JPH06110504A - Method and device for estimating characteristic of process, method for monitoring process by using the estimating method and method for controlling process - Google Patents

Abstract

PURPOSE:To estimate the characteristic of a process with high precision even while improving responsiveness and the followup ability in the case of estimating the characteristic of the process. CONSTITUTION:Based on sampling values, multiple regression analysis is performed for each prescribed term by sampling the plural kinds of state amounts x1-xp expressing the states of a process 1 and an output amount (y). Among eccentric recursive coefficients calculated by a first multiple regression analysis part 4, the partial regression coefficient not to be statistically significant is replaced with an alternative partial regression coefficient. A second multiple regression analysis part 9 performs multiple regression analysis excluding variables applied the alternative partial regression coefficient. A second certification part 10 repeats the replacement of the partial regression coefficient and the multiple regression analysis until the partial regression coefficient calculated by the second multiple regression analysis part 9 includes no partial regression coefficient which is not statistically significant. A multiple regression expression making the alternative partial regression coefficient and the partial regression coefficient calculated by the second multiple regression analysis part 9 correspondent to respective variables is adopted as a state equation expressing the characteristic of the process 1.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a process characteristic estimating method and apparatus, and a process monitoring method and process controlling method using the method.

[0002]

2. Description of the Related Art Generally, when a process whose state changes over time, such as a manufacturing plant, is monitored or controlled to a desired state, it is abnormal if the characteristics of the process can be grasped to detect the variation. The occurrence of such a state can be notified or avoided, and the operation amount can be adjusted according to the characteristics of the process, so that stable and accurate control can be performed. Further, when the process includes dead time or delay time, if the monitoring and control are performed based on the predicted value of the state, it is possible to quickly respond to the change.

However, there are many types of variables (manipulation variables, internal state detection variables, control variables, etc.) that represent the state of a process, and there are dependency relationships and interferences among the variables. It is often difficult to understand the characteristics theoretically and accurately. Therefore, in the actual process, the value of the variable representing the state is measured in-process, and the characteristic of the process is estimated by setting the state equation representing the characteristic of the process at each time point based on the measured value of the variable. It is necessary.

As a method for estimating the state of a process based on a large number of variables, a method using multiple regression analysis based on a multiple linear regression model is known. To use multiple regression analysis to estimate the variability of the characteristics of a process, multiple variables are sampled over time to obtain measurements,
After correcting the dead time and delay time for the past multiple measured values for the measurement time point, multiple regression equation is obtained based on each measured value pair, and this multiple regression equation is used as a state equation representing the state of the process. Can be used.

By the way, in an actual process, the characteristics of the process may change from moment to moment due to changes in characteristics of various materials, facility performance, environment, etc. Therefore, in order to make an estimation with good responsiveness and followability. The set of measurement values used for multiple regression analysis must be within the shortest possible time.

[0006]

However, when a set of measured values obtained within a short time is used for multiple regression analysis, a measured value with almost no fluctuation tends to occur, and partial regression corresponding to such a variate is likely to occur. The coefficient is likely to be not statistically significant. In addition, if a partial regression coefficient that is not statistically significant is included in the multiple regression equation, it affects other partial regression coefficients, and the reliability of the predicted value obtained by using the multiple regression equation as the state equation is low. May be.

That is, it is desirable that the set of measured values used for multiple regression analysis be evenly distributed in the widest possible range, but if the set of measured values within a short time is used, the measured values of the variables are relatively stable. However, there is a problem in that the measured value of the desired distribution state cannot be obtained and the accuracy of the estimation of the process characteristics decreases. This will be described in more detail. As shown in FIG. 2, the measured values of the state quantities x _{1 to} x _p , which are the manipulated variables of the process and the detected quantities of the process, and the output amount y, which is the controlled variable of the process (here, the dead time of the measured values and The delay time is assumed to be corrected), and changes with the passage of time. Here, the state quantities x ₁ and x ₂ have a period with a small fluctuation range,
In particular, there is a period in which the state quantity x ₂ hardly changes.
Further, the state quantity x _p has a time-varying period, and within such a period, the measured value hardly changes in each of the sections T ₁ , T ₂ , .... In this way, in the section T ₁ , T ₂ , ... Which obtains the measurement value used for the multiple regression analysis, when the fluctuation range of the measurement value is small or the change of the measurement value is small, it is relatively difficult in the actual process. It occurs frequently. In such a case, the sections T ₁ , T ₂ ,
The distribution of the measured values of the corresponding state quantities x ₁ , x ₂ , x _p in ...... will be biased. Therefore, if a multiple regression analysis based on such measured values is performed, a statistically significant partial regression coefficient cannot be obtained, and the reliability of the result becomes low.

An object of the present invention is to solve the above-mentioned problems, and to make it possible to estimate the characteristics of a process with high accuracy while improving the response and tracking when estimating the characteristics of the process. Another object of the present invention is to provide a method for estimating the characteristics of the process and an apparatus therefor, and a method for monitoring the process and a method for controlling the process using the estimation method.

[0009]

According to a first aspect of the present invention, there is provided a process characteristic estimating method for achieving the above object, wherein measured values of a plurality of kinds of variables representing a process state are sampled with respect to time, A first step of performing multiple regression analysis using a set of variable measurement values obtained by sampling multiple times within a section from the time of measurement to a predetermined time, and partial regression obtained by multiple regression analysis in the first step When some coefficients are not statistically significant, the substitute partial regression coefficient obtained based on the default partial regression coefficient is applied to at least one variable corresponding to the statistically insignificant partial regression coefficient. , Substituting the product of the partial regression coefficient and the measured value of the corresponding variable from the value of the variable that serves as the reference variable to form the intermediate reference variable, and using the intermediate reference variable as the reference variable, the set of the measured values of the remaining variables Due to A second process of performing regression analysis, and a third step of all the partial regression coefficient obtained by the multiple regression analysis in the second step repeats the second step until the statistical significance,
When all partial regression coefficients become statistically significant in the third process, the multiple regression equation in which the partial regression coefficient applied in the second process and the partial regression coefficient obtained in the third process are used as partial regression coefficients for each variable Fourth, the process characteristic is estimated by using
It consists of a process and.

In the second aspect of the present invention, the partial regression coefficient for substitution applied in the second step is the latest partial regression coefficient which is statistically significant. Claim 3
In the invention, the substitute regression coefficient applied in the second process is determined based on the statistical amount of the statistically significant past partial regression coefficient. In the invention of claim 4, the substitute regression coefficient applied in the second step is determined by applying an empirical rule based on the time series of statistically significant past partial regression coefficients and the statistic.

According to a fifth aspect of the present invention, there is provided a process characteristic estimation device to which the above estimation method is applied, the process state detection means for sampling measurement values of a plurality of kinds of variables representing the state of the process with respect to time. A data storage means for storing the sampling value, a first multiple regression analysis means for performing multiple regression analysis using a set of variable measurement values obtained by sampling a plurality of times from a measurement time point to a predetermined time before,
First test means for testing the partial regression coefficient obtained by the first multiple regression analysis means, analysis result storage means for storing the partial regression coefficient obtained by the first multiple regression analysis means, and first test means When a partial regression coefficient that is not statistically significant is detected by, at least one variable corresponding to the partial regression coefficient that is not statistically significant is added to the default partial regression coefficient stored in the analysis result storage means. Intermediate reference variable calculation means for obtaining an intermediate reference variable by subtracting the product of the substitute partial regression coefficient obtained based on the measured value of the corresponding variable from the measured value of the variable serving as the reference variable, and the intermediate reference variable and Intermediate variable data storage means for storing a partial regression coefficient for substitution,
A second multiple regression analysis means for performing a multiple regression analysis using a set of measurement values of the remaining variables with an intermediate reference variable as a reference variable, and a partial regression coefficient obtained by the second multiple regression analysis means are tested to obtain a second Second test means for repeating the calculation by the intermediate reference variable calculation means and the second multiple regression analysis means until all the partial regression coefficients obtained by the multiple regression analysis means of No. When all the partial regression coefficients calculated by the regression analysis means become statistically significant, the partial partial regression coefficient calculated by the intermediate reference variable calculation means and the partial regression coefficient calculated by the second multiple regression analysis means are calculated for each variable. It is provided with a characteristic estimating calculation means for estimating a characteristic of the process by setting a multiple regression equation having a partial regression coefficient as a state equation.

According to a sixth aspect of the present invention, there is provided a process monitoring method, which monitors a change in a process state using a state equation obtained by the process characteristic estimating method according to the first to fourth aspects. A seventh aspect of the present invention is a process control method, wherein the process state is maintained in a prescribed state by using the state equation obtained by the process characteristic estimation method according to the first to fourth aspects. It controls the operation amount of.

[0013]

According to the method of claim 1, when the partial regression coefficient obtained based on the measured values obtained in each section is not statistically significant, the partial regression coefficient is calculated based on the known partial regression coefficient. Since the multiple regression equation obtained by substituting the partial regression coefficient obtained by using is used as the state equation, the accurate partial regression coefficient for the measurement value used for multiple regression analysis at the time of measurement is unknown, but a relatively good partial regression coefficient is used. Therefore, it is possible to suppress deterioration of reliability. Therefore, it is possible to use the measurement values used in the multiple regression analysis within a comparatively short time, and as a result, it is possible to estimate the process characteristics with good responsiveness and trackability even if the values of the variables fluctuate relatively quickly. Can be done.

The method according to any one of claims 2 to 4 is an example of a desirable method of obtaining a partial regression coefficient to be substituted.
In the above method, the latest one of the statistically significant partial regression coefficients in the past is used, so that the influence of the time lag on the characteristic variation of the process can be reduced.
Further, in the method of claim 3, the average value of the statistically significant past partial regression coefficients is used as the partial regression coefficient to be substituted, or the test statistic is most suitable for the threshold value that divides the rejection area from the acceptance area. Since the partial regression coefficient on the significant side is used, it is possible to reduce the influence of variations in process characteristic variations. Furthermore, according to the method of claim 4, the empirical rule is applied to determine the partial regression coefficient to be substituted based on the statistically significant time series of partial regression coefficients in the past and the test statistic. It is possible to select a desirable partial regression coefficient that is small in both influences of variation in fluctuation and time lag.

A fifth aspect of the present invention is the configuration of an apparatus for estimating the characteristics of a process according to the above method. That is, the measurement value of each variable of the process is sampled and stored in the data storage means, and the result of performing the multiple regression analysis by the first multiple regression analysis means by the data stored in the data storage means is the first test. Test by means to determine whether each partial regression coefficient is statistically significant. The partial regression coefficient obtained by the first multiple regression analysis means is stored in the analysis result storage means, and the partial regression coefficient not statistically significant with respect to the partial regression coefficient stored in the analysis result storage means is analyzed result storage means. The partial regression coefficient obtained based on the known partial regression coefficient stored in is substituted. Further, the intermediate reference variable calculation means subtracts the product of the partial regression coefficient of substitution and the measured value of the corresponding variable from the value of the variable serving as the reference variable to obtain the intermediate reference variable, and the intermediate reference variable and the substitute partial deviation are calculated. The regression coefficient is stored in the intermediate variable data storage means. Using the intermediate reference variable thus obtained as a reference variable, the second multiple regression analysis means performs a multiple regression analysis using a set of measurement values of the remaining variables, and calculates the partial regression coefficient obtained by the second multiple regression analysis means. The calculation by the intermediate standard variable calculation means and the second multiple regression analysis means is repeated until all partial regression coefficients obtained by the second multiple regression analysis means are statistically significant after being tested by the second verification means. . When all the partial regression coefficients calculated by the second multiple regression analysis means become statistically significant, the partial partial regression coefficient calculated by the intermediate reference variable calculation means and the partial regression coefficient calculated by the second multiple regression analysis means The characteristic of the process is estimated by setting a multiple regression equation with a partial regression coefficient for each variable as a state equation. After all, as in the method of claim 1, since the multiple regression equation is set only by the statistically significant partial regression coefficient and this multiple regression equation is used for the state equation for estimating the characteristic of the process, a relatively good partial regression is obtained. The coefficient can be used to reliably estimate the process characteristics.
As a result, even if the multiple regression analysis is performed using the measured values within a relatively short time, the reliability does not decrease, and the process corresponding to the relatively rapid fluctuation of the variable can be accurately estimated.

According to the method of claim 6, the characteristics of the process are monitored by the state equation obtained as described above, and it is possible to quickly detect the progress to an abnormal state or the occurrence of an abnormal state. , It becomes possible to report an abnormality without delay. Also, in processes that include dead time and delay time, it is possible to predict the fluctuation of the reference variable by applying the state equation to the current value of each variable to obtain the predicted value, and to predict the abnormality. It is also possible to do.

According to the method of claim 7, the characteristic of the process is controlled by the state equation obtained as described above, and an abnormal state is avoided or the characteristic is controlled to stabilize the quality of the product. You can Also, in processes that include dead time and delay time, it is possible to apply a state equation to the current value of each variable to obtain a predicted value, so predict the fluctuation of the reference variable and adjust the manipulated variable. Thus, stable control with good responsiveness can be performed.

[0018]

【Example】

(Embodiment 1) In this embodiment, a multiple regression equation in which the manipulated variable that is an input to the process and the detected amount of the state of each part in the process are used as explanatory variables and the reference variable of the controlled variable that is the output amount of the process is used. Then, the characteristic of the process is estimated by using this multiple regression equation as a state equation. That is,
The predicted value of the output amount (control amount) is obtained by using the manipulated variable, control amount, and detected amount of the process as variables, and the behavior of the partial regression coefficient is investigated. These variables are sampled at a predetermined time interval, and a value obtained by correcting a time shift such as a dead time or a delay time is used as a measurement value. In the following description, the measured value is treated as having been corrected for the time lag.

The measured values obtained by correcting the time lag between the state quantities x _{1 to} x _p , which are the manipulated variables of the process and the detected quantities of the process, and the output amount y, which is the controlled quantity of the process, are as shown in FIG. Fluctuates with the passage of. A specific range of the output amount y of the process is defined by the upper limit value UL and the lower limit value LL with respect to the target value DV, and the process is feedback-controlled so that the output amount y does not deviate from the predetermined range. . However, the characteristics of the process may change,
When the disturbance becomes large, an abnormality may occur in which the output amount y deviates from the predetermined range. In the present embodiment, the occurrence of such an abnormality is monitored to notify that an abnormal state has occurred, or control for avoiding the abnormality is performed.

That is, as shown in FIG. 1, in order to analyze the characteristics of the process at a specific time, the process 1
State quantities x _{1 to} x _p obtained from the sensor output, etc.
The output amount y and the output state y are sampled at predetermined time intervals in the process state detection unit 2, and the measured value with the time difference corrected is stored in the data storage unit 3. With respect to the measured values stored in the data storage unit 3 in this way, as shown in FIG. 2, the past predetermined sections T ₁ , T ₂ , ... (Predetermined time interval s ₁ , s ₂ , ... In general, T ₁ = T ₂ = ...
... = constant), multiple regression analysis is performed by the first multiple regression analysis unit 4 using a plurality of sets of measurement values, and each section T ₁ , T ₂ ,.
For each ..., obtain the multiple regression equation as shown below. Y _j = α _j0 + α _j1 x _j1 + α _j2 x _j2 + ... + α _jp x _jp (1) Here, the intervals T ₁ , T ₂ , ... are set to the shortest possible time, and the fluctuation of the process characteristics is set. It responds to and can respond quickly. Y _j is the output value y
_{It is} a predicted value of j, and the subscript j indicates that it is a value at the measurement time point s _j obtained using the measurement value of the section T _j .

When the multiple regression equation is obtained, the obtained predicted value Y
_j , partial regression coefficients α _j0 , α _j1 , ..., α _jp are stored in the analysis result storage unit 5. Therefore, the predicted value Y obtained in the past
_ji and partial regression coefficient α _{(ji) 0} , α _{(ji) 1} , ……, α
_{(ji) p is} also stored in the analysis result storage unit 5. The analysis result storage unit 5 stores the predicted value Y _ji and the partial regression coefficients α _{(ji) 0} , α of the measurement time point s _j for a certain number of past times (for example, ten times), that is, within a past predetermined period.
_{(ji) 1} , ..., α _{(ji) p} are stored. That is, if the past ten times are stored, i = 1, 2, ...
…, 10

Next, the partial regression coefficients α _j0 , α _j1 , ..., α _jp
Is tested by the first test unit 6 to determine whether it is statistically significant. That is, it is determined whether the obtained partial regression coefficients α _j0 , α _j1 , ..., α _jp are statistically significant by _obtaining a test statistic and comparing it with a separately determined threshold. Various types of test statistics are known here, but in this embodiment, the t value known as the test statistic using the t distribution is used. In the case of t-value, the larger the absolute value, the higher the probability of being statistically significant.

When the first test unit 6 _judges that all the partial regression coefficients α _j0 , α _j1 , ..., α _jp are statistically significant, the partial regression coefficients α _j0 , α _j1 , ......, estimates the state of the process 1 by using a multiple regression formula using the alpha _uk as a state equation of the monitoring control unit 12. That is, by _obtaining the predicted value y _j of the output amount by the equation (1) and checking the behavior of the partial regression coefficients α _j0 , α _j1 , ..., α _jp , an alarm output is issued for the abnormal state of the process 1. It can be generated and notified, and the process 1 can be controlled.

On the other hand, when the first test unit 6 determines that even one of the partial regression coefficients α _j0 , α _j1 , ..., α _jp is not statistically significant, the intermediate reference variable calculation unit At 7, the following calculation is performed. That is, at least one state quantity x _jk (kε {n}: {n} is a set of n) corresponding to the statistically insignificant partial regression coefficient α _jn (n is an indefinite natural number) is stored as an analysis result. Partial regression coefficient α _{(ji) k} (i = 1, 2, ...
...), the latest partial regression coefficient among the statistically significant partial regression coefficients corresponding to the state quantity x _jk is used as a partial regression coefficient β _jk for substitution. For example, as shown in FIG. 3, when the absolute value of the t value that is the test statistic of the partial regression coefficient α _jk at the measurement time point s _j does not exceed the threshold value t _s , the t value in the past predetermined period L The latest partial partial regression coefficient α _{(j−2) k} among those that exceed the threshold t _s is adopted as the partial partial regression coefficient β _jk for substitution. It should be noted that even when the t value is negative, the same determination can be made by using the absolute value.

After obtaining the substitute partial regression coefficient in this way,
The sum of the products of the state quantity x _jk and the substitute partial regression coefficient β _jk is calculated. For example, the partial regression coefficients α _j1 , α _j2 , and α _jp are not statistically significant, and all corresponding state quantities x _j1 , x _j2 , x
_If the partial regression coefficients β _j1 , β _j2 , and β _jp are used for _jp , the following values are obtained. B _j = β _j1 x _j1 + β _j2 x _j2 + ... + β _jp x _jp (2) Further, the value obtained by subtracting the value B _j obtained in the equation (2) from the output value y _{j is} expressed by the following equation: Let the intermediate reference variable z _j . z _j = y _j −B _j (3) The intermediate reference variable calculation unit 7 calculates the formula (3) for all measured values in the section T _j corresponding to the current measurement time point s _j. Then, an intermediate reference variable z _j is obtained. The intermediate reference variable z _j is stored in the intermediate variable data storage unit 8.

Next, in the second multiple regression analysis unit 9, the state quantities x _j1 , x corresponding to the substitute partial regression coefficients β _j1 , β _j2 , β _jp.
State quantities x _j3 , ..., X in the section T _j excluding _j2 and x _jp
and _{j (p-1),} using an intermediate reference variable z _j, performing a multiple regression analysis again with respect to interval T _j, obtaining the following multiple regression equation. Z _j = γ _j0 + γ _j3 x _j3 + γ _j4 x _j4 + ... + γ _{j (p-1)} x _{j (p-1)} (4) where Z _j is the predicted value of the intermediate reference variable z _j is there.

Partial regression coefficient γ thus obtained_j0, Γ
_j3, ……, γ_{j (p-1)}For the first verification unit 10
The test is performed in the same manner as the test unit 6 of the
Partial regression coefficient obtained by comparing the magnitude relationship with the threshold of
γ_j0, Γ_j3, ……, γ_{j (p-1)}Is statistically significant
To determine. Where the partial regression coefficient γ_j0, Γ_j3, ……,
γ _{j (p-1)}Contains something that is not statistically significant
The partial regression coefficient that substitutes the partial regression coefficient
And repeat the calculation from the intermediate reference variable calculator 7.
And calculate until all partial regression coefficients are statistically significant
repeat.

Substitute partial regression coefficients β _j1 , β _j2 , stored in the intermediate variable data storage unit 8 obtained as described above,
β _jp and partial regression coefficient γ obtained by the second multiple regression analysis unit 9
_j0 , γ _j3 , ..., γ _{j (p-1)} corresponding state quantities x _{j1 to} x _jp
A multiple regression equation such as the equation (5) is set as the partial regression coefficient for, and the following equation is calculated in the characteristic estimation calculation unit 11 by using this multiple regression equation as the state equation. Y _j = Z _j + B _j (5) (= γ _j0 + γ _j3 x _j3 + γ _j4 x _j4 + ... + γ _{j (p-1)} x _{j (p-1)} + β _j1 x _j1 + β _j2 x _j2 + ... (+ Β _jp x _jp ) That is, the equation (5) is used as a state equation showing the state of the process 1, and the supervisory control unit 12 performs supervisory control using the predicted value Y _j and partial regression coefficient. Here, in the above example, it is assumed that all partial regression coefficients become statistically significant at one time in the second test unit 10, but when it does not become statistically significant at one time, As a matter of course, the number of substitute partial regression coefficients increases.

The monitor control unit 12 monitors the variation of the coefficient value of each variable in the state equation and compares it with the past coefficient value to detect an abnormal state or determine the location of the abnormality. Further, each manipulated variable can be quickly adjusted according to each coefficient of the state equation so that the output of the process 1 approaches the target value, and the responsiveness and accuracy of control are improved.

In the process 1 including the dead time and the delay time, the predicted value for the output value of the process 1 can be obtained by giving the measured value at the time of measurement to the state equation. It becomes possible to control each operation amount with high responsiveness before it occurs. By the way, as a method of determining the partial regression coefficient β _jk as a substitute, in addition to the method described above, the average value of the statistically significant partial regression coefficients among the partial regression coefficients within the past predetermined period corresponding to the same variable. _May be used as a substitute partial regression coefficient β _jk . For example, according to the example of FIG. 3, regarding the partial regression coefficients α _{(j-2) k} , α _{(j-8) k} , α _{(j-9) k} , α _{(j-10) k in} the period L. t
Since the value exceeds the threshold t _s , (α _{(j-2) k} + α _{(j-8) k}
＋ α _{(j-9) k} ＋ α _{(j-10) k} ) / 4 substitute partial regression coefficient β _jk
Is used as. Also, within the past predetermined period, t
For each partial regression coefficient whose value exceeds the threshold t _s , the partial regression coefficient β for which the weighted average value with the weight of the t value as a substitute is used.
_It may be adopted as _jk . Further, for each partial regression coefficient whose t value exceeds the threshold value t _s within a predetermined period in the past,
A weighted average value using a weight based on the time difference from the measurement time point s _j may be used as the substitute partial regression coefficient β _jk .

Further, the size of the test statistic (for example, t value) is compared with respect to the partial regression coefficient within a predetermined period in the past, and the partial regression coefficient whose statistical test value is the most significant side with respect to the threshold value is determined. The partial regression coefficient β _jk may be adopted as a substitute. For example, in the example of FIG.
_{Since the} t value for _{(j-9) k} is the maximum, this partial regression coefficient is adopted as the partial regression coefficient β _jk as a substitute.

In the above embodiment, the state quantities x _{1 to} x _p to be subjected to the multiple regression analysis are not limited to the measured values directly obtained from the process 1, but the predetermined four arithmetic operations, differential operations,
It is also possible to use a value processed by performing integration calculation, filtering, or the like, use a difference in variable at a predetermined time interval, or use a value obtained by combining measurement values from the measurement time point to a predetermined time period before combining. Although the t value is used as the test statistic, a test statistic that follows the F distribution can be used.

(Embodiment 2) In the present embodiment, an example of determining a substitute partial regression coefficient β _jk by integrating the time series of past partial regression coefficients and the statistics of past partial regression coefficients by an empirical rule Indicates. Here, of the partial regression coefficients within a predetermined period in the past, the latest partial regression coefficient that is statistically significant and the partial regression coefficient having the maximum t value are associated with each other. To determine the partial regression coefficient β _jk for substitution. That is,
In the example of FIG. 3, the latest partial regression coefficient is α _{(j-2) k} , and t
Since the partial regression coefficient having the maximum value is α _{(j-9) k} , the difference in t value Δt _j between the two partial regression coefficients α _{(j-2) k} and α _{(j-9) k.}
(= T _{(j-2) k-} t _{(j-9) k} ) and the partial regression coefficient α _{(j-2) k} ,
Difference ΔL _j (= s _(j-2) − at the time of measurement when α _{(j-9) k} was obtained
s _(j-9) ) as the antecedent part, and the coupling coefficient f (0 ≦ f
The coupling coefficient f by fuzzy inference with ≦ 1) as the consequent part
And the substitute partial regression coefficient β _j is determined by the following equation. β _j = (1-f) · α _{(j-2) k} + f · α _{(j-9) k} Thus, the latest partial regression coefficient α _{(j-2) k} is (1-f)
And the partial regression coefficient α _{(j-9) k} having the maximum t value is multiplied by f to determine the partial regression coefficient β _jk as a substitute.

The coupling coefficient f is determined as follows. First, the difference in t value Δt _j and the difference in measurement time ΔL _j are:
A membership function as shown in FIG. 5 is applied. Here, the difference in t value Δt _j and the difference in measurement time ΔL _j are S
The labels are classified by three levels of labels (small), M (medium), and B (large), and the label of the coupling coefficient f is determined according to the rule map arranged as shown in FIG. Regarding the coupling coefficient f, the labels are VS (minimum), S (small), M (medium), and B.
(Large), VB (maximum) is classified into five stages, as shown in FIG.
Apply the membership function shown in to determine the definite value by fuzzy operation.

As described above, the membership function,
An empirical rule can be incorporated into the rule map and the method of determining a definite value. The partial regression coefficient β _{jk can} be substituted by _combining the time series of partial regression coefficients and the statistics of past partial regression coefficients with empirical rules.
Can be determined. The other configuration is the first embodiment.
Is the same as. (Third Embodiment) In the second embodiment, fuzzy inference is performed in order to obtain the coupling coefficient f of the two partial regression coefficients, but in this embodiment, the t value exceeds the threshold value t _s within the past predetermined period. An example is shown in which a weighted average value is adopted as a substitute partial regression coefficient β _jk for each partial regression coefficient, and the weighting coefficient for obtaining the weighted average value is determined by fuzzy inference.

That is, a rule map is constructed with the antecedent part of the magnitude of the t value and the time difference between the measurement time point s _j and the time point at which each t value is obtained, and the weighting factor is used as the consequent part for fuzzy calculation. The determined value of the weighting factor is obtained by.
By determining the weighting coefficient in this manner, the empirical rule can be incorporated into the weighted average value, and a value considered desirable as the partial regression coefficient β _j can be selected. Other configurations are similar to those of the first embodiment.

When fuzzy inference is used as in the second and third embodiments, the partial regression coefficients α _j0 , α _j1 , ..., α _jp obtained by the first multiple regression analysis unit 4 are The partial regression coefficients α _j0 , α are also set for those that are statistically significant, rather than performing fuzzy inference that gives a substitute partial regression coefficient β _jk only for those that are not statistically significant.
_j1, ......, by incorporating information about alpha _uk fuzzy inference, partial regression coefficients α _j0, α _j1, ......, it is possible to correct the alpha _uk to a more desirable value. For example,
_Assuming that the partial regression coefficient α _jk is statistically significant, the measurement time point s _j is replaced with the latest partial regression coefficient in the past in the second embodiment.
Information about the partial regression coefficient α _jk of is used, and in the third embodiment, in addition to the information about the partial regression coefficient in the past, the measurement time point s _j
The information about the partial regression coefficient α _jk of is used.

[0038]

As described above, according to the present invention, when the partial regression coefficient obtained based on the measurement values obtained in each section is not statistically significant, this partial regression coefficient is converted into a known partial regression coefficient. Since the multiple regression equation substituting the partial regression coefficient obtained based on this is used as the state equation, a relatively good partial regression coefficient can be used, and the decrease in reliability can be suppressed. That is, it is possible to use the measurement values used in the multiple regression analysis within a relatively short time, and as a result, it is possible to estimate the process characteristics with good responsiveness and trackability even if the values of the variables fluctuate relatively quickly. There is an effect that can be performed.

Further, by monitoring and controlling the process according to the state equation obtained as described above,
There are advantages that it is possible to promptly notify the occurrence of an abnormal state, avoid the occurrence of an abnormal state, and control the characteristics to stabilize the product quality. In particular, in a process that includes dead time and delay time, the state equation can be applied to the current value of each variable to obtain the predicted value.
There is an advantage that it is possible to predict the occurrence of an abnormal state by predicting the fluctuation of the reference variable and to perform stable control with good responsiveness by predictively adjusting the manipulated variable.

[Brief description of drawings]

FIG. 1 is a block circuit diagram showing a first embodiment.

FIG. 2 is an operation explanatory diagram showing an example of a change in a variable obtained from the process in the first embodiment.

FIG. 3 is an operation explanatory diagram illustrating an example of a relationship between a partial regression coefficient and a t value according to the first embodiment.

FIG. 4 is a diagram showing a membership function applied to the antecedent part of fuzzy inference in Example 2;

FIG. 5 is a diagram showing a rule map applied to fuzzy inference in a second embodiment.

FIG. 6 is a diagram showing a membership function applied to a consequent part of fuzzy inference in Example 2;

[Explanation of symbols]

1 process 2 process state detection unit 3 data storage unit 4 first multiple regression analysis unit 5 analysis result storage unit 6 first test unit 7 intermediate reference variable calculation unit 8 intermediate variable data storage unit 9 second multiple regression analysis unit 10 second test 11 characteristic estimating arithmetic unit 12 monitors the control unit x ₁ ~x _p state quantity y output quantity

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Masaki Inoue 1048, Kadoma, Kadoma, Osaka Prefecture Matsushita Electric Works Co., Ltd.

Claims (7)

[Claims] 1. A set of measured values of variables obtained by sampling the measured values of a plurality of kinds of variables representing the state of a process with time and sampling a plurality of times within a section from a measurement time point to a predetermined time before. Corresponding to a partial regression coefficient that is not statistically significant when there is a statistical regression that is not significant in the first process in which multiple regression analysis is performed using the multiple regression analysis in the first process For at least one variable, apply the partial regression coefficient for substitution obtained based on the default partial regression coefficient, and calculate the product of the partial regression coefficient for substitution and the measured value of the corresponding variable from the value of the variable as the reference variable. The second step of performing a multiple regression analysis using a set of measurement values of the remaining variables with the intermediate reference variable subtracted and the intermediate reference variable as the reference variable, and all the deviations obtained by the multiple regression analysis in the second step. Regression coefficient The third process, in which the second process is repeated until it becomes statistically significant, and when all partial regression coefficients become statistically significant in the third process, the substitute partial regression coefficient applied in the second process and the third process are obtained. And a partial regression coefficient for each variate, and a multiple regression equation for estimating a process characteristic as a state equation. 2. The process characteristic according to claim 1, wherein the substitute partial regression coefficient applied in the second step is the latest partial regression coefficient which is statistically significant. Estimation method. 3. The characteristic estimation of the process according to claim 1, wherein the substitute regression coefficient applied in the second step is determined based on a statistical amount relating to a statistically significant past partial regression coefficient. Method. 4. The substitute regression coefficient applied in the second process is determined by applying an empirical rule based on a statistically significant time series of partial regression coefficients in the past and statistics. The method for specific estimation of a process according to claim 1. 5. A process state detecting means for sampling the measured values of a plurality of kinds of variables representing the state of the process with time, a data storing means for storing the sampled values, and a plurality of times from the time of measurement to a predetermined time before. A first multiple regression analysis means for performing a multiple regression analysis using a set of variable measurement values obtained by sampling, and a first testing means for testing a partial regression coefficient obtained by the first multiple regression analysis means. ,
It is not statistically significant when the analysis result storage means for storing the partial regression coefficient obtained by the first multiple regression analysis means and the partial regression coefficient which is not statistically significant are detected by the first test means. For at least one variable corresponding to the partial regression coefficient, the product of the substitute partial regression coefficient obtained based on the predetermined partial regression coefficient stored in the analysis result storage means and the corresponding measured value of the variable is used as the reference variable. Intermediate reference variable calculation means for obtaining the intermediate reference variable by subtracting from the measured value of the variable, intermediate variable data storage means for storing the intermediate reference variable and the partial regression coefficient for substitution, and the intermediate reference variable as the reference variable The second multiple regression analysis means for performing multiple regression analysis by the set of the remaining measured values of the variables and the partial regression coefficient obtained by the second multiple regression analysis means were tested to obtain by the second multiple regression analysis means. Everything Second test means for repeating the calculation by the intermediate reference variable calculation means and the second multiple regression analysis means until the partial regression coefficient of the above becomes statistically significant, and all the deviations obtained by the second multiple regression analysis means. When the regression coefficient becomes statistically significant, a multiple regression equation in which the substitute partial regression coefficient obtained by the intermediate standard variable calculating means and the partial regression coefficient obtained by the second multiple regression analyzing means are used as the partial regression coefficient for each variable A process characteristic estimating device comprising: a characteristic estimating calculation means for estimating a characteristic of a process by setting it as a state equation. 6. A method for monitoring a process, wherein a change in the state of the process is monitored by using the state equation obtained by the method for estimating the characteristic of the process according to any one of claims 1 to 4. 7. An operation amount of a process is controlled so that the state of the process is maintained in a prescribed state by using a state equation obtained by the process characteristic estimating method according to any one of claims 1 to 4. Control method of the process.