JPH10260722A - Abnormal value detecting device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make an accurate decision even when a measured value increases gradually and when the measured value varies abruptly as operation is changed by checking one kind of measured value from various angles and totally deciding the check result by employing main component analysis. SOLUTION: An abnormal value detection logic arithmetic unit 2 consists of two 1st and 2nd devices. The 1st abnormal value detection logic arithmetic unit consists of a basic data arithmetic unit 211, a variation quantity overestimate arithmetic unit 212, a similar instrument comparing arithmetic unit 213, and a prediction error arithmetic unit 214. The 2nd abnormal value detection logic arithmetic unit consists of a trend arithmetic unit 221 and a variation quantity underestimate arithmetic unit 222. The measured value is checked from various angles such as the measured value, variation quantity overestimation, variation quantity underestimation, comparison with similar instruments, trends, and relation with other instruments. Then main component analysis is employed to totally decide the result of abnormal value detection logic.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an abnormal value detecting device for detecting abnormal data from a measuring instrument used in a plant such as a water and sewage treatment system.

[0002]

2. Description of the Related Art In processes such as water treatment and sewage treatment plants, a large number of measuring instruments installed in a plant are monitored in a central monitoring room. When an abnormality occurs in each measured value, the result is reported to an operator. Notify. FIG. 7 shows an abnormal value detection device in a conventional water and sewage treatment plant. 1 is a measured value accumulating device for storing measured values collected and collected from the plant, 9 is a standard deviation calculating device for calculating the standard deviation of past measured values, and 10 is a standard deviation calculated by the standard deviation calculating device 9. A criterion value calculating device for calculating a criterion value (upper limit value, lower limit value) of a measured value, 11 compares the measured value with a value calculated by the criterion value calculating device and determines whether the measured value exceeds the criterion value. An abnormal value judgment device 8 for judging whether or not the value is a value calculated by the abnormal value judgment device 11, that is, an abnormal value detection result display device for displaying whether the measured value is normal or abnormal. Such an abnormal value detection device sometimes determines a normal value to be an abnormal value or an abnormal value to be a normal value when dirt accumulates in a measuring instrument or when operating conditions are changed. For example, in an MLSS (activated sludge suspended solids) concentration meter, contamination may accumulate in a sensor unit due to a failure in a cleaning device, and the measured value may increase or decrease unilaterally. In the abnormal value detection based on the upper and lower limits of the conventional method, since the change in the measured value is gradual, such abnormalities in the measured value are determined as normal values.
Abnormal measurement values due to accumulation of dirt could not be detected. In addition, in the DO (dissolved oxygen) meter, when the aeration air volume was increased, the value of the DO meter rapidly increased. Therefore, the conventional method deviated from the upper and lower limits, and determined that the measured value was abnormal. However, the rise in the DO meter is a phenomenon caused by increasing the amount of aeration air, and the measuring instrument is operating normally. That is, in the conventional method, since the abnormal value is not detected in consideration of the influence of the operation change, it may be determined that the measurement device is abnormal even when the measuring device is operating normally.

[0003]

As described above, in the conventional configuration in which the abnormal value of the measuring instrument is detected by checking the upper and lower limits, the measured value unilaterally increases or decreases due to accumulation of dirt in the sensor section. It is difficult to detect an abnormal value corresponding to a case where the measured value shows a sudden change due to a change in operation or a change in operation. Therefore, the present invention checks one type of measured value from various angles and comprehensively determines the result using principal component analysis, so that when the measured value gradually increases or when driving changes. It is an object of the present invention to provide an abnormal value detection device that can perform accurate determination even when a measured value shows a sudden change.

[0004]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, the present invention provides a measured value accumulating device for accumulating measured values collected from measuring instruments installed in a plant, and a measuring device for measuring measured values from the measured value accumulating device. A basic data calculation device for calculating a value, an excessive change amount arithmetic device for calculating an amount of change for checking an excessive amount of change, a prediction error arithmetic device for calculating an error predicted from a prediction model using a statistical method, and the plant. A first abnormal value detection logic operation device comprising a similar measurement device comparison operation device for comparing with a measurement value from a measurement device installed at a different point in the inside, and an average value of a long cycle from the measurement value accumulation device measurement value And a second abnormal value detection logic operation device comprising a trend operation device that calculates the amount of change and a change amount undercalculation device that calculates the amount of change to check the amount of underchange, and A data aggregating device for aggregating the results from the one abnormal value detection logic arithmetic device, a principal component analysis aggregating device for principal component analysis of the result from the data aggregating device, and a value calculated by the principal component analysis aggregating device Index calculating device for calculating the distance of Mahalanobis from one to one index, a general index determining reference value calculating device for calculating a determination reference value based on the result from the general index calculating device, and the second abnormal value detection A trend judgment reference value operation device for calculating a judgment reference value based on the result from the trend operation device of the logic operation device, and an under-change amount judgment reference value operation for calculating a judgment reference value based on the result from the under-change amount operation device A determination reference value calculation device including a device, and an abnormal value is determined from a result from the general index determination reference value calculation device and a result from the general index calculation device. A trend abnormal value determination device that determines an abnormal value from a result from the general abnormal value determination device and the trend calculation device and a result from the trend determination reference value calculation device, a result from the change amount under-calculation device, and the change amount An abnormal value judging device composed of a change amount undersized abnormal value judging device for judging an abnormal value from a result from the under judging reference value calculating device, and an abnormal value detecting result display device displaying a result from the abnormal value judging device; Is provided. By the above means, one type of measured value is checked from various angles, so that an abnormal value can be accurately detected even with a change in various measured values.

[0005]

Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 1 is a functional block diagram showing an apparatus for detecting an abnormal value of a measured value of a sewage treatment plant of the present invention. The same components are denoted by the same reference numerals. In the figure, 1 is a measurement value storage device, 2 is an abnormal value detection logic operation device, 3 is a data aggregation device, 4 is a principal component analysis operation device,
Reference numeral 5 denotes a general index calculation device, 6 denotes a judgment reference value calculation device, 7 denotes an abnormal value judgment device, and 8 denotes an abnormal value detection result display device.
The measured value accumulating device 1 is a measured value accumulating device that accumulates measured values obtained from a measuring device provided in a process of a sewage treatment plant. The abnormal value detection logic operation device 2 includes two devices, a first device and a second device. The first abnormal value detection logic operation device is a basic data operation device 211, an excessive change amount operation device 212, a similar measurement device comparison operation device 21
3. It is composed of four devices of the prediction error calculation device 214. The second abnormal value detection logic operation device is the trend operation device 2
21 and an under-change amount calculation device 222. Here, the basic data calculation device 211 outputs the measured value X (n) at the time point n stored in the measured value storage device 1. The excessive change amount arithmetic unit 212 outputs a change amount Y (n) defined by the following equation. Y (n) = X (n) −X (n−1) (1) where X (n) is a measured value at the time point n. The similar measurement device comparison / calculation unit 213 has the same processing system and different MLS
The difference between the measured values of channel 1 and channel 2 such as S concentration meter and DO meter,
That is, Y (n) defined by the following equation is output. Y (n) = X ₁ (n) −X ₂ (n) (2) where X ₁ (n) is the measured value of channel 1 at time n, and X ₂ (n) is the value measured at time n It is a measured value of the waterway 2.

[0006] The prediction error calculation unit 214 predicts a measured value in consideration of the movement of the manipulated variable by an autoregressive model, and outputs the prediction error. An autoregressive model composed of k variables is defined by the following equation (3), where X (n) is a k-dimensional vector of the state of the process at time n.

(Equation 1) Here, A (m) is the regression coefficient of the autoregressive model, U (n) is its own eigen noise vector, and M is the optimal order of the autoregressive model. The prediction error calculation unit 214 considers a part obtained by removing U (n) from X (n) as a predicted value Xp (n), and outputs a prediction error Y (n) defined by the following equation. Y (n) = Xp (n) -X (n) (4) where Xp (n) is a predicted value at n time and X (n) is a measured value at n time. The trend calculation device 221 detects an abnormality when the measured value unilaterally increases or decreases due to accumulation of dirt on the sensor unit. For this reason, the average value and its change amount in a relatively long cycle are calculated. Since the measurement data of the sewage treatment plant has a 24-hour cycle, the data is handled on a daily basis, and the amount of change in the daily average value is set as target data for abnormal value detection. The change amount undercalculation device 222 detects an abnormality such as a failure of a measuring instrument when there is almost no change in the measured value. Therefore, the amount of change in the measured value obtained from the measured value accumulating device 1, that is, the expression (1) is used as the target data for abnormal value detection. The data aggregating device 3 includes a basic data aggregating device 31, an excessively large change amount aggregating device 32, a similar measurement value difference aggregating device 33, and a prediction error aggregating device. Here, the basic data aggregating device 31 determines the last three values of the measured values stored in the measured value storage 1.
The data for the 0th day and the current measurement value are aggregated. The excessive change amount aggregation device 32 aggregates the change amount in the last 30 days and the current change amount. The similar measurement value difference aggregating device 33 aggregates the data of the latest 30 days of the measurement value difference of the similar measurement values of the same processing system but different waterways and the current similar measurement value difference. The prediction error aggregating device 34 aggregates the latest 30 days of the difference (prediction error) between the value predicted by the autoregressive model and the measured value and the current prediction error.

[0007] The principal component analysis operation unit 4 performs a principal component analysis of multivariate analysis using the results of the abnormal value detection logic operation unit 2 aggregated by the data aggregation unit 3 as target data. Principal component analysis, a p-type variables _{X 1, X 2, ··· X} p,
Using a given coefficient a ₁ , a ₂ ,... A _p , a comprehensive index (composite variable) is created by the following linear expression. z = a ₁ X ₁ + a ₂ X ₂ +... + a _p X _p = a ^′ X (5) where a = (a ₁ ,..., a _p ) ^′ and X = (X ₁ ,…
·, X _P ) ^' . The z thus combined must be well representative of the p variables X. The variance of the principal component z is maximized as a criterion for that. The formulation of principal component analysis using this criterion (maximization of variance of principal components) will be described below. The variance of z when combining from the p variables by equation (5) is given by equation (6).

(Equation 2) Where sjk is the covariance of xjk and xk, and S is sjk
This is the variance-covariance matrix of the (j, k) element. Z in equation (5)
Means that the z axis is taken in the OZ direction from the origin O in the p-dimensional space, and the scale of the z coordinate is x
₁ x ₂ ... X If the _P axis is the same, the coefficient a
_1, a ₂ ··· a _p each becomes the direction cosines of the straight line OZ, meet ^{_{a 'a = a 1 2 +}} a 2 2 + ··· + a p 2 = 1 ···· (7). Therefore, the problem is to maximize equation (6) under the constraints of equation (7). Such a maximization problem with constraints is a problem of maximizing F (a, λ) = a'Sa-λ '(a'a-1) (8) using the Lagrange multiplier λ. Is transformed into If the equation (8) is partially differentiated with respect to each element of a and set to zero, an eigenvalue problem of the matrix S as in the equation (9) is obtained.

(Equation 3) S is a symmetric matrix, and a'Sa = V (a '
x) Since it is a non-negative value from ≧ 0, it has p real non-negative eigenvalues λ ₁ ≧ λ ₂ ≧... ≧ λ _p ≧ 0. If each eigenvalue vector is a ₁ , a ₂ , a ₃ ,..., A _p, then from equation (9), Sa _ij = λ _j a _j , and λ _j is equal to the variance of z _j = a _j ^′ X . Therefore, the principal component that maximizes the variance is the eigenvector a ₁ = (a ₁₁ , a ₂₁ ,..., Ap _p1 ′) corresponding to the maximum eigenvalue λ _1.
Become z ₁ = a ₁ as the coefficient _{elements' x = a 11 x 1 +} ··· + a n1 x n ···· (10). This is called a first principal component. The variance of the first principal component z ₁ is λ _1, which is a composite variate that is representative of p variates. When the variation of the original p-dimensional data is not sufficiently explained only by the first principal component, the linear combination z by the equation (5) is considered again. Here, z is the first principal component z ₁
In because those considered to explain the portion which can not be explained, taken in so as to mutually uncorrelated and z and z _1.
That is, Cov (z, z ₁ ) = a′Sa ₁ = λ ₁ aa ₁ = 0... (11) Under the constraints, the equations (7) and (11), Maximize variance. Two Lagrange multipliers lambda, the use of ν F (a, λ, ν ) = a'Sa -λ (a'a-1) -ν to (A'A ₁₎ maximization problem .. (12) Be transformed. If this is partially differentiated by each element of a and set to zero, the equation (13) is obtained.

(Equation 4) This both sides, in consideration of the multiplying a ₁ from the left (11) becomes [nu = 0, (13) equation reduces to the same eigenvalue problem as obtained in the first principal component. This time, the eigenvector a ₂ = (a ₁₂ , a ₂₂ ,...) Corresponding to the _second largest eigenvalue λ ₂
a _p2) 'as the coefficients of the elements of z ₂ = a _2' synthesized as _{x = a 12 x 1 + ···} + ap P2 xp ··· (14). This is called a second principal component. In the same manner, the dispersion is lambda _3, can be obtained .., third principal component such that lambda _P, ..., the first p principal component. The outline is shown in FIG. The first principal component score represents a value perpendicular to the axis of the first principal component from the point (a), and the second principal component score is
(a) represents a value perpendicular to the axis of the second principal component from the point. The coefficients a _1p , a _2p ...
・ Data x of abnormal value detection logic of ap _pp and current time n
_{1 (n), x 2 (} n), ···· x p n) from each of the principal component scores z
Calculate _p (n). _{z p (n) = a 1p} x 1 (n) + a 2p x 2 (n) + ···· an np x p (n) ·· (15)

[0008] The comprehensive index computing device 5 computes a comprehensive index from the eigenvalues obtained by the principal component analysis computing device 4 and the principal component score using the Mahalanobis distance. When the principal component scores up to the second principal component are taken in, the first principal component and the second principal component are uncorrelated with each other, so that an arbitrary point (z ₁ , z ₂ ) and a center of gravity (0,0)
Square distance D between defined as follows as the sum of squared distances squared distance and z ₂ direction of z ₁ direction. D = ((Z ₁ ² / V (Z ₁ ))) + (Z ₂ ² / V (Z ₂ ))) ^1/2 (16) The same applies when calculating the distance of the Mahalanobis to be taken up to the third principal component. To do. The Mahalanobis distance D is used as an overall index of the abnormal value detection logic output result. The outline is shown in FIG. Mahalanobis distance is calculated by the first principal component analysis.
The variance of the main component and the second main component are made equal, and the distance from the center of gravity at that time is shown. The determination reference value calculation device 6 includes a general index determination reference value calculation device 61, a trend determination reference value calculation device 62, and a change amount under-determination reference value calculation device 63. The comprehensive index determination reference value calculation device 61 stores a determination reference value of the comprehensive index in order to determine whether the value obtained by the comprehensive index calculation device 5 is normal or abnormal. The judgment reference value is determined by manual input or automatic setting by a computer. The automatic setting by the computer is performed on the basis of the quartile using the comprehensive index as the target data. The quartiles are the 25%, 50%, and 75% points of the relative cumulative frequency, from 25% to 50%
Is the hinge width 1 and the hinge width 2 is 50% to 75%.
The reference value is set based on one hinge width. Those that are at least 6 times the hinge width 1 from the 25% point and at least 6 times the hinge width 2 from the 75% point are called far-outs and are abnormal values. That is, in the determination method based on the quartile, the lower limit and the upper limit of the following equation are set as the determination reference values. Lower limit value = 25% point−6 × hinge width 1 (17) Upper limit value = 75% point−6 × hinge width 2 (18) The trend determination reference value calculation device 62 is a trend calculation device. In order to determine whether or not the value obtained in step 25 is normal, a reference value for determining the amount of change in the daily average is stored. The judgment reference value is determined by manual input or automatic setting by a computer. Here, in order to detect when the amount of change in the daily average value indicates positive or negative for several consecutive days as an abnormality, the judgment reference value is manually set to 0 and stored. The under-change amount determination reference value calculation device 63 stores the under-change amount determination reference value to determine whether the value calculated by the under-change amount calculation device 26 is normal. The judgment reference value is determined by manual input or automatic setting by a computer. Here, in order to detect an abnormality when there is no change in the measured value due to a failure of the measuring instrument or the like, the judgment reference value is manually set as 0 and stored.

The abnormal value judging device 7 comprises a general abnormal value judging device 71, a trend abnormal value judging device 72, and a change amount undersized abnormal value judging device 73. Total abnormal value determination device 7
1 determines whether or not the comprehensive index calculated by the general index calculating device 5 exceeds the judgment reference value calculated by the general index judgment reference value calculating device 61. The trend abnormal value determination device 72 determines whether or not the amount of change in the daily average value obtained by the trend calculation device 25 exceeds the determination reference value stored in the trend determination reference value storage device 62 for several consecutive days. Is determined. The under-change amount abnormal value determination device 73 determines the determination reference value stored in the under-change amount determination reference value storage device 63 for the change amount of the measurement value obtained by the under-change amount calculation device 26 continuously for several days. It is determined whether it is within the reference value. The abnormal value detection result display device 8 is a
1. A display device for notifying the operator of an abnormality when an abnormality is detected by any of the trend abnormality determination device 72 and the change amount under-determination abnormality determination device 73. Next, an example in which the device of the present invention is applied to an MLSS concentration meter of a sewage plant will be described. FIG. 4 shows a situation where an abnormality is detected by the abnormal value detection logic operation device of FIG. From the figure, it can be seen that points detected as abnormal by each detection logic output different results. Since the six detection logics look at the data from different angles, even when the excessive change amount is abnormal, the basic data, the prediction error, and the similar measurement device may output different results such as normal. In addition, even when the basic data and the excessive change amount are abnormal, the prediction error and the similar measurement device may output a normal value. By checking the measured values from various angles in this way, it is less likely that the normal value is an abnormal value and the abnormal value is erroneously determined as a normal value.

FIG. 5 shows the change over time of the total index (maharanobis distance) principal component score calculated by the comprehensive judgment operation device and the judgment reference value (lower limit value, upper limit value). As a result of applying the principal component analysis to obtain the comprehensive index of each detection logic, the contribution ratio in this example is about 43% for the first principal component, about 29% for the second principal component, and about 29% for the third principal component. 15%, 4th main component is about 13
%. Here, in order to capture the cumulative contribution rate of 70% or more, the integration of the first principal component score and the second principal component score is performed using the Mahalanobis distance, and this distance is a determination reference value obtained from the data of the last 30 days. Above is considered abnormal. FIG. 6 shows points detected as abnormal in FIG. 5 and measured values. The calculation of the comprehensive index based on the Mahalanobis distance is based on the result of each abnormal value detection logic detected in FIG.
It is summarized in one index, and online data can be checked comprehensively and accurately. As in the conventional case, the operator comprehensively determines the result of the detection of an abnormality by the abnormal value detection logic operation device every day, and determines whether or not the measured value is abnormal. Thereby, this monitoring work can be largely automated. In this embodiment, an example in which the present invention is applied to a lower treatment apparatus is shown. However, the present invention is not limited to this, and may be applied to other treatment plants such as a water treatment apparatus and a waste treatment apparatus.

[0011]

As described above, according to the present invention, measured values can be measured from various angles, such as measured values, excessive change, excessive change, comparison with similar measuring devices, trends, and relationships with other measuring devices. Is checked, the accuracy of abnormal value detection can be improved. In addition, by using the principal component analysis, it is possible to check data that comprehensively determines the result of the abnormal value detection logic.

[Brief description of the drawings]

FIG. 1 is a functional block diagram illustrating an outline of an abnormal value detection device according to the present invention.

FIG. 2 is an explanatory diagram showing the principle of principal component analysis.

FIG. 3 is an explanatory diagram for calculating a Mahalanobis distance.

FIG. 4 is a characteristic diagram showing an abnormal value detection result of the MLSS concentration by the abnormal value detection logic of the present invention.

FIG. 5 is a characteristic diagram showing a change over time of a comprehensive index of the present invention.

FIG. 6 is a characteristic diagram showing a result of detecting an abnormal value of the MLSS concentration according to the present invention.

FIG. 7 is a block diagram illustrating an outline of a conventional abnormal value detection device.

[Explanation of symbols]

 1: Measured value accumulation device 2: Abnormal value detection logic operation device 21: First abnormal value detection logic operation device 211: Basic data operation device 212: Excessive change amount operation device 213: Similar measurement device comparison operation device 214: Prediction error Arithmetic unit 22: Second abnormal value detection logic arithmetic unit 221: Trend arithmetic unit 222: Under-change amount arithmetic unit 3: Data aggregator 31: Basic data aggregator 32: Change aggregator 33: Similar measured value difference aggregator 34: Prediction error aggregation device 4: Principal component analysis calculation device 5: Comprehensive index calculation device 6: Judgment reference value calculation device 61: Comprehensive index judgment reference value calculation device 62: Trend judgment reference value calculation device 63: Change amount underdetermination reference Value calculation device 7: Abnormal value judgment device 71: Comprehensive index abnormal value judgment device 72: Trend abnormality judgment device 73: Change amount too small abnormality judgment device 8,12: abnormal value detection result display unit 9: standard deviation calculation unit 10: determination reference value calculation unit 11: an abnormal value determining device

Claims (1)

[Claims] 1. A measurement value accumulating device for accumulating measurement values collected from a measuring instrument installed in a plant, a basic data calculation device for calculating a measurement value from the measurement value accumulating device, and checking an excessive change amount An excessive change amount arithmetic device for calculating an amount of change, a prediction error arithmetic device for calculating an error predicted from a prediction model using a statistical method, and a measuring device installed at a different point in the plant. A first abnormal value detection logic operation device comprising a similar measurement device comparison operation device for comparing with a measurement value; and a trend operation device for calculating an average value of a measurement value from the measurement value storage device over a long period and its change amount. And a second abnormal value detection logic operation device including a change amount undercalculation device that calculates the change amount to check the underchange amount, and from the first abnormal value detection logic operation device A data aggregation device that aggregates the results of the above, a principal component analysis device that performs principal component analysis on the result from the data aggregation device, and a Mahalanobis distance calculated from the value calculated by the principal component analysis device. An overall index operation device serving as an index, an overall index determination reference value operation device for calculating a judgment reference value based on a result from the overall index operation device, and the trend operation device of the second abnormal value detection logic operation device A determination reference value calculation device for calculating a determination reference value based on the result from the above, and a change amount under-determination reference value calculation device for calculating a determination reference value based on the result from the under-change amount calculation device A calculating device; a general abnormal value determining device for determining an abnormal value from a result from the general index determining reference value calculating device and a result from the general index calculating device; A trend abnormal value judging device for judging an abnormal value from the result from the rendering operation device and the result from the trend judgment reference value operation device, and a result from the under-change amount operation device and the under-change amount reference value operation device An abnormal value detection device comprising: an abnormal value determination device configured to determine an abnormal value from the result from the abnormal value determination device; and an abnormal value detection result display device configured to display a result from the abnormal value determination device. apparatus.