JP2001175326A - Multi-variable data monitoring device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize the plant state monitoring of high performance, to reduce the load of cost and to effectively utilize past stored data in a multi-variable data monitoring device for inputting plural processes of a plant to be monitored and analyzing main components. SOLUTION: The multi-variable data monitoring device is provided with an analytical data preprocessing unit for calculating moving average based on polynomial approximation and prescribed sampling period data suited to analysis are generated from various sampling period data corresponding to respective process quantity data. Consequently high frequency noise can be cut out and quality analytical time-series data of a fixed sampling period are obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a multivariate data monitoring device for monitoring a state of a plant using a plurality of process quantities of the plant as keys.

[0002]

2. Description of the Related Art FIG. 17 is a block diagram showing a general configuration of a multivariate data monitoring device. The multivariate data monitoring device 1 is a device for comprehensively judging a large amount of process in a power plant or an industrial plant and monitoring whether or not the state of the plant 10 is abnormal.

[0003] For this monitoring, a process amount collecting device 12 for taking in a plurality of process amounts of the plant 10, a collected process amount and, when an alarm is generated, the type of the generated alarm and an operator are operating. A process amount database 16 for storing the contents of the operation,
A feature amount conversion device 14 that performs principal component analysis on the collected process amounts to calculate principal component scores, singular values, and principal component loadings; and all or any of the obtained principal component scores, singular values, and principal component loadings. And a display device 16 for displaying the principal component score on a coordinate system in which the horizontal axis represents time and the vertical axis represents the value of the principal component score. The plant 10 is monitored by a graph of the principal component score with less.

[0004] As a result, only the operator 20 can monitor the plant 10, and the plant 10 can be effectively monitored by a small number of operators.

[0005]

The time series data to be subjected to the principal component analysis is based on a certain fixed sampling period, and it is necessary to sample each process amount at a predetermined period when collecting data. .

In general, many plants 10 record and store a large amount of data over the period from the past to the present, but for data recording, the efficiency of storage for a predetermined capacity is improved. Therefore, the sampling period is changed according to the time constant of the corresponding portion. That is, a portion having a long time constant has a large sampling period, and a portion having a short time constant has a small sampling period, and more information is recorded in a predetermined capacity. Therefore, the principal component analysis is
It is not possible to directly use precious accumulated data of the past. This means that past case studies cannot be performed, and is particularly fatal when used for case inference.
Therefore, for a process amount having a large sampling period, it is conceivable to adjust the data sampling period by linear interpolation or spline interpolation.

However, in these methods, when high-frequency noise is received, data with low reliability is created. In particular, in the case of spline interpolation, the influence of the high-frequency noise may be increased. In other words, this is a factor that also deteriorates the effect of the principal component analysis.

[0008] Even for new data, collecting data in accordance with the minimum sampling period imposes an extremely heavy burden on costs.

SUMMARY OF THE INVENTION It is an object of the present invention to realize a high-performance plant state monitoring without a large cost burden.
An object of the present invention is to provide a multivariate data monitoring device that can effectively use past accumulated data.

[0010]

According to a first aspect of the present invention, there is provided a multivariate data monitoring apparatus for performing a principal component analysis by taking in a plurality of process quantities of a plant to be monitored, and analyzing data for executing a moving average by a polynomial approximation. A multivariate data monitoring device comprising a preprocessing device and generating predetermined sampling cycle data suitable for analysis from various sampling cycle data corresponding to each process amount.

Therefore, in the multivariate data monitoring apparatus according to the first aspect of the present invention, various sampling period data corresponding to each process amount is subjected to a moving average process by a polynomial approximation to obtain a predetermined sampling period suitable for analysis. Since the data is generated, high-frequency noise can be cut, and high-quality analysis time-series data with a constant sampling period is generated.

Therefore, high-performance plant state monitoring can be realized, there is no great burden on costs, and past accumulated data can be effectively used.

Next, the invention according to claim 2 is based on claim 1.
A cubic approximation is employed as a polynomial approximation used for a moving average in the described analysis data preprocessing device, and various sampling period data corresponding to each process amount is generated as predetermined sampling period data suitable for analysis. A multivariate data monitoring device characterized in that:

Therefore, in the multivariate data monitoring apparatus according to the second aspect of the present invention, various sampling period data corresponding to each process amount is subjected to a moving average process by cubic approximation. Since the predetermined sampling period data suitable for the analysis is generated by the same operation as in the first embodiment, the same effect as in the first embodiment can be obtained.

According to a third aspect of the present invention, there is provided the analysis data pre-processing device according to the first or second aspect, wherein an amount of collected data required for analysis cannot be obtained, and the corresponding time is the center of the original data of the approximate expression. If the data is not between the two data, the central part of the predetermined number of collected data selects the collected data closest to the corresponding time, derives an approximate expression, and obtains various values according to each process amount. A multivariate data monitoring apparatus characterized in that the sampling cycle data is supplemented by a shortage of predetermined sampling cycle data suitable for analysis.

Therefore, in the multivariate data monitoring apparatus according to the third aspect of the present invention, the amount of collected data required for the analysis cannot be obtained, and the corresponding time is between the two central data of the original data of the approximate expression. , The central part of the predetermined number of collected data selects the collected data closest to the corresponding time, derives an approximate expression, and performs the same operation as in the case of claim 1 described above. In order to compensate for the lack of the predetermined sampling period data suitable for the analysis, the same effects as those of the first and second aspects can be obtained.

According to a fourth aspect of the present invention, there is provided the analysis data pre-processing device according to the first or second aspect, wherein an amount of collected data necessary for analysis cannot be obtained, and the corresponding time is within the original data of the approximate expression. If it does not exist, select the collected data closest to the relevant time and derive an approximation formula, and compensate for the lack of the predetermined sampling cycle data suitable for analysis of various sampling cycle data corresponding to each process amount. A multivariate data monitoring device characterized in that:

Therefore, in the multivariate data monitoring apparatus according to the fourth aspect of the present invention, even when the amount of collected data required for analysis cannot be obtained and the corresponding time does not exist inside the original data of the approximate expression, An approximate expression is derived by selecting the collected data closest to the corresponding time, and the shortage of the predetermined sampling period data suitable for the analysis is supplemented by extrapolation by the same operation as in the case of claim 1 described above. Therefore, claim 1,
The same effect as in the case of No. 2 can be obtained.

According to a fifth aspect of the present invention, there is provided the analysis data pre-processing device according to the first or second aspect, wherein an amount of collected data required for analysis cannot be obtained, and the corresponding time is within the original data of the approximate expression. If it does not exist, an AR model is constructed from the collected data, and the process amount of the portion without the collected data at the corresponding time is estimated. Extrapolate the data so as to obtain an approximate expression based on the extrapolated data and the collected data, and divide the various sampling cycle data corresponding to each process amount into a predetermined sampling cycle data shortage suitable for analysis. Is a multivariate data monitoring device characterized in that:

Therefore, in the multivariate data monitoring device according to the fifth aspect of the present invention, even if the amount of collected data required for analysis cannot be obtained and the corresponding time does not exist inside the original data of the approximate expression, An AR model is constructed from the collected data, and the data is extrapolated so that the corresponding time is between the two data at the center of the data that is the source of the approximate expression. Is derived and the same effect as in claim 1 described above is used to compensate for the lack of predetermined sampling period data suitable for analysis, so that the same effect as in claims 1 and 2 is obtained. Can be.

According to a sixth aspect of the present invention, there is provided the analysis data pre-processing device according to the first or second aspect, wherein an amount of collected data required for analysis cannot be obtained, and the corresponding time is within the original data of the approximate expression. If it does not exist, the FIR filter is applied to the collected data to estimate the process amount of the portion where there is no collected data at the corresponding time, and the corresponding time is the data of the two central data of the data that is the base of the approximate expression. Extrapolate the data so that it is in between, and derive an approximation formula based on this extrapolated data and the collected data, and deficiency of predetermined sampling cycle data suitable for analysis of various sampling cycle data corresponding to each process amount It is a multivariate data monitoring device characterized by supplementing the minutes.

Therefore, in the multivariate data monitoring apparatus of the invention according to claim 6, even if the amount of collected data required for analysis cannot be obtained and the corresponding time does not exist inside the original data of the approximate expression, Applying an FIR filter to the collected data, extrapolating the data so that the relevant time is between the two data at the center of the data that is the basis of the approximate expression, and using the extrapolated data and the collected data to approximate the expression Is derived and the same effect as in claim 1 described above is used to compensate for the lack of predetermined sampling period data suitable for analysis, so that the same effect as in claims 1 and 2 is obtained. Can be.

[0023]

Embodiments of the present invention will be described below in detail with reference to the drawings.

(First Embodiment) FIG. 1 is a block diagram showing a configuration example of a multivariate data monitoring device 1 according to a first embodiment. The same parts as those in FIG. 17 are denoted by the same reference numerals.

In FIG. 1, a multivariate data monitoring device 1
Is a process amount collecting device 12 for taking in a plurality of process amounts of the plant 10, a collected process amount and the type of alarm that has been generated when an alarm has been generated, and the content of the operation when the operator is operating. A process amount database 16 to be stored and a data preprocessing device 13 for converting the collected process amount into a predetermined sampling period
A feature amount conversion device 14 that performs principal component analysis on a process amount unified at a predetermined sampling period and calculates a principal component score, a singular value, and a principal component loading; A feature database 18 storing all or any of the principal component loadings
And a display device 16 for displaying the principal component score on a coordinate system in which the horizontal axis represents time and the vertical axis represents the value of the principal component score.

As shown in FIG. 2, the data preprocessor 13 outputs a sample time 136 which is a time at which process data for analysis is required.
And the target sample data 137 which becomes the original data of the approximate expression using the process amount sample data 135 and the sample time 136 from the process amount collecting device 12 as inputs.
, A least-squares polynomial approximator 132 that receives the target sample data 137 as an input and outputs a least-squares approximation polynomial 138, and a process at a sample time 136 using the least-squares approximation polynomial 138. And an estimator 133 for outputting the amount estimation sample data 139 to the feature converter.

The operation of the multivariate data monitoring device 1 configured as described above will be described. The multivariate data monitoring apparatus 1 collects a predetermined monitoring target process amount of the plant 10 at time intervals according to the process amount.
That is, although it is necessary to collect data for each process amount at regular time intervals, this time interval may be changed according to the process amount. Therefore, the process data at the sample time 136 that is the time at which the process data for analysis is required does not always exist in the process amount sample data 135 from the process amount collection device 12. Therefore, the time-series data extractor 131 extracts some target sample data 137 from the process amount sample data 135. The least-squares polynomial approximator 132 approximates the process amount to the extracted target sample data 137 by a polynomial 138 relating to time by a least-squares approximation technique. The estimator 133 calculates the least square approximation polynomial 138
Is used to calculate the process amount estimation sample data 139 at the sample time 136. Hereinafter, the plant state monitoring is performed in the same manner as in the related art.

Hereinafter, an example will be described in detail with reference to FIGS. FIG. 3 shows a sampling cycle of a certain process amount. A certain process amount is a value x1 at time t1, a value x2 at time t2,..., A value x1 at time t12.
2. At time t13, values x13,... Are collected at a sampling period of Δt. On the other hand, the sampling period of the data for analysis is Δτ, and is from τ1 to τ2, τ3, τ4, τ
The data of 5, ... is required.

FIG. 4 shows how some target sample data 137 are extracted from the process amount sample data 135 at time τ1. Four pieces of target sample data 137 (t3, x
3), (t4, x4), (t5, x5), (t6, x6)}, four after the time τ1 {(t7, x7), (t8, x8), (t9, x
9), (t10, x10)｝ A total of eight (black circles) are extracted.

FIG. 5 is a diagram in which a least squares approximation polynomial 138 is derived by performing curve fitting by the least squares method on eight sets of the target sample data 137.

FIG. 6 shows the sample time 136 for the independent variable (time variable) of the least square approximation polynomial 138.
(Time τ1) is given, and the process amount estimation sample data 139 (y1: white circle) at time τ1 is calculated.

FIG. 7 shows how the process amount estimation sample data 139 (y2: white circle) at time τ2 is calculated at time τ2 in the same manner as at time τ1.

FIG. 8 shows how the process amount estimation sample data 139 (y3: white circle) at time τ3 is calculated at time τ3 in the same manner as at time τ1.

In the following, data up to a target time can be prepared by a weighted moving average method.

Therefore, even if the sampling cycle of the process amount changes for each process amount, the feature amount conversion device 14
A predetermined time interval data of a predetermined monitoring target process amount is input.

Therefore, high-performance plant state monitoring can be realized, and there is no great burden on costs, and past accumulated data can be effectively used.

In the multivariate data monitoring apparatus 1 according to the present embodiment, the case where four data before and after the relevant time are extracted has been described as an example. However, the present invention is not limited to this. Even with a small number, the same effect can be obtained by the same operation, and the same effect can be obtained by the same operation even if the number is different before and after the relevant time.

Further, in the multivariate data monitoring apparatus 1 according to the present embodiment, the case where the sampling period (Δt) of the process amount is longer than the sampling period (Δτ) for analysis has been described as an example. When the sampling period (Δt) of the process amount is smaller than the sampling period (Δτ) for the analysis or is different, the sampling period (Δt) of the process amount and the sampling period (Δτ) for the analysis are the same. Even when the time is off, the same effect can be obtained by the same operation.

(Second Embodiment) FIG. 9 is a block diagram showing a configuration example of a data preprocessing device 13 according to a second embodiment. Since the multivariate data monitoring device 1 is the same as that of the first embodiment as shown in FIG. 1, the description will be omitted, and only the data preprocessing device 13 will be described.

The data preprocessing device 13 outputs a sample time 136 which is a time at which process data for analysis is required, as shown in FIG.
And the target sample data 137 which becomes the original data of the approximate expression using the process amount sample data 135 and the sample time 136 from the process amount collecting device 12 as inputs.
, A least-squares cubic approximator 232 that receives the target sample data 137 as an input and outputs a least-squares approximation cubic expression 238, and a least-squares approximation cubic expression 238. And an estimated amount calculator 133 that outputs the process amount estimated sample data 139 at the sample time 136 to the feature amount conversion device.

The operation of the multivariate data monitoring device 1 configured as described above will be described. The multivariate data monitoring apparatus 1 collects a predetermined monitoring target process amount of the plant 10 at time intervals according to the process amount.
That is, although it is necessary to collect data for each process amount at regular time intervals, this time interval may be changed according to the process amount. Therefore, the process data at the sample time 136 that is the time at which the process data for analysis is required does not always exist in the process amount sample data 135 from the process amount collection device 12. Therefore, the time-series data extractor 131 extracts some target sample data 137 from the process amount sample data 135. The least-squares cubic approximator 232 approximates the process amount to the extracted target sample data 137 by a cubic formula 238 relating to time by a least-squares approximation method. The estimator 133 calculates the least square approximation cubic expression 238
Is used to calculate the process amount estimation sample data 139 at the sample time 136. Hereinafter, the plant state monitoring is performed in the same manner as in the related art.

Therefore, high-performance plant state monitoring can be realized, there is no great burden on costs, and past accumulated data can be effectively used.

The multivariate data monitoring device according to the present embodiment can be variously modified and implemented without departing from the gist, similarly to the multivariate data monitoring device according to the first embodiment.

(Third Embodiment) FIG. 10 is a block diagram showing a configuration example of a data preprocessing device 13 according to a third embodiment. Since the multivariate data monitoring device 1 is the same as that of the first embodiment as shown in FIG. 1, the description will be omitted, and only the data preprocessing device 13 will be described.

As shown in FIG. 10, the data pre-processing device 13 outputs a sample time 136 which is a time at which process data for analysis is required.
And the target sample data 137 which becomes the original data of the approximate expression using the process amount sample data 135 and the sample time 136 from the process amount collecting device 12 as inputs.
, A least-squares cubic approximator 232 that receives the target sample data 137 as an input and outputs a least-squares cubic expression 238, and a least-squares cubic expression 238. And an estimated amount calculator 133 that outputs the process amount estimated sample data 139 at the sample time 136 to the feature amount conversion device.

The operation of the multivariate data monitoring device 1 configured as described above will be described. The multivariate data monitoring apparatus 1 collects a predetermined monitoring target process amount of the plant 10 at time intervals according to the process amount.
That is, although it is necessary to collect data for each process amount at regular time intervals, this time interval may be changed according to the process amount. Therefore, the process data at the sample time 136 that is the time at which the process data for analysis is required does not always exist in the process amount sample data 135 from the process amount collection device 12. Therefore, when the time-series data extractor 331 extracts some target sample data 137 from the process amount sample data 135, the amount of collected data required for analysis cannot be obtained, and the corresponding time is equal to the original data of the approximate expression. If the data is not between the two central data, the collected data in which the two central data of the predetermined number of collected data are closest to the corresponding time are extracted by the time-series data extractor 331 from the process amount sample data 135 into the target sample. It is extracted as data 137. The least-squares cubic approximator 232 extracts the extracted target sample data 137.
Is approximated by a cubic expression 238 relating to time by the least squares approximation method. The estimator 133 calculates
Using the least squares approximation cubic equation 238, the sample time 13
6, the process amount estimation sample data 139 is calculated.

Hereinafter, a detailed description will be given using an example with reference to FIG. FIG. 11 shows a sampling cycle of a certain process amount. A certain process quantity has a value x at time t1.
1, value x2 at time t2, value x12 at time t12,
At time t13, 13 sets of data of value x13 are collected at a sampling period of Δt. On the other hand, the sampling period of the data for analysis is Δτ, and the required data is τ1 to τ2, τ3, τ4, τ5,.

First, at the time before the time τ1, four ｛(t3,
x3), (t4, x4), (t5, x5), (t6, x6)}, four after the time τ1, {(t7, x7), (t8, x8), (t9,
x9), (t10, x10)｝ A total of eight pieces are extracted, and a cubic curve fitting by the least square method is performed on the eight sets of the target sample data 137 to obtain a least square approximation cubic equation 238. Is derived, and the process amount estimation sample data 139 (y1) not shown is calculated.

Hereinafter, similarly, at time τ2,
At the time before the time τ2, four ｛(t3, x3), (t4, x
4), (t5, x5), (t6, x6)}, 4 after time τ2
｛(T7, x7), (t8, x8), (t9, x9), (t1
(0, x10)｝ A total of eight are extracted, and the process amount estimation sample data 139 (y2) not shown is calculated, and the time τ3
, Four ｛(t4, x
4), (t5, x5), (t6, x6), (t7, x7)}, four after time τ3 (t8, x8), (t9, x9), (t1
(0, x10), (t11, x11) {8 totals are extracted, and the process amount estimation sample data 139 (y
3), and for the time τ4, four ｛(t5, x5), (t6, x6), (t7, x7), (t
8, x8)}, four after time τ4 (t9, x9), (t
(10, x10), (t11, x11), (t12, x12)} A total of eight pieces are extracted, and the process amount estimation sample data 139 (y4), not shown, is calculated. In the previous time, four (t5, x5), (t6, x6),
(t7, x7), (t8, x8) {, four after time τ5}
(t9, x9), (t10, x10), (t11, x11), (t
12, x12)｝ A total of eight pieces are extracted, and process amount estimation sample data 139 (y5), not shown, is calculated.
6, four ｛(t6, x
6), (t7, x7), (t8, x8), (t9, x9)}, four after time τ6 (t10, x10), (t11, x1
1), (t12, x12), (t13, x13)｝ a total of eight (black circles) are extracted, and process amount estimation sample data 139 (y6) not shown is calculated. At the time before ｛(t6, x6), (t7, x
7), (t8, x8), (t9, x9)｝, 4 after time τ7
｛(T10, x10), (t11, x11), (t12, x
12), (t13, x13)｝ A total of eight (black circles) are extracted, and the process amount estimation sample data 139 (y
7) is calculated.

Next, at time τ8, four ｛(t7, x7), (t8, x8), (t9, x
9), (t10, x10)} can be extracted, but after time τ3, three ({t11, x11), (t12, x12), (t1)
3, x13) Only｝ can be extracted. Therefore, in this case, the same data 時刻 (t6, x6), (t7, x7),
(t8, x8), (t9, x9), (t10, x10), (t1
1, x11), (t12, x12), (t13, x13)｝ 8 in total
(Black circles), and a cubic curve fitting by the least squares method is performed on the eight sets of data of the target sample data 137 to derive a least squares approximation cubic equation 238 to obtain a process amount estimation sample. Data 139 (y8: white circle) is calculated.

Similarly for time τ9, the same data ｛(t6, x6), (t7, x7), (t8, x8),
(t9, x9), (t10, x10), (t11, x11), (t
(12, x12), (t13, x13)｝ A total of eight (black circles) are extracted, and the process amount estimation sample data 139 (y9: white circles) is calculated.

Thereafter, plant condition monitoring is performed in the same manner as in the conventional case.

Therefore, high-performance plant state monitoring can be realized, there is no great burden on costs, and past accumulated data can be effectively used.

The multivariate data monitoring device according to the present embodiment can be variously modified and implemented within the scope of the gist, similarly to the multivariate data monitoring device according to the second embodiment.

(Fourth Embodiment) FIG. 12 is a block diagram showing a configuration example of a data preprocessing device 13 according to the present embodiment. The multivariate data monitoring device 1 is the same as that of the first embodiment as shown in FIG.
Only the data preprocessor 13 will be described.

The data preprocessor 13 outputs a sample time 136 which is a time at which process data for analysis is required, as shown in FIG.
And the target sample data 137 which becomes the original data of the approximate expression using the process amount sample data 135 and the sample time 136 from the process amount collecting device 12 as inputs.
, A least-squares cubic approximator 232 that receives the target sample data 137 as an input and outputs a least-squares approximation cubic expression 238, and a least-squares approximation cubic expression 238. And an estimated amount calculator 133 that outputs the process amount estimated sample data 139 at the sample time 136 to the feature amount conversion device.

The operation of the multivariate data monitoring device 1 configured as described above will be described. The multivariate data monitoring apparatus 1 collects a predetermined monitoring target process amount of the plant 10 at time intervals according to the process amount.
That is, although it is necessary to collect data for each process amount at regular time intervals, this time interval may be changed according to the process amount. Therefore, the process data at the sample time 136 that is the time at which the process data for analysis is required does not always exist in the process amount sample data 135 from the process amount collection device 12. Therefore, when the time-series data extractor 431 extracts some target sample data 137 from the process amount sample data 135, the amount of collected data required for analysis cannot be obtained, and the corresponding time is equal to the original data of the approximate expression. If it does not exist inside, the collected data closest to the relevant time is extracted as the target sample data 137 from the process amount sample data 135 by the time series data extractor 431. The least square cubic approximator 232 outputs the extracted target sample data 13
7, the process amount is approximated by a cubic expression 238 relating to time by the least squares approximation method. Estimator 133
Calculates the process amount estimation sample data 139 at the sample time 136 using the least square approximation cubic expression 238.

Hereinafter, an example will be described in detail with reference to FIG. FIG. 13 shows a sampling cycle of a certain process amount. A certain process quantity has a value x at time t1.
1, value x2 at time t2, value x12 at time t12,
At time t13, 13 sets of data of value x13 are collected at a sampling period of Δt. On the other hand, the sampling cycle of the data for analysis is Δτ, and the required data requires 13 sets of data from τ1 to τ2, τ3, τ4, τ5,.

As in the second embodiment, at time τ1
From time τ7, process amount estimation sample data 139 (y1 to y7) not shown is calculated. Time τ
From time 8 to time τ12, the process amount estimation sample data 13 (not shown) is set in the same manner as in the third embodiment.
9 (y8 to y12) is calculated.

Next, with respect to time τ13, time τ13
There is no data for the time zone including. That is, at time τ13
There is only the collected data of the process amount at the time before. Therefore, in this case, the same data ｛(t6, x6) as at time τ7,
(t7, x7), (t8, x8), (t9, x9), (t10, x
10), (t11, x11), (t12, x12), (t13,
x13)｝ A total of eight (black circles) are extracted, and a cubic curve fitting by the least square method is performed on the eight sets of the target sample data 137 to derive a least square approximation cubic equation 238. , Process amount estimation sample data 13
9 (y13) is calculated by extrapolation.

Thereafter, the state of the plant is monitored in the same manner as in the prior art.

Therefore, high-performance plant state monitoring can be realized, there is no great burden on costs, and past accumulated data can be effectively used.

The multivariate data monitoring device according to the present embodiment can be variously modified and implemented without departing from the gist, as in the multivariate data monitoring device according to the second embodiment.

(Fifth Embodiment) FIG. 14 is a block diagram showing a configuration example of a data preprocessor 13 according to the present embodiment. The multivariate data monitoring device 1 is the same as that of the first embodiment as shown in FIG.
Only the data preprocessor 13 will be described.

As shown in FIG. 14, the data pre-processing device 13 outputs a sample time 136 which is a time at which the process data for analysis is required.
And the target sample data 137 which becomes the original data of the approximate expression using the process amount sample data 135 and the sample time 136 from the process amount collecting device 12 as inputs.
A time-series data extractor 531 for extracting the data and an auto-regression model (Auto-Regre
AR model identifier 532 that identifies a model by sssiveModel (hereinafter referred to as an AR model) and outputs an AR model 538
An additional data calculator 533 that outputs the estimation target sample data 539 in the shortage time of the process amount sample data using the identified AR model 538, the target sample data 137 and the estimation target sample data 53
9 as an input, and outputs a least-squares approximation cubic expression 238, and a least-squares approximation cubic expression 23
8 and an estimator 133 for outputting process amount estimation sample data 139 at the sample time 136 to the feature amount converter.

The operation of the multivariate data monitoring device 1 configured as described above will be described. The multivariate data monitoring apparatus 1 collects a predetermined monitoring target process amount of the plant 10 at time intervals according to the process amount.
That is, although it is necessary to collect data for each process amount at regular time intervals, this time interval may be changed according to the process amount. Therefore, the process data at the sample time 136 that is the time at which the process data for analysis is required does not always exist in the process amount sample data 135 from the process amount collection device 12.

Therefore, when the time-series data extractor 531 extracts some target sample data 137 from the process amount sample data 135, the amount of collected data necessary for analysis cannot be obtained, and the corresponding time is represented by an approximate expression. If the data is not located between the two data at the center of the original data, the data at the center of the predetermined number of collected data is extracted from the collected data closest to the corresponding time. , The collected data closest to the time is extracted by the time-series data extractor 531 into the process amount sample data 13.
5 is extracted as target sample data 137.

On the other hand, the target sample data 137 is also supplied to the AR model identifier 532, and the AR model 538 is identified and output to the additional data calculator 533. The additional data calculator 533 uses the identified AR model 538 to supply the estimation target sample data 537 of the time at which the process amount calculation data 135 is insufficient to the least square cubic approximator 232 as additional data. Least-squares cubic approximator 232
Approximates the process amount to the extracted target sample data 137 and estimation target sample data 537 by a cubic expression 238 relating to time by a least squares approximation method. The estimation amount calculator 133 calculates the process amount estimation sample data 139 at the sample time 136 by using the least square approximation cubic expression 238.

Hereinafter, a detailed description will be given using an example with reference to FIG. FIG. 15 shows a sampling cycle of a certain process amount. A certain process quantity has a value x at time t1.
1, value x2 at time t2, value x12 at time t12,
At time t13, 13 sets of data of value x13 are collected at a sampling period of Δt. On the other hand, the sampling cycle of the data for analysis is Δτ, and the required data requires 13 sets of data from τ1 to τ2, τ3, τ4, τ5,.

As in the second embodiment, at time τ1
From time τ7, process amount estimation sample data 139 (y1 to y7) not shown is calculated. Time τ
8 to time τ12, as in the third embodiment, process amount estimation sample data 1 (not shown).
39 (y8 to y12) is calculated.

Next, with respect to time τ13, time τ13
There is no data for the time zone including. That is, at time τ13
There is only the collected data of the process amount at the time before. Therefore, in this case, the following seven sets of data ｛(t7, x7), (t
8, x8), (t9, x9), (t10, x10), (t11, x
11), (t12, x12), (t13, x13)｝ (black circle)
Is used to estimate the process amount [x14] (gray circle) at time t14 based on the identified AR model. 7 sets of target sample data 137
｛(T7, x7), (t8, x8), (t9, x9), (t10,
x10), (t11, x11), (t12, x12), (t1
(3, x13)｝ (black circle) represents the sample data 537 to be estimated.
｛(T14, [x14])｝ (gray circle) is added, and a cubic curve fitting by the least square method is performed on a total of eight sets of data to derive a least square approximation cubic equation 238. The process amount estimation sample data 139 (y13) is calculated.

Thereafter, the state of the plant is monitored in the same manner as in the prior art.

As another example of the fifth embodiment, the following multivariate data monitoring device can be configured. That is, as the analysis data pre-processing device of the first and second embodiments, if the relevant time (current time) does not exist inside the original data of the approximate expression in the online analysis, an AR model is constructed from the collected data and Extrapolation of one sampling (one piece) of data is performed on a process amount basis for a portion where there is no collected data at a time, and an approximate expression is derived based on the extrapolated data and the collected data to obtain various expressions corresponding to each process amount. A multivariate data monitoring device that generates on-line periodic data at the current time from the sampling periodic data is configured.

In this multivariate data monitoring apparatus, even if the corresponding time (current time) does not exist inside the original data of the approximate expression in the online analysis, the AR
By constructing a model, extrapolating data of one sampling (one piece) on a process amount basis for a portion having no collected data at the corresponding time, deriving an approximate expression based on the extrapolated data and the collected data, An approximate expression is derived based on the interpolated data and the collected data, and by the same operation as in the case of the above-described claim 1, the periodic data at the current time is obtained online from various sampling periodic data corresponding to each process amount. Therefore, the same effects as in the first and second aspects can be obtained.

Therefore, in the fifth embodiment, high-performance plant state monitoring can be realized, there is no great burden on costs, and past accumulated data can be effectively used.

The multivariate data monitoring device according to the present embodiment can be variously modified and implemented without departing from the gist, similarly to the multivariate data monitoring device according to the second embodiment.

(Sixth Embodiment) FIG. 16 is a block diagram showing a configuration example of a data preprocessing device 13 according to the present embodiment. The multivariate data monitoring device 1 is the same as that of the first embodiment as shown in FIG.
Only the data preprocessor 13 will be described. The data preprocessing device 13 includes a sample time generator 130 that outputs a sample time 136 that is a time at which process data for analysis is required, as shown in FIG. 16, and process amount sample data from the process amount collection device 12. A time-series data extractor 631 that extracts target sample data 137 serving as original data of an approximate expression by using the input of the sample data 135 and the sample time 136, an FIR filter designer 632 that designs an FIR filter by using the target sample data 137 as an input, An additional data calculator 6 that outputs estimated target sample data 639 in the shortage time of the process amount sample data using the designed FIR filter 638
33 and the target sample data 137 and the estimation target sample data 639 as inputs, the least square approximation cubic expression 238
From the least squares cubic approximator 232 and the estimated amount calculator 133 that outputs the process amount estimation sample data 139 at the sample time 136 to the feature amount conversion device using the least squares approximated cubic expression 238. Be composed.

The operation of the multivariate data monitoring device 1 configured as described above will be described. The multivariate data monitoring apparatus 1 collects a predetermined monitoring target process amount of the plant 10 at time intervals according to the process amount.
That is, although it is necessary to collect data for each process amount at regular time intervals, this time interval may be changed according to the process amount. Therefore, the process data at the sample time 136 that is the time at which the process data for analysis is required does not always exist in the process amount sample data 135 from the process amount collection device 12.

Therefore, when the time-series data extractor 631 extracts some target sample data 137 from the process amount sample data 135, the amount of collected data required for analysis cannot be obtained, and the corresponding time is determined by an approximate expression. If the data is not located between the two data at the center of the original data, the data at the center of the predetermined number of collected data is extracted from the collected data closest to the corresponding time. , The collected data closest to the time is extracted by the time-series data extractor 531 into the process amount sample data 13.
5 is extracted as target sample data 137.

On the other hand, the target sample data 137 is also supplied to the FIR filter designer 632, and the FIR filter
38 is designed and output to the additional data calculator 633. The additional data calculator 633 is the designed FIR filter 63
8, the sample data 637 to be estimated at the time when the process amount calculation data 135 is insufficient is supplied to the least square cubic approximator 232 as additional data. The least-squares cubic approximator 232 extracts the extracted target sample data 137.
The process amount is approximated to the estimation target sample data 637 by a cubic expression 238 relating to time by a least squares approximation technique. The estimation amount calculator 133 calculates the process amount estimation sample data 139 at the sample time 136 by using the least square approximation cubic expression 238.

Since this specific calculation method is performed in the same manner as in the above-described fifth embodiment according to FIG. 15, the description thereof is omitted.

As another example of the sixth embodiment, the following multivariate data monitoring device can be configured. That is, as the analysis data pre-processing device of the first and second embodiments, when the corresponding time (current time) does not exist inside the original data of the approximate expression in the online analysis, the FIR filter is applied to the collected data, One sample (one piece) of data is extrapolated on a process amount basis for a portion where there is no collected data at the corresponding time, and an approximation formula is derived based on the extrapolated data and the collected data to obtain various values corresponding to each process amount. A multivariate data monitoring device that generates on-line the periodic data at the current time from the sampling periodic data.

In the multivariate data monitoring apparatus, even if the corresponding time (current time) does not exist inside the original data of the approximate expression in the online analysis, the FIR is added to the collected data.
By applying a filter, the portion without collected data at the corresponding time is processed by one sampling (1
2), extrapolating the data of the data, and deriving an approximate expression based on the extrapolated data and the collected data, deriving an approximate expression based on the extrapolated data and the collected data, By generating the cycle data at the current time on-line from various sampling cycle data corresponding to each process amount by the same operation as in the above, the same effect as in the first and second aspects can be obtained.

According to the sixth embodiment, high-performance plant state monitoring can be realized, there is no great burden on costs, and past accumulated data can be effectively used.

The multivariate data monitoring device according to the present embodiment can be variously modified and implemented without departing from the gist, similarly to the multivariate data monitoring device according to the second embodiment.

[0086]

As described above, according to the present invention, by providing the data pre-processing device 13 for converting a large number of process quantities having different sampling periods into data of a predetermined sampling cycle for each process quantity, It is possible to provide a multivariate data monitoring device that enables the use of conventionally accumulated process data having different sampling periods and that does not reduce the data recording efficiency.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration example of a multivariate data monitoring device according to an embodiment.

FIG. 2 is a block diagram showing a configuration example of a data preprocessing device according to the first embodiment of the present invention.

FIG. 3 is an explanatory diagram showing an operation example 1 of the data preprocessing device according to the first embodiment of the present invention.

FIG. 4 is an explanatory diagram showing an operation example 2 of the data preprocessing device according to the first embodiment of the present invention.

FIG. 5 is an explanatory diagram showing an operation example 3 of the data preprocessing device according to the first embodiment of the present invention.

FIG. 6 is an explanatory diagram showing an operation example 4 of the data preprocessing device according to the first embodiment of the present invention.

FIG. 7 is an explanatory diagram showing an operation example 5 of the data preprocessing device according to the first embodiment of the present invention.

FIG. 8 is an explanatory diagram showing an operation example 6 of the data preprocessing device according to the first embodiment of the present invention.

FIG. 9 is a block diagram showing a configuration example of a data preprocessing device according to a second embodiment of the present invention.

FIG. 10 is a block diagram showing a configuration example of a data preprocessing device according to a third embodiment of the present invention.

FIG. 11 is an explanatory diagram showing an operation example of the data preprocessing device according to the third embodiment of the present invention.

FIG. 12 is a block diagram showing a configuration example of a data preprocessing device according to a fourth embodiment of the present invention.

FIG. 13 is an explanatory diagram showing an operation example of a data preprocessing device according to a fourth embodiment of the present invention.

FIG. 14 is a block diagram showing a configuration example of a data preprocessing device according to a fifth embodiment of the present invention.

FIG. 15 is an explanatory diagram showing an operation example of the data preprocessing device according to the fifth embodiment of the present invention.

FIG. 16 is a block diagram showing a configuration example of a data preprocessing device according to a sixth embodiment of the present invention.

FIG. 17 is a block diagram showing a configuration example of a conventional multivariate data monitoring device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 Plant 12 Process amount collection apparatus 13 Data preprocessing apparatus 14 Feature amount conversion apparatus 15 Display device 16 Process data database 18 Feature amount database 19 Feature amount search apparatus 20 Operator 130 Sample time generator 131 Time series data extractor 132 Least square Polynomial approximator 133 Estimator calculator 135 Process amount sample data 136 Sample time 137 Target sample data 138 Least square approximation polynomial 139 Process amount estimation sample data 232 Least square cubic approximator 238 Least square approximation cubic 331 Time series data extractor 431 Time series data extractor 531 Time series data extractor 532 AR model identifier 533 Additional data calculator 537 Estimation target sample data 538 AR model 539 Estimation target sample data 31 time-series data extractor 632 FIR filter design unit 633 adds data calculator 637 estimates the target sample data 638 FIR filter 639 estimation target sample data

 ────────────────────────────────────────────────── ─── Continued on the front page (72) Inventor Masanori Yukitomo 1 Toshiba-cho, Fuchu-shi, Tokyo F-term in Fuchu factory, Toshiba Corporation 5H223 AA02 CC08 EE02 EE06 FF03

Claims (6)

[Claims] 1. A multivariate data monitoring method apparatus which takes in a plurality of process quantities of a plant to be monitored and performs a multivariate analysis, analyzes various sampling cycle data corresponding to each process quantity by using a moving average based on polynomial approximation. A multivariate data monitoring device, comprising: an analysis data preprocessing device that corrects the data to a predetermined sampling cycle data suitable for the data. 2. The multivariate data monitoring device according to claim 1, wherein a cubic approximation is employed as a polynomial approximation used for a moving average in said analysis data preprocessing device. . 3. The multivariate data monitoring device according to claim 1, wherein the analysis data pre-processing device has two data at the center of the original data whose approximate time is approximated due to lack of collected data. If not, a multivariate data monitoring device characterized by selecting the collected data whose time is closest to the two central parts and deriving an approximate expression. 4. The multivariate data monitoring device according to claim 1, wherein the analysis data pre-processing device is configured such that when there is only one of the collected data before and after the relevant time due to a shortage of the collected data. A multivariate data monitoring device characterized in that the selected data closest to the time is selected, an approximate expression is derived, and the data at the time is obtained by extrapolation. 5. The multivariate data monitoring device according to claim 1, wherein the analysis data pre-processing device is configured such that when there is no collected data before or after the time due to lack of collected data, A multivariate data monitoring device characterized by constructing an AR model from collected data, estimating a process amount of a portion having no collected data at a corresponding time, and deriving an approximate expression based on the estimated data and the collected data. 6. The multivariate data monitoring device according to claim 1 or 2, wherein the analysis data pre-processing device is configured such that when there is only one of the data before and after the relevant time due to a shortage of the collected data, A multivariate data monitoring device characterized by applying an FIR filter to collected data, estimating a process amount for a portion having no collected data at a corresponding time, and deriving an approximate expression based on the estimated data and the collected data. .