable corresponding to the output variable drifts, the estimated deviation will change. Can do. Thus, it can be applied to various abnormalities.

  Furthermore, since the management threshold for the monitoring index is set in consideration of statistical characteristics, the dependency of the abnormality detection performance on the reference data is small. In addition, the adoption rate within a certain time range is obtained from the results of the sequential probability ratio test, and monitoring is performed by setting a management threshold considering its statistical characteristics. The monitoring results are highly reliable. As described above, also in this embodiment, the same effect as that of the first embodiment can be obtained.

[Third Embodiment]
A plant monitoring apparatus and a plant monitoring method according to a third embodiment of the present invention will be described. In addition, the same code | symbol is attached | subjected to the component same as plant monitoring apparatus 1A, 1B, and the overlapping description is abbreviate | omitted.

  The configuration of the plant monitoring apparatus 1C according to the third embodiment of the present invention is the same as that shown in FIG. 1 except that the plant monitoring apparatus 1, the monitoring model creating means 12 and the monitoring processing means 13 are respectively the plant monitoring apparatus 1C and the monitoring model creating. The configuration is replaced with the means 12C and the monitoring processing means 13C. That is, the plant monitoring apparatus 1C includes a plant data input / storage unit 11, a monitoring model creation unit 12C, and a monitoring processing unit 13C.

  FIG. 13 is a schematic diagram showing the configuration of a monitoring model creation means (hereinafter referred to as “third monitoring model creation means”) 12C included in the plant monitoring apparatus 1C according to the third embodiment of the present invention. It is.

  As shown in FIG. 13, the monitoring model creation means 12C performs principal component analysis, selects up to s principal components whose cumulative contribution rate from the smallest eigenvalue is equal to or less than a preset value, A second principal component analysis unit 41 that creates a load matrix having columns of corresponding s eigenvectors as a monitoring model, and characteristics of principal component scores generated by the monitoring model created by the second principal component analysis unit 41 A principal component score characteristic evaluation unit 42 that performs evaluation is provided.

  FIG. 14 is a schematic diagram showing the configuration of the monitoring processing means 13C (hereinafter referred to as “third monitoring processing means”) provided in the plant monitoring apparatus 1C.

  The monitoring processing unit 13C applies the monitoring model created by the monitoring model creation unit 12C to the monitoring variable data to calculate the principal component score, and the main component score calculation unit 45 calculates the principal component score. Principal component score test processing unit 46 that tests the hypothesis as to whether or not the variance value of the principal component score has changed by applying a sequential probability ratio test to the component score, and the variance change of the result of the sequential probability ratio test A hypothesis adoption rate determination unit 27 that detects the abnormality by calculating the hypothesis adoption rate and determining that the state in which the acceptance rate exceeds the management threshold value is abnormal.

  The processing contents executed by the plant monitoring apparatus 1C are the same as the processing contents executed by the plant monitoring apparatus 1B, although the processing models created and applied are different, but the essential processing steps are the same. That is, in the plant monitoring apparatus 1C, first, a plant data input / storing step is executed, and subsequently, a monitoring model creation / storing step is executed. In this processing step, a history stored in the plant data input / storing means 11 Using the data, a load matrix serving as a monitoring model necessary for the monitoring execution of the plant 2 is created and stored by the third monitoring model creating means 12C.

  Subsequent to the monitoring model creation / storage step, a monitoring processing step is executed. In this processing step, a plant to which the monitoring model stored in the third monitoring model creation means 12C is applied by the third monitoring processing means 13C. Two abnormalities are detected. After the monitoring processing step, a monitoring processing result output step and a monitoring processing result display step are executed, and the result of detection of the presence or absence of abnormality of the plant 2 by the third monitoring processing means 13C is displayed on the display unit of the display device 3. Is done.

  Next, the processing contents executed by the third monitoring model creation means 12C (hereinafter referred to as “third monitoring model creation / storing step”) and the processing contents executed by the third monitoring processing means 13C ( The “third monitoring processing step” will be described in more detail.

  FIG. 15 is a process flow diagram showing a third monitoring model creation / storing step executed by the third monitoring model creation means 12C.

  As shown in FIG. 15, the third monitoring model creation / storing step includes a step of setting reference data (step S21), a step of performing principal component analysis (step S22), and a load on a principal component having a small contribution rate. A step of constructing a matrix (step S43), a step of calculating principal component scores when the load matrix (monitoring model) constructed in step S43 is applied (step S44), and a principal component score calculated in step S44 (Step S45).

  In the third monitoring model creation / storing step, first, similarly to the second monitoring model creating / storing step, steps S21 and S22 are executed. When the principal component analysis of step S22 is completed, in the subsequent step S43, the second principal component analysis unit 41, based on the result of the principal component analysis performed in step S22, the cumulative contribution rate from the smallest eigenvalue. Select up to s principal components that are equal to or less than a preset value, and construct a load matrix A having corresponding s eigenvectors in columns. The load matrix A constructed in step S43 is a monitoring model applied by the plant monitoring apparatus 1C.

Subsequent to step S43, in step S44, the principal component score test processing unit 46 applies the load matrix A constructed in step S43 to the characteristic evaluation data X set in step S21. The given principal component score T is calculated.

  Subsequent to step S44, in step S45, the principal component score test processing unit 46 calculates an average value and a variance value which are statistical characteristic values of the principal component score T calculated in step S44. These characteristic values are used for the sequential probability ratio test processing of the third monitoring processing means 13C.

When the mean and variance values, which are statistical characteristic values of the principal component score T, are calculated in step S45, the third monitoring model creation / storing step (steps S21 to S45) is completed after completing all the processing steps. It becomes.

  FIG. 16 is a process flow diagram showing the third monitoring processing step executed by the third monitoring processing means 13C.

  As shown in FIG. 16, the third monitoring processing step includes a step of setting monitoring data (step S31), a step of calculating principal component scores based on the load matrix (step S52), and a principal component score variance value. A step of performing a sequential probability ratio test (step S53) and a step of monitoring a hypothesis acceptance rate (step S14).

  In the third monitoring process step, step S31 is first executed as in the second monitoring process step. When the monitoring data setting (step S31) is completed, in the subsequent step S52, the principal component score calculation unit 45 applies the load matrix created by the monitoring model creation means 12C to the monitoring data to obtain the principal component score. calculate.

  When the calculation of the principal component score using the load matrix (step S52) ends, in the subsequent step S53, the principal component score test processing unit 46 applies the sequential probability ratio test to the principal component score calculated in step S52, It is determined whether or not the variance value of the principal component score has changed. When the determination of whether or not the variance of the principal component score has changed (step S53) is completed, step S14 is subsequently executed. When step S14 is completed, the third monitoring processing steps (steps S31 to S14) are performed. Then, all processing steps are completed and the process ends.

  Whereas the second plant monitoring device 1B detects that a new component has occurred due to the influence of an abnormality, which cannot be explained by the correlation between variables during normal operation, the third plant monitoring device 1C It can be said that it detects that the correlation component, which was so small that it was hardly recognized at normal time, increased due to the occurrence of an abnormality.

  FIG. 17 is an explanatory diagram for explaining how the third plant monitoring device 1C monitors the plant 2, and FIG. 6 shows an example of the monitoring processing result of the first plant monitoring device 1A. It is a figure equivalent to FIG. 12 which showed an example of the monitoring process result of the 2nd plant monitoring apparatus 1B.

  More specifically, each of the diagrams of FIGS. 17A to 17E is a graph in which the horizontal axis indicates the number of steps, and the vertical axis indicates the change in the operating condition of the plant 2 in FIG. 17 is a variable z corresponding to the output that changes under the influence of FIG. 17B, a variable u corresponding to the input that directly represents a change in the operating condition of the plant 2 in FIG. 17B, a principal component score in FIG. In (d), it is set as a test index, and in FIG. 17 (e), it is expressed as a hypothesis acceptance rate.

  Note that the monitoring variable x is an example configured by Expression (9) as in the case of FIG. 12, and a load matrix is configured from the eigenvector corresponding to the minimum eigenvalue. In the example shown in FIG. 12, the cumulative contribution ratio of the two principal components is 99% or more. However, in the example shown in FIG. 17, the cumulative contribution ratio is 1% or less. ) And the estimated value 29 shown in FIG. 17 (b) are significantly different from (deviated from) the measured value 28.

  The principal component score T shown in FIG. 17 (c) shows a large change after the 1000th step when the abnormality occurred, and the test index for the variance increase hypothesis of the principal component score shown in FIG. Very large values are shown after the step. Further, the hypothesis adoption rate shown in FIG. 17 (e) is the adoption rate of the variance increase hypothesis calculated by giving a time width of 400 steps, and the management threshold indicated by the broken line in the figure is 1000 steps. This is greatly exceeded after the first eye, and an abnormality is detected.

  According to the present embodiment, by configuring the monitoring variable by a variable corresponding to the input / output variable of the plant, a change appears in the estimated deviation even when a drift occurs in the measurement signal of the variable corresponding to the output variable. It can be detected correctly and can be applied to various abnormalities.

  In addition, since the management threshold for the monitoring index is set in consideration of statistical characteristics, the dependence on the reference data is small. In addition, the adoption rate within a certain time range is obtained from the results of the sequential probability ratio test, and monitoring is performed by setting a management threshold considering its statistical characteristics. The monitoring results are highly reliable. As described above, also in this embodiment, the same effect as that of the first embodiment can be obtained.

[Fourth Embodiment]
A plant monitoring apparatus and a plant monitoring method according to the fourth embodiment of the present invention will be described. In addition, the same code | symbol is attached | subjected to the component same as plant monitoring apparatus 1A, 1B, 1C, and the overlapping description is abbreviate | omitted.

  The configuration of the plant monitoring apparatus 1D according to the fourth embodiment of the present invention is the same as that shown in FIG. 1 except that the plant monitoring apparatus 1, the monitoring model creating means 12, and the monitoring processing means 13 are the plant monitoring apparatus 1D and the monitoring model creating respectively. The configuration is replaced with the means 12D and the monitoring processing means 13D. That is, the plant monitoring apparatus 1D includes a plant data input / storage unit 11, a monitoring model creation unit 12D, and a monitoring processing unit 13D.

  FIG. 18 is a schematic diagram showing the configuration of a monitoring model creation means (hereinafter referred to as “fourth monitoring model creation means”) 12D provided in the plant monitoring apparatus 1D according to the fourth embodiment of the present invention. It is.

  As shown in FIG. 18, the fourth monitoring model creation unit 12D includes a first principal component analysis unit 51 including a first principal component analysis unit 31 and a second principal component analysis unit 41, And a statistical data characteristic evaluation 52 including an estimated deviation and a principal component score generated by the monitoring model created by the principal component analysis unit 51.

  The statistic data characteristic evaluation 52 has the functions of the second estimated deviation characteristic evaluation part 32 and the principal component score characteristic evaluation part 42, and is generated by the monitoring model created by the first principal component analysis part 31. The characteristic evaluation of the estimated deviation of the monitoring variable is the function of the second estimated deviation characteristic evaluation unit 32, and the characteristic evaluation of the principal component score generated by the monitoring model created by the second principal component analysis unit 41 is mainly used. Characteristic evaluation is performed using the function of the component score characteristic evaluation unit 42.

  FIG. 19 is a schematic diagram showing the configuration of the monitoring processing means 13D (hereinafter referred to as “fourth monitoring processing means”) provided in the plant monitoring apparatus 1D.

  As shown in FIG. 19, the fourth monitoring processing unit 13D applies the load matrix, which is the monitoring model created by the monitoring model creation unit 12D, to the input / output variable data, and the estimated deviations and principal components of the input / output variables. A statistical data calculation unit 55 for calculating scores, and a probability ratio test applied to the statistical data composed of the estimated deviation and principal component scores calculated by the statistical data calculation unit 55 to distribute the statistical data A statistical data test processing unit 56 for testing a hypothesis as to whether or not the value has changed, and a rate of adoption of a variance change hypothesis as a result of the sequential probability ratio test, and the rate of adoption exceeds a management threshold And a hypothesis adoption rate determination unit 27 that detects an abnormality by determining that it is abnormal.

  The statistic data calculation unit 55 has the functions of the second output estimation deviation calculation unit 35 and the principal component score calculation unit 45, and is estimated by applying the monitoring model created by the first principal component analysis unit 31. For the calculation of the deviation, the function of the second output estimated deviation calculation unit 35 is applied, and for the calculation of the principal component score by applying the monitoring model created by the second principal component analysis unit 41, the principal component score calculation unit 45 The calculation is performed using the function.

  The statistic data test processing unit 56 has the functions of the second estimated deviation test processing unit 36 and the principal component score test processing unit 46, and the second estimated deviation test processing unit for testing the estimated deviation calculated. The function of 36 is calculated by using the function of the principal component score test processing unit 46 for the calculation of the calculated principal component score.

  The processing contents executed by the plant monitoring device 1D are first executed by the plant data input / storing step by the plant data input / storing means 11 in the same manner as the processings executed by the other plant monitoring devices 1A, 1B, 1C, etc. Subsequently, the monitoring model creation / storage step by the fourth monitoring model creation means 12D is executed, and then the monitoring processing step by the fourth monitoring processing means 13D is executed. After the monitoring process step, the monitoring process result output step and the monitoring process result display step are executed, and the result of the fourth monitoring processing unit 13D detecting whether or not there is an abnormality in the plant 2 is the display unit of the display device 3. Is displayed.

  Further, the processing content executed by the plant monitoring device 1D can be considered as a step in which some processing steps are combined with the processing content executed by the plant monitoring devices 1B and 1C. That is, in the processing contents executed by the fourth monitoring model creating means 12D (hereinafter referred to as “fourth monitoring model creating / saving step”), first, the reference data is set by the principal component analysis unit 51. S21, step S22 for principal component analysis, step S23 for constructing a load matrix for a principal component with a large contribution ratio, and step S43 for constructing a load matrix for a principal component with a small contribution ratio are executed. Subsequently, as a characteristic evaluation of the statistic data, the statistic data characteristic evaluation unit 52 calculates the estimated deviation when the load matrix (monitoring model) configured in step S23 is applied, and is calculated in steps S24 and S24. Characteristic evaluation of the principal component score calculated in step S44 and step S44 for calculating the principal component score when the load matrix (monitoring model) configured in step S25 and step S43 for evaluating the characteristic of the estimated deviation is applied. Step S45 is executed.

  When the principal component analysis unit 51 and the statistic data characteristic evaluation unit 52 complete a series of processing steps, the fourth monitoring model creation / storage step ends. In other words, the fourth monitoring model creation / storing step includes a second monitoring model creating / storing step and a third monitoring model creating / storing step. This is a processing step for separately processing the monitoring model creation / storage step.

  In the processing contents (referred to as “fourth monitoring processing step”) executed by the fourth monitoring processing means 13D, first, the statistical data calculation unit 55 sets monitoring data by step S31, estimation by load matrix. Step S32 for calculating the deviation and step S52 for calculating the principal component score by the load matrix are executed. Subsequently, step S33 for performing the sequential probability ratio test of the variance value of the estimated deviation dX and the step S53 of performing the sequential probability ratio test of the principal component score variance value are executed by the statistical data test processing unit 56. Subsequently, the hypothesis adoption rate determination unit 27 calculates the adoption rate of the variance change hypothesis as a result of the sequential probability ratio test for the statistical data performed by the statistical data test processing unit 56, and the calculated hypothesis adoption rate is managed. Abnormality is detected by judging that it is abnormal when the value is exceeded. When the step of monitoring the hypothesis acceptance rate (step S14) is finished, the fourth monitoring processing step is completed after completing all the processing steps.

  As described above, according to the present embodiment, the same effects as those of the plant monitoring apparatuses 1A, 1B, 1C and the respective plant monitoring methods described in the first to third embodiments can be obtained, and between normal variables. High versatility applicable to both a wide range of abnormalities that cause new components that cannot be explained by the correlation, and types of abnormalities that increase weak correlation components that exist even during normal operation Monitoring becomes possible.

[Fifth Embodiment]
A plant monitoring apparatus and a plant monitoring method according to the fifth embodiment of the present invention will be described. In addition, the same code | symbol is attached | subjected to the component same as plant monitoring apparatus 1A, 1B, 1C, 1D, and the overlapping description is abbreviate | omitted.

  The configuration of the plant monitoring apparatus 1E according to the fifth embodiment of the present invention is the same as that shown in FIG. 1, except that the plant monitoring apparatus 1, the monitoring model creating means 12 and the monitoring processing means 13 are respectively the plant monitoring apparatus 1E and the monitoring model creating. The configuration is replaced with the means 12E and the monitoring processing means 13E. That is, the plant monitoring apparatus 1E includes a plant data input / storage unit 11, a monitoring model creation unit 12E, and a monitoring processing unit 13E.

  FIG. 20 shows an outline of a plant monitoring apparatus 1E according to the fifth embodiment of the present invention, and in particular, a monitoring model creating means (hereinafter referred to as “fifth monitoring model creating means”) provided in the plant monitoring apparatus 1E. It is a schematic diagram showing the configuration of 12E in detail.

  As shown in FIG. 20, in addition to the output error model identifying unit 21 and the first estimated deviation characteristic evaluating unit 22, the fifth monitoring model creating unit 12E directly causes the change in the operating state of the plant 2. An input variable characteristic evaluation unit 58 for evaluating variable characteristics is further provided. That is, the fifth monitoring model creation unit 12E is configured by further including an input variable characteristic evaluation unit 58 in addition to the first monitoring model creation unit 12A.

  FIG. 21 shows an outline of the plant monitoring apparatus 1E, and in particular, an outline showing in detail the configuration of the monitoring processing means 13E (hereinafter referred to as “fifth monitoring processing means”) provided in the plant monitoring apparatus 1E. FIG.

As shown in FIG. 21, the fifth monitoring processing unit 13E includes the first monitoring processing unit 13A and an input abnormality monitoring processing unit 60 that monitors whether there is an abnormality in the input variable characteristics.

  The processing content executed by the plant monitoring device 1E is similar to the processing content executed by the other plant monitoring devices 1A, 1B, 1C, 1D, etc. First, the plant data input / storage step by the plant data input / storage means 11 is performed. Then, the monitoring model creation / storage step by the fifth monitoring model creation means 12E is executed, and then the monitoring processing step by the fifth monitoring processing means 13E is executed. Then, after the monitoring process step, the monitoring process result output step and the monitoring process result display step are executed, and the result of the fifth monitoring processing unit 13E detecting whether there is an abnormality in the plant 2 is the display unit of the display device 3. Is displayed.

  Of the monitoring processing steps by the fifth monitoring processing means 13E, the processing contents performed by the first monitoring processing means 13A are as described in the first embodiment. In the plant monitoring apparatus 1E, the value of the output variable of the plant 2 is estimated from the measured value of the input variable of the plant 2 by the output error model, and the deviation between the measured value of the output variable and the estimated value is monitored. Monitor changes in output variables that cannot be explained by changes. That is, in the first monitoring processing unit 13A, as long as the change in the input / output variable follows the output error model, no abnormality is detected regardless of the change in the input variable.

On the other hand, the input abnormality monitoring processing means 60 monitors whether or not the variance value of the input variable within the predetermined time width is the same as the variance value in the normal state by a variance test. The variance test is performed as follows.

  FIG. 22 is a diagram showing an example of the input abnormality monitoring processing result of the fifth plant monitoring device 1E, and shows an example applied to data when the variance value of the input variable changes.

  22A is a graph showing the measured values of the input variables, FIG. 22B is a graph showing the dispersion monitoring index calculated by the equation (11), and FIG. 22C is the adoption rate of the dispersion change hypothesis. That is, the abnormality detection rate. In FIG. 22C, n in the equation (12) is 200 steps, and the time width for calculating the abnormality detection rate is 400 steps. According to FIG. 22, the abnormality detection rate starts to change greatly after the 1000th step when an abnormality has occurred, and an abnormality is detected when the management threshold value indicated by the broken line is exceeded.

  As described above, according to the present embodiment, in addition to the same effects as the first embodiment, it is also possible to detect abnormal changes in plant input variables that cannot be detected in the first embodiment, The scope of application can be further expanded.

[Sixth Embodiment]
A plant monitoring apparatus and a plant monitoring method according to a sixth embodiment of the present invention will be described. In addition, the same code | symbol is attached | subjected to the component same as plant monitoring apparatus 1A, 1B, 1C, 1D, 1E, and the overlapping description is abbreviate | omitted.

  The configuration of the plant monitoring device 1F according to the sixth embodiment of the present invention is the same as that shown in FIG. 1 except that the plant monitoring device 1, the monitoring model creation means 12, and the monitoring processing means 13 are the plant monitoring device 1F and the monitoring model creation, respectively. The configuration is replaced with the means 12F and the monitoring processing means 13F. That is, the plant monitoring apparatus 1F includes a plant data input / storage unit 11, a monitoring model creation unit 12F, and a monitoring processing unit 13F.

  FIG. 23 shows a plant monitoring apparatus 1F according to the sixth embodiment of the present invention, a monitoring model creation means (hereinafter referred to as “sixth monitoring model creation means”) 12F included in the plant monitoring apparatus 1F, and It is the schematic which showed the structure of the monitoring process means 13F (henceforth "the 6th monitoring process means").

  As shown in FIG. 23, the sixth monitoring model creation unit 12F includes a principal component analysis unit 51, a statistic data characteristic evaluation unit 52, and an input variable characteristic evaluation unit 58. The sixth monitoring processing unit 13F includes a fourth monitoring processing unit 13D and an input abnormality monitoring processing unit 60.

  The processing content executed by the plant monitoring device 1F is the same as the processing content executed by the other plant monitoring devices 1A, 1B, 1C, 1D, 1E, etc. A storage step is executed, followed by a monitoring model creation / storing step (hereinafter referred to as a “sixth monitoring model creation / storing step”) by the sixth monitoring model creation means 12F, followed by a sixth monitoring. A monitoring process step (referred to as “sixth monitoring process step”) by the processing means 13F is executed.

  After the sixth monitoring processing step is executed, the monitoring processing result output step and the monitoring processing result display step are executed, and the result of the sixth monitoring processing means 13F detecting whether there is an abnormality in the plant 2 is displayed. It is displayed on the display unit of the device 3.

  Of the sixth monitoring model creation / storage step, the processing contents executed by the principal component analysis means 51 and the statistic data characteristic evaluation unit 52 are the same as the fourth monitoring model creation / storage step. is there. Further, the processing content executed by the input variable characteristic evaluation unit 58 is processing content executed in the same manner as the monitoring model creation / storage step by the fifth monitoring model creation means 12E.

  On the other hand, in the sixth monitoring processing step, the processing content executed by the fourth monitoring processing means 13D is the processing content executed in the same manner as the fourth monitoring processing step, and the input abnormality monitoring processing means 60 As to the processing content to be executed, as described in the fifth monitoring processing step, whether or not the variance value of the input variable within the predetermined time width is the same as the variance value in the normal state is monitored by variance test.

  As described above, according to this embodiment, in addition to the same effects as the fourth embodiment, it is also possible to detect abnormal changes in plant input variables that cannot be detected by the fourth embodiment, The scope of application can be further expanded.

Further, in the second embodiment and the third embodiment described above, as an object of principal component analysis, the measured values at different times of the plant input variable u and the output variable z are expressed as shown in Equation (9). In the plant where the configured monitoring variable x is used but the input / output relationship is not necessarily clear, all are regarded as the output variable z, and the monitoring variable is configured as shown in the following formula (13) or (14). May be.
[Formula 14]
x (t) = z (t) (13)
Or x (t) = [z (t−1) z (t)] (14)

Furthermore, when the input time interval of the plant data is shorter than the dynamic characteristics of the plant 2, the monitoring variable may be configured with data of a longer section as shown by the following equation (15). .
[Equation 15]
x (t) = [z (t−h)... z (t−1) z (t)] (15)

  Note that the present invention is not limited to the above-described embodiments as they are, and may be embodied by modifying constituent elements without departing from the scope of the invention in the implementation stage. For example, in the example of the plant monitoring device 1 shown in FIG. 1, the plant monitoring device 1 and the display device 3 are shown as independent devices, but may be configured as an integrated device. That is, the plant monitoring apparatus 1 may include a display unit 3 or a display unit having a function equivalent to this.

  Also, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the above embodiments, or some constituent elements can be deleted from all the constituent elements shown in the embodiments. The present invention may be configured as follows.

  Further, the method described in each of the above-described embodiments is applied as a program that can be executed by a computer to a storage medium such as a magnetic disk, an optical disk, or a semiconductor memory and applied to various apparatuses, or transmitted by a communication medium. It can also be applied to various devices. A computer that implements the apparatus reads the program recorded in the storage medium, and executes the above-described processing by controlling the operation by the program.

The schematic diagram showing the basic composition of the plant monitoring device concerning the present invention. The schematic diagram of the monitoring model preparation means with which the plant monitoring apparatus which concerns on the 1st Embodiment of this invention is equipped. The schematic diagram of the monitoring processing means with which the plant monitoring apparatus concerning a 1st embodiment of the present invention is equipped. The processing flowchart which showed the processing content which the monitoring model preparation means with which the plant monitoring apparatus which concerns on the 1st Embodiment of this invention comprises is equipped. The processing flowchart which showed the processing content which the monitoring process means with which the plant monitoring apparatus which concerns on the 1st Embodiment of this invention is equipped performs. It is the figure which showed an example of the monitoring process result of the plant monitoring apparatus which concerns on the 1st Embodiment of this invention, Comprising: (a) is an output (measurement value, estimated value) on a vertical axis | shaft, and a horizontal axis | shaft is the number of steps. (B) is a graph in which the vertical axis is input, the horizontal axis is the number of steps, (c) is a graph in which the vertical axis is the estimated deviation, and the horizontal axis is the number of steps, (d) is the test index, the vertical axis. A graph with the horizontal axis representing the number of steps, (e) is a graph with the vertical axis representing the hypothesis acceptance rate and the horizontal axis representing the number of steps. The partial enlarged view which expanded and showed the range of 900 steps to 1100 steps in the graph of Fig.6 (a). The schematic of the monitoring model preparation means with which the plant monitoring apparatus which concerns on the 2nd Embodiment of this invention is equipped. The schematic of the monitoring processing means with which the plant monitoring apparatus which concerns on the 2nd Embodiment of this invention is equipped. The processing flowchart which showed the processing content which the monitoring model preparation means with which the plant monitoring apparatus which concerns on the 2nd Embodiment of this invention comprises is executed. The processing flowchart which showed the processing content which the monitoring process means with which the plant monitoring apparatus which concerns on the 2nd Embodiment of this invention is equipped performs. It is the figure which showed an example of the monitoring process result of the plant monitoring apparatus which concerns on the 2nd Embodiment of this invention, Comprising: (a) is a vertical axis | shaft output (measurement value, estimated value), and a horizontal axis | shaft is set with the number of steps. (B) is a graph in which the vertical axis is input, the horizontal axis is the number of steps, (c) is a graph in which the vertical axis is the estimated deviation, and the horizontal axis is the number of steps, (d) is the test index, the vertical axis. A graph with the horizontal axis representing the number of steps, (e) is a graph with the vertical axis representing the hypothesis acceptance rate and the horizontal axis representing the number of steps. The schematic of the monitoring model preparation means with which the plant monitoring apparatus which concerns on the 3rd Embodiment of this invention is equipped. Schematic of the monitoring process means with which the plant monitoring apparatus which concerns on the 3rd Embodiment of this invention is equipped. The processing flowchart which showed the processing content which the monitoring model preparation means with which the plant monitoring apparatus which concerns on the 3rd Embodiment of this invention comprises is equipped. The processing flowchart which showed the processing content which the monitoring process means with which the plant monitoring apparatus which concerns on the 3rd Embodiment of this invention is equipped performs. It is the figure which showed an example of the monitoring process result of the plant monitoring apparatus which concerns on the 3rd Embodiment of this invention, Comprising: (a) is a vertical axis | shaft output (measurement value, estimated value), and a horizontal axis | shaft is the number of steps. (B) is a graph in which the vertical axis is input (measurement value, estimated value), the horizontal axis is the number of steps, (c) is a graph in which the vertical axis is the estimated deviation, and the horizontal axis is the number of steps, (d) Is a graph with the test index on the vertical axis and the number of steps on the horizontal axis. (E) is a graph with the hypothesis acceptance rate on the vertical axis and the number of steps on the horizontal axis. The schematic of the monitoring model preparation means with which the plant monitoring apparatus which concerns on the 4th Embodiment of this invention is equipped. Schematic of the monitoring process means with which the plant monitoring apparatus which concerns on the 4th Embodiment of this invention is equipped. The schematic diagram of the monitoring model creation means with which the plant monitoring apparatus which concerns on the 5th Embodiment of this invention, and a plant monitoring apparatus is equipped. The schematic diagram of the monitoring processing means with which the plant monitoring apparatus which concerns on the 5th Embodiment of this invention, and a plant monitoring apparatus is equipped. It is the figure which showed an example of the input abnormality monitoring process result of the plant monitoring apparatus which concerns on the 5th Embodiment of this invention, Comprising: (a) is a vertical axis | shaft input (measurement value, estimated value), and a horizontal axis is a step. (B) is a graph with the vertical axis representing the distributed monitoring index, the horizontal axis representing the number of steps, and (e) a graph with the vertical axis representing the abnormality detection rate and the horizontal axis representing the number of steps. Schematic of the plant monitoring apparatus which concerns on the 6th Embodiment of this invention, the monitoring model preparation means with which a plant monitoring apparatus is equipped, and the monitoring process means.

Explanation of symbols

1 (1A to 1F) Plant monitoring device 2 Plant 3 Display device 4 Input means 11 Plant data input / storage means 12 (12A to 12F) Monitoring model creation means 13 (13A to 13F) Monitoring processing means 21 Output error model identification unit 22 First estimation deviation characteristic evaluation unit 25 First output estimation deviation calculation unit 26 First estimation deviation test processing unit 27 Hypothesis adoption rate determination unit 31 First principal component analysis unit 32 Second estimation deviation characteristic evaluation unit 35 Second output estimated deviation calculation unit 36 Second estimated deviation test processing unit 41 Second principal component analysis unit 42 Principal component score characteristic evaluation unit 45 Principal component score calculation unit 46 Principal component score test processing unit 51 Principal component analysis unit 52 statistic data characteristic evaluation unit 55 statistic data calculation unit 56 statistic data test processing unit 58 input variable characteristic evaluation unit 60 input abnormality monitoring processing means 61 statistic calculation unit 62 Hypothesis testing processor

Claims (10)

Plant data input / storage means for inputting measured data of plant input variables and output variables and storing them as time series data,
Monitoring model creating means for creating a monitoring model based on the time-series data stored in the plant data input / storing means;
A monitoring processing unit that applies the monitoring model to the measurement data to determine normality / abnormality of a plant state and outputs the determination result;
The monitoring model creating means includes an output error model identifying unit that identifies an output error model using the time series data;
An estimated deviation characteristic evaluation unit that evaluates a statistical characteristic of the estimated deviation of the output variable calculated by applying the output error model to the time series data, and
The monitoring processing means applies an output error model to the measurement data to calculate an estimated deviation of the output variable;
An estimated deviation test processing unit that tests a variance value change hypothesis of the estimated deviation by applying a sequential probability ratio test to the estimated deviation;
A hypothesis adoption rate determination unit that monitors plant abnormalities based on the adoption rate of the variance variation hypothesis of the estimated deviation;
A plant monitoring apparatus comprising: Plant data input / storage means for inputting measured data of plant input variables and output variables and storing them as time series data,
Monitoring model creating means for creating a monitoring model based on the time-series data stored in the plant data input / storing means;
A monitoring processing unit that applies the monitoring model to the measurement data to determine normality / abnormality of a plant state and outputs the determination result;
The monitoring model creating means applies a principal component analysis to the time series data to constitute a load matrix for a plurality of principal components having a large contribution rate;
An estimated deviation characteristic evaluation unit that evaluates a statistical characteristic of the estimated deviation calculated by applying the load matrix to the time series data; and
The monitoring processing means includes an output estimated deviation unit that calculates an estimated deviation by applying the load matrix to the measurement data;
An estimated deviation test processing unit that tests a variance value change hypothesis of the estimated deviation by applying a sequential probability ratio test to the estimated deviation;
A hypothesis adoption rate determination unit that monitors a plant abnormality based on an adoption rate of a variance value change hypothesis of the estimated deviation. Plant data input / storage means for inputting measured data of plant input variables and output variables and storing them as time series data,
Monitoring model creating means for creating a monitoring model based on the time-series data stored in the plant data input / storing means;
A monitoring processing unit that applies the monitoring model to the measurement data to determine normality / abnormality of a plant state and outputs the determination result;
The monitoring model creating means applies a principal component analysis to the time series data to constitute a load matrix for a plurality of principal components having a small contribution rate;
A principal component score characteristic evaluation unit that evaluates a statistical characteristic of the principal component score calculated by applying the load matrix to the time series data, and
The monitoring processing means includes a principal component score calculation unit that calculates a principal component score by applying the load matrix to the measurement data;
A principal component score test processing unit that tests a variance value change hypothesis of the principal component score by applying a sequential probability ratio test to the principal component score;
A hypothesis adoption rate determination unit that monitors a plant abnormality based on an acceptance rate of a variance value change hypothesis of the principal component score. Plant data input / storage means for inputting measured data of plant input variables and output variables and storing them as time series data,
Monitoring model creating means for creating a monitoring model based on the time-series data stored in the plant data input / storing means;
A monitoring processing unit that applies the monitoring model to the measurement data to determine normality / abnormality of a plant state and outputs the determination result;
The monitoring model creating means applies a principal component analysis to the time series data to form a first load matrix for a plurality of principal components having a large contribution rate;
A second principal component analysis unit constituting a second load matrix for a plurality of principal components having a small contribution rate from the result of the principal component analysis;
An estimated deviation characteristic evaluation unit that evaluates a statistical characteristic of the estimated deviation calculated by applying the first load matrix to the time series data;
A principal component score characteristic evaluation unit that evaluates a statistical characteristic of the principal component score calculated by applying the second load matrix to the time-series data; and
The monitoring processing means applies an estimated deviation by applying the first load matrix to the measurement data;
A principal component score calculating unit for calculating a principal component score by applying the second load matrix to the measurement data;
An estimated deviation test processing unit that tests a variance value change hypothesis of the estimated deviation by applying a sequential probability ratio test to the estimated deviation;
A principal component score test processing unit that tests a variance value change hypothesis of the principal component score by applying a sequential probability ratio test to the principal component score;
And a hypothesis adoption rate determination unit that monitors plant abnormalities based on the adoption rate of the variance deviation hypothesis of the estimated deviation and the adoption rate of the variance change hypothesis of the principal component score. . The plant monitoring apparatus according to any one of claims 1 to 4, further comprising an input abnormality monitoring processing unit that monitors a plant state by a variance test of the input variable.

A plant data input saving step for inputting measurement data of plant input variables and output variables and saving them as time series data,
A monitoring model creating step for creating a monitoring model based on the time-series data saved in the plant data input saving step;
A monitoring process step of applying the monitoring model to the measurement data to determine normality / abnormality of the plant state and outputting the determination result,
The monitoring model creation step includes an output error model identification step for identifying an output error model using the time series data;
An estimated deviation characteristic evaluation step for evaluating a statistical characteristic of an estimated deviation of the output variable calculated by applying the output error model to the time series data, and
The output monitoring processing step calculates an estimated deviation of the output variable by applying the output error model to the measurement data,
An estimated deviation test processing step of testing a variance value change hypothesis of the estimated deviation by applying a sequential probability ratio test to the estimated deviation;
A hypothesis adoption rate determination step for monitoring a plant abnormality based on the adoption rate of the variance value change hypothesis. A plant data input saving step for inputting measurement data of plant input variables and output variables and saving them as time series data,
A monitoring model creating step for creating a monitoring model based on the time-series data saved in the plant data input saving step;
A monitoring process step of applying the monitoring model to the measurement data to determine normality / abnormality of the plant state and outputting the determination result,
The monitoring model creation step includes applying a principal component analysis to the time series data to form a first load matrix for a plurality of principal components having a large contribution rate;
An estimated deviation characteristic evaluation step for evaluating a statistical characteristic of the estimated deviation calculated by applying the first load matrix to the time series data, and
The output monitoring processing step includes an output estimated deviation step of calculating an estimated deviation by applying the load matrix to the measurement data;
An estimated deviation test processing step of testing a variance value change hypothesis of the estimated deviation by applying a sequential probability ratio test to the estimated deviation;
And a first hypothesis adoption rate determination step of monitoring a plant abnormality based on an adoption rate of the estimated deviation variance change hypothesis. The monitoring model creation step includes applying a principal component analysis to the time series data to configure a second load matrix for a plurality of principal components having a small contribution rate;
A principal component score characteristic evaluation step for evaluating a statistical characteristic of the principal component score calculated by applying the second load matrix to the time series data;
The output monitoring processing step includes a principal component score calculating step of calculating a principal component score by applying the second load matrix to the measurement data;
A principal component score test processing step of testing a variance value change hypothesis of the principal component score by applying a sequential probability ratio test to the principal component score;
9. The plant monitoring method according to claim 8, further comprising a second hypothesis adoption rate determination step of monitoring an abnormality of the plant based on an acceptance rate of the variance value change hypothesis of the principal component score. The monitoring processing step includes an output monitoring processing step for monitoring the plant state by applying the monitoring model to the time series data, and an input abnormality monitoring processing step for monitoring the plant state by a variance test of the input variable. 10. The plant monitoring method according to any one of claims 7 to 9, wherein:

Images