JP2551166B2 - Process control equipment - Google Patents

Description

The present invention relates to a process control device for controlling a process, and more particularly to a process control device having a function of automatically adjusting a control constant to an optimum value.

<Prior Art> In order to control a process to be controlled under optimum conditions, it is necessary to obtain dynamic characteristics of this process and successively obtain control constants for calculating a manipulated variable. As such a process control device, open loop type, limit sensitivity type, response waveform type, and adaptive control type process control devices are known.

Figure 12 shows the configuration of the open-loop type process controller. In normal control, the switch 1 is set to the control calculation unit 2 side. The set amount and the process amount are input to the deviation calculation unit 3 to take their deviations, and the control calculation unit 2 calculates the manipulated variable by PID calculation or the like and outputs it to the process 4. Periodically or when the deviation becomes large, the switch 1 is moved to the identification signal generator 5 side, and a step signal or a pulse signal is applied to the process 4. From the response of process 4 at this time
A control constant is calculated by the Ziegler-Nichols method or the like and set in the control calculation unit 2.

Fig. 13 shows the configuration of a process control device of the limit sensitivity type. Output of control calculation unit 2 and limit cycle generation unit 6
Are added by the adder unit 7 and output to the process 4 as a manipulated variable. The limit cycle generator 6 periodically fluctuates the operation amount input to the process 4 to the maximum value and the minimum value, and from the limit cycle and the limit sensitivity at that time, the Ziegler-Nicho
A control constant is calculated by the ls method or the like and set in the control calculation unit 2.

FIG. 14 shows the configuration of the process controller of the response waveform diagram, in which the set amount and the deviation are input to the response waveform rule 8.
The response waveform rule 8 is composed of a tuning rule such as a fuzzy rule. The rule is applied to the overshoot of the process amount, the damping ratio, the period, etc. to obtain the control constant, which is set in the control calculation unit 2. .

Figure 15 shows the configuration of an adaptive control type process controller. The output of the control calculator 2 and the output of the identification signal generator 9 are added by the adder 10 and input to the process 4. Also,
The output of the process amount and addition unit 10 is input to the dynamic characteristic identification unit 11, the control constant is estimated, and the result is output to the control calculation unit 2. That is, a slight disturbance is generated in the identification signal generator 9, and a control constant is obtained from the dynamic characteristic of the process amount at that time and set in the control calculator 2.

<Problems to be Solved by the Invention> However, such a process control device has the following problems. In the open loop type and the adaptive control type, an identification signal must be given to the process, and in the limit sensitivity type, a continuous oscillation must be generated. That is, since a disturbance must be given to the process, there are restrictions when trying to obtain the control constant during the operation. The response waveform type requires several times of waveform observation in order for the control constants to converge, which requires a certain amount of time.

Therefore, none of the control devices could freely reset the control constant.

<Object of the Invention> An object of the present invention is to provide a process control device capable of resetting a control constant in a short time and without giving a disturbance to the process.

<Means for Solving the Problems> In order to solve the above problems, according to the present invention, an operation amount and a process amount to be output to a process are collected by a data collection unit for a predetermined time so that the collected operation amount and the collected process amount are matched. Adjust the internal parameters of the process model in the modeling section,
And the accuracy of this process model is calculated by the estimated accuracy calculation unit, and the modeling unit is activated when the manipulated variable or process amount fluctuates by a predetermined value or more, and the outputs of this modeling unit and the estimated accuracy calculation unit. The control constant of the control calculation unit is obtained from

Further, the gain constant of the process model is determined from the output of the process model and the process amount.

Further, the simplex method is used for the search, the length of the side component of the initial polygon is made longer than the sampling period collected by the data collection unit, and after the first convergence, the simplex is recreated at that point. It is the one.

Further, a stepwise operation amount is applied to the process at the beginning of control, and the sampling period and the initial values of the control constants are obtained from the response of the process amount at that time.

<Operation> The control constant can be obtained in a short time without giving disturbance.

<Embodiment> FIG. 1 shows an embodiment of the process control apparatus according to the present invention. In FIG. 1, reference numeral 20 denotes a control calculation unit, which inputs a set amount which is a target value and a process amount which is an output of the process 21, and performs a proportional integral differential calculation on these deviations to calculate a manipulated variable. 21 is a process to be controlled,
The operation amount, which is the output of the control calculation unit 20, is input, and the process amount is output. A data collection unit 22 receives an operation amount and a process amount, and holds these data as time series data for a certain period. 23 is a modeling section,
A preprocessing unit 231, a process model 232, an internal parameter adjusting unit 233, and a comparing unit 234 are included therein. The output of the data collection unit 22 is subjected to preprocessing such as filtering by the preprocessing unit 231, the operation amount is input to the process model 232, and the process amount is input to the comparison unit 234. The process model 232 is a simulation of the process 21. The output of the process model 232 is input to the comparison unit 234 and compared with the separately input process amount. The comparison result is input to the internal parameter adjusting unit 233. The internal parameter adjusting unit 233 adjusts the internal parameters of the process model 232 so that the output of the process 21 and the output of the process model 232 match. An estimated accuracy calculation unit 24 receives the output of the modeling unit 23 and calculates the certainty of the process model 232. 25 is a control constant calculator, which is an internal parameter adjuster.
The outputs of 233 and the estimation accuracy calculation unit 24 are input, and the control constants are calculated from these outputs and output to the control calculation unit 20.
The control calculator 20 calculates the manipulated variable based on this new control constant. Reference numeral 26 is a calculation command unit, which inputs a process amount and an operation amount, and when these change by a predetermined value or more, operates switches 27 to 30 to output a calculation command to the modeling unit 23 and Starts action 23.

Next, the operation of this embodiment will be described based on the flowchart of FIG. Note that, regardless of this operation, the data collection unit 22 collects the operation amount and the process amount as time series data at regular intervals. In FIG. 2, the calculation command unit 26 calculates the process amount (PV) and the operation amount (MV) at regular intervals.
, And determine if their variations are greater than a predetermined value. If the fluctuation is small, the process ends without issuing a calculation command. This judgment is performed by calculating the area (hatched portion) by integrating the deviations of the process amount and the manipulated variable from the steady state as shown in FIG. 3, for example. When the fluctuation is large, the modeling unit 23 causes the process amount and operation amount time series data {PV}, {M} collected by the data collecting unit 22 to be collected.
V} is read and the preprocessing unit 231 performs filtering to remove stationary components and noise. This filtering is performed, for example, by the following calculation, where PV _F is the time series data of the process amount and PV ₀ is the stationary component.

PV _F = (1-α) · PV _F + α · (PV-PV ₀ ) α: Filter constant Time series data PV _F and MV _F of filtered process amount and manipulated variable are input to process model 232 and model output Is calculated. This operation is different by the process model, the model outputs the n-th and O _n, for example, _{O n = β M · O n} -1 + (1-β M) · K M · MV F (n-L M) (1) K _M : gain of process model L _M : dead time of process model β _M : first-order lag coefficient of process model MV _F (n−L _M ): by calculation of the value of MV _F before L _M time Done. This operation is reached or the maximum number of iterations is predefined, or output O _n of process model 232 is performed until close enough to the time series data PV _F process variable. Number of repetitions does not reach the maximum number of iterations, and the difference between the time series data PV _F output O _n and process of the process model 232 as shown in FIG. 4 is not smaller, so that this difference becomes small ( That is, the gain K _M , dead time L _M , and first-order lag coefficient β _M of the process model 232 are adjusted so that the process model 232 is activated again so that the process model 232 moves in the direction of the arrow in FIG. The maximum number of iterations has been reached, or the process model 232 output O _n and process amount time series data PV _F
When the difference between the two is sufficiently small, the estimated accuracy calculator 24 calculates the accuracy K of the process model 232. This accuracy K is obtained by the following equation (2) or equation (3), for example.

K = 1- [Σ {PV _F (i) -O (i)} ² / ΣPV
_F (i) ² ] (2) K = Σ {PV _F (i)} ² / Σ {O (i)} ² (3) where PV _F (i) and O (i) are respectively It is the time series data of the process amount at the time point i and the output of the process model 232. If the estimated accuracy is higher than a predetermined value, the control constant calculator 25 calculates the proportional, integral, and differential constants, that is, the control constants from the obtained first-order lag process model 232 by the Ziegler-Nichols method. The control calculator 20 calculates the manipulated variable using the newly obtained control constant. It should be noted that the process model does not necessarily have to be the first-order lag system as shown in the equation (1). Also, if the process 21 has a measurable disturbance, such as a load change, then this disturbance may be used in the model calculation.

In the embodiment of FIG. 1, the internal parameters of the process model 232, that is, the gain K _M , the dead time L _M , and the first-order lag coefficient β _M must all be changed little by little and the calculation must be repeated. There is a problem that it takes a lot of calculation time. An example of solving this problem will be described below. In this embodiment, the gain K _M is obtained from the integral ratio of the process amount and the output of the process model when the process gain is set to 1, and the dead time L _M and the temporary delay coefficient β _M are obtained using the iterative search method. It was done. FIG. 5 shows the configuration of this embodiment. However, the data collection unit 22
Since parts other than the modeling part are the same as those in FIG. 1, description thereof will be omitted. In FIG. 5, reference numeral 31 is a modeling unit, which has the same function as the modeling unit 23 in FIG. 31
Reference numeral 1 denotes a pre-processing unit, which has the same function as the pre-processing unit 231 in FIG. That is, the process amount and the operation amount data held in the data collection unit 22 are filtered to remove the stationary noise. Reference numeral 312 denotes a process model, which simulates an actual process like the process model 232. The operation amount preprocessed by the preprocessing unit 311 is input to the process model 312. 313 is a gain calculation unit, which is a process model 3
12 outputs and process quantities are input and process model 31
Calculate the gain of 2. Reference numeral 314 is a comparison unit, which inputs the output of the process model 312 and the process amount and compares these values. Reference numeral 315 denotes an internal parameter adjustment unit, which receives the output of the comparison unit 314 and the output of the gain calculation unit 313 and changes the internal parameters of the process model 312 so that the difference between the output of the process model 312 and the process amount becomes sufficiently small. To The output of the internal parameter adjusting unit 315 is input to the estimation accuracy calculating unit 24 and the control constant calculating unit 25 shown in FIG. 1, and the control constant of the control calculating unit 20 is determined. The calculation command unit 26 controls the switches 32-35.

Next, the operation of this embodiment will be described with reference to FIG. First, the calculation command unit 26 shown in FIG. 1 periodically reads the operation amount and the process amount, and only when the fluctuations are large, the following processing is performed. That is, the pre-processing unit 311 filters the time series data {PV} and {MV} of the operation amount and the process amount to remove noise for a stationary portion, and then the process model 312 calculates the output of the process model. In this calculation, the process gain K _{M is set} to 1 and is performed by the following equation (4).

_{O n = β n · O n} -1 + (1-β n) · MV F (n-L M) ...... (4) provided that the meaning of each coefficient is the same as above (1). Next, the gain calculation unit 313 calculates the process gain K _{M according} to the following equation (5).

K _M = [Σ {O (n) · PV _F (n)} / {ΣO (n) ² }
(5) However, O (n) is the value obtained by the equation (4), that is, the output of the process model 312 when the process gain K _M is 1. This operation reaches a predefined maximum number of iterations, or the process model 31
2 output O _n and process gain calculated by gain calculation unit 313
It is performed until the product of K _M approaches the time series data PV _F of the process quantity sufficiently. If none of these conditions are met,
The dead time L _M and the first-order delay coefficient β _M of the process model 312 are adjusted so that this difference becomes small, the process model 312 is started again, and the output is calculated by the equation (4).
The operation of calculating the process gain K _{M according} to the equation (5) is repeated. When the condition is satisfied, the estimation accuracy calculation unit 24
When the accuracy K of the process model 312 is calculated and the estimated accuracy is smaller than the predetermined value, the control constant calculation unit
From the process model 312 of the first-order lag model obtained, 25 is proportional, integral, differential constant, by the Ziegler-Nichols method,
That is, the control constant is calculated. The control calculator 20 calculates the manipulated variable using the newly obtained control constant. The internal parameter adjusting unit 315 is configured to obtain the dead time L _M and the first-order delay coefficient β _M by the iterative search method using the simplex method. Also, if the process gain K _M is needed at the beginning of the iteration, then the previous process gain
Try to use the value of K _M. In this embodiment, the process gain K _M is automatically determined when the dead time L _M and the first-order delay coefficient β _M are determined, so that the number of repeated searches can be reduced.

As a method for the internal parameter adjusting unit to search for internal parameters of the process model, for example, the simplex method can be used. The simplex method is a search by taking several points with a geometrical arrangement on R ⁿ and comparing the values of the objective function at those points.
For example, it is described on pages 284 to 287 of the nonlinear programming method (written by Hiro Konno and Hiroshi Yamashita) published by Nikkan Giren. However, the normal simplex method has a problem that since the input and output data are basically discrete values, the objective function also becomes a space of discrete values and the search cannot be performed when the search point enters the subspace. There was also a constraint that parameters such as time constant and dead time must be positive numbers. FIG. 7 shows a search procedure using the simplex method and excluding the above constraints. In this figure, the process data collected by the data collecting unit 22 is a geometric point x on R ⁿ , and the function f () at this point is the objective function. The search is performed by comparing the values of the objective function f (). Also, of the points x, the point that gives the maximum value of f () is x ^h , the point that gives the second largest value is x ^s , the point that gives the maximum value is x ^l , and the point x
Let x ^m be the centroid of the figure with the vertex at. Furthermore, the reflection x ^r = (1 + α) · x ^m −α x ^h α> 0 expansion x ^e = γ · x ^r + (1-γ) · x ^m γ> 1 contraction x ^c = β · x ^h + (1 −β) ・ x ^m β ∈ (0,1) reduction Reduces all vertices in the direction of x ^l .

Is defined. In FIG. 7, after initialization, the first simplex, that is, (n + 1) affine-independent convex hulls on R ⁿ are generated. Then, the above-mentioned x ^h , x ^s , x ^l ,
Determine x ^m . Then, it is determined whether to converge. The judgment of convergence is Will be satisfied. However, Is. When the convergence condition is satisfied, if it is the first convergence, the first simplex is generated at that point, and x ^h , x ^s ,
Repeat from the decision of x ^l , x ^m . If it converges for the second time or more, it ends. When it does not converge, x ^r and f (x ^r ) are obtained, and it is determined whether or not f (x ^r ) ≦ f (x ^s ) is satisfied. If it is not satisfied, it is determined whether f (x ^r ) <f (x ^h ). If it is satisfied, x ^h is replaced by x ^r , and if not satisfied, x ^c and f (x ^c ) are obtained as they are. Then, f (x ^c ) <f (x ^h ) is determined. If this expression is satisfied, replace x ^h with x ^c , if not, replace x ⁱ with (x ⁱ + x ^l ) / 2 to determine x ^h , x ^s , x ^l , x ^m Return to the place. On the other hand, if f (x ^r ) ≦ f (x ^s ) is satisfied, it is determined whether f (x ^r ) ≧ f (x ^l ) is satisfied, and if not, x ^e ,
Obtain f (x ^e ), determine whether f (x ^e ) <f (x ^l ) is satisfied, and if so, replace x ^h with x ^e ,
When not satisfied and when f (x ^r ) ≧ f (x ^l ) is satisfied, x ^h is replaced with x ^r and the process returns to the determination of x ^h , x ^s , x ^l , and x ^m . In this embodiment, in comparison with the normal simplex method, a simplex is generated at that point at the time of the first convergence, and the simplex method is applied again, so that it is possible to prevent the subspace from entering and also to obtain a negative parameter. It is made to be applicable. In this embodiment, the length of the side component of the initial polygon is set longer than the sampling period, and the initial polygon is recreated at least once. However, only one of them may be performed.

FIG. 8 shows still another embodiment. In the embodiment shown in FIG. 1, the initial values of the proportional, integral and differential coefficients and the sampling period must be set prior to the control operation. If this setting is not appropriate, automatic adjustment of control constants will not work well,
The control characteristics deteriorate. In this embodiment, in order to avoid such a situation, the control constant and the sampling period are tuned in advance. Note that the same elements as those in FIG. 8th
In the figure, reference numeral 40 denotes a step input applying means, which applies a stepwise operation amount to the process 21. By inputting this operation amount, the process amount gradually increases as shown in FIG. Reference numeral 41 denotes a first monitoring means, which monitors the process amount that is the output of the process 21, and after the step-like operation amount is applied by the step input applying device 40, the process amount is a first predetermined value, for example, full scale. The time width t _{1 to} reach 1% of, that is, the first time width is measured. Reference numeral 42 denotes a sample period calculation means, to which the first time width measured by the first monitoring means 41 is input, and from this value, the data collection unit
Calculate the sampling period for 22. Reference numeral 43 denotes a second monitoring means, which monitors the process amount which is the output of the process 21, and after the step-like operation amount is applied by the step input applying device 40, the process amount is larger than the first predetermined value. The time width t ₂ until the second predetermined value, for example, 2% of full scale, that is, the second time width, is measured. Reference numeral 44 denotes a sample cycle correction means, which receives the second time width measured by the second monitoring means 43 and corrects the sample cycle calculated by the sample cycle calculation means 42. A switch 45 constitutes a sampler, which is first turned on for each cycle calculated by the sample cycle calculation means 42 and later for each cycle corrected by the sample cycle correction means 44, and the process amount data is collected by the data collection unit 22. Collected by. This collected data is input to the modeling unit 23, and its internal parameters are adjusted so that the output of the process model matches the collected process amount as described with reference to FIG. Is calculated.

Next, the operation of this embodiment will be described based on the flowchart of FIG. This operation is executed at regular intervals in the initial period of control. First, it is determined whether it is an initial start, and if it is an initial start, a stepwise operation amount is applied to the process 21. If it is not the first step, the process amount is monitored to determine which of the following three steps is applicable. If none of the steps apply, the process ends without doing anything.

When the process amount PV exceeds 1% of full scale.

When the process amount PV exceeds 2% of full scale.

When the process amount PV is settled to a constant value or becomes an integral multiple of the sample period obtained by the sample period correction means 44.

If it corresponds to a step, the elapsed time t ₁ after applying the stepwise operation amount to the process, that is, the first time width is measured, and an integral fraction of this time is set as the sample time t _s . For example, the t _s = t _1/4. If it corresponds to a step, the elapsed time t ₂ after applying the stepwise operation amount to the process, that is, the second time width is measured, and this time width t ₂ is twice the time t ₁ obtained in step 1 Judge whether it is large or not. twice larger than the t _1, as t _s = t _2/4, by interpolating the previously determined process quantity is corrected to a value obtained by sampling the process variable at the new sampling time t _s. When the step is applied, the process amount data collection is ended, the process amount and operation amount time series data {PV}, {MV} accumulated in the data collection unit 22 is arranged, and modeling is performed based on this time series data. The internal parameter of the process model 232 is adjusted by the unit 23 to obtain the control constant. This control constant is the initial value. After that, data is collected at the cycle of this sample time t _s , and the control constant is calculated by the procedure described in FIG.

In this embodiment, the stepwise operation amount is applied only once in the initial stage of control, but in the case of the localization system,
Although the process amount PV converges as shown in FIG. 11 (A), an offset remains as shown in FIG. 11 (b) in the case of the integral system. Therefore, after the process amount is settled, stepwise signals are added in the opposite direction to remove the offset, and then the sampling of the process data is continued. If the change in the process amount exceeds the allowable range, stepwise signals are applied in the opposite direction even before the settling to prevent the process amount from becoming excessive. Further, it is possible to determine whether or not the process is an integral system from the offset amount. In FIG. 11 (b), the time T for applying the stepwise operation amount is set to be a predetermined time or a change amount of the process amount is larger than ΔP ^max .

Further, in these embodiments, the control constant is obtained from the internal parameters of the modeling unit by the Ziegler-Nichols method, but the least squares method or the maximum likelihood method may be used.

<Effects of the Invention> As described above in detail with reference to the embodiments, in the present invention, the process data is accumulated in the data collecting unit, and the modeling unit is activated only when the process data fluctuates. The internal parameters of the model were adjusted, the estimated accuracy of the process model was calculated, and the control constants were obtained from these. Therefore, the identification signal is unnecessary, and there is an effect that unnecessary disturbance is not given to the process.

Further, since the estimation accuracy of the process model is calculated, the adjustment of the control constant can be reduced or stopped when it is determined that the model is not adapted, so that the runaway of the process control device can be prevented.

Further, since the process model is calculated while temporarily accumulating the process data in the data collecting unit, the time for collecting the data becomes unnecessary.

Further, the search time can be shortened by calculating the process gain from the output of the process amount and the process model.

Further, since the simplex method is used as the search method and the simplex is recreated once after the first convergence, the defect of entering the subspace can be eliminated.

Further, since a stepwise operation amount is given to the process at the beginning of control and the initial values of the sampling period and the control constant are obtained from the change in the process amount at that time, more appropriate control can be realized.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a process control device according to the present invention, FIG. 2 is a flow chart for explaining the operation thereof, and FIGS. 3 and 4 are explanations of the operation of the embodiment of FIG. FIG. 5 and FIG. 6 are diagrams showing the configuration and operation of another embodiment, FIG. 7 is a flow chart of a search method using the simplex method, and FIGS. FIG. 12 and FIG. 15 are diagrams showing an example of the configuration of a conventional process control device. 20 ... Control calculation part, 21 ... Process, 22 ... Data collection part, 23 ... Modeling part, 24 ... Estimated accuracy calculation part, 25 ...
... control constant calculator, 26 ... calculation command section, 40 ... step input applying means, 41 ... first monitoring means, 42 ... sample period calculating means, 43 ... second monitoring means, 44 ... Sample period correction means, 232,312 …… Process model, 233,315…
… Internal parameter adjuster,

Claims (5)

(57) [Claims] 1. A control calculation unit which inputs a process amount and a set value, calculates a manipulated variable of the process from these values and outputs the manipulated variable to the process, and data which collects the manipulated variable and the process amount for a predetermined time. A collection unit and a process model inside the collection unit, the operation amount collected by the data collection unit is input, and the output of the process model is adjusted so as to be most consistent with the process amount collected by the data collection unit. A modeling unit that adjusts parameters and an estimated accuracy calculation unit that calculates the certainty of this process model are provided, and the modeling unit is activated when the manipulated variable or the process amount fluctuates by a predetermined value or more. And a control constant of the control calculation unit from the outputs of the estimation accuracy calculation unit. 2. The gain constant, which is one of the internal parameters of the process model, is obtained from the output of the process model and the process amount when the gain constant of the process model is set to 1. The process control device according to item 1. 3. The simplex method is used to adjust the internal parameters of the process model, and the length of the side component of the initial polygon is set longer than the sampling period collected by the data collection unit. Item 2. The process control device according to item 1. 4. The simplex method is used to adjust the internal parameters of the process model, and after the internal parameters have converged, the initial polygon is recreated and searched at the coordinates at least once. The process control device according to claim 1. 5. A control calculation unit which inputs a process amount and a set value, calculates a manipulated variable of the process from these values and outputs the manipulated variable to the process, and data which collects the manipulated variable and the process amount for a predetermined time. A collecting unit; a step input applying unit that applies a step input to the process; a first monitoring unit that monitors a first time width until the process amount changes by a predetermined first level; A sampling period calculation means for sampling the process amount by the data collection unit at the sampling period with a value obtained by dividing the time width of the above, and until the process amount changes to a second level larger than the first level. Second monitoring means for monitoring the second time width of the, and the sampling period calculation means calculates from the second time width and the first time width. Sampling cycle correction means for correcting the sampling cycle, and a modeling section for calculating a control constant from the process data collected by the data collecting section by a process model, and at the beginning of control, the step input applying means steps the process. The process control device is characterized in that a control constant is obtained by applying a constant operation amount and activating a modeling unit from time series data of the process at that time.