JP6904282B2 - Plant control regulator - Google Patents

Description

The present invention relates to a plant control and adjustment device in plant equipment.

As a method for evaluating control performance in conventional plant control, especially in process system feedback control, a method called a minimum variance benchmark or "Harris index" as described in Non-Patent Documents 1 and 2 below is widely known. , It is possible to detect plant fluctuations using this method.

If the fluctuation of the plant can be detected, this fluctuation can be used for self-diagnosis of the adjustment timing in self-tuning, or the adjustment timing can be notified to the manager.

The Harris index does not require the installation of a new measurement sensor because the data used is only the measured value of the plant, and a specific frequency is injected into the system to measure noise attenuation, or, for example. Data that has been exchanged in the past without stopping the operation operation without inputting a specific command value such as step input, changing the program or modifying the device for that purpose, and included in it This is an excellent method that can be evaluated only by the noise generated.

Lane Desborough et al. "Performance Assessment Measures for Univariate Feedback Control" The Canadian Journal of Chemical Engineering Vol.70, 12, 1992 P.1186-1197 Hiroshi Maruta, 4 others, "Chapter 1 Method using minimum distributed control as a benchmark", [online], Japan Society for the Promotion of Science Process Systems Engineering 143rd Committee Workshop No. 25 Final Report, P.M. 224-234, [Search on December 5, 2017], Internet <http://ws25.pse143.org/Part3.pdf> Yoshihiro Onishi, 1 other person, "Design of smart PID control system based on low frequency component evaluation of control error", 2017 IEEJ Industrial Application Division Conference 2-OS1-2, [II-55]-[II- 58]

However, since the Harris index uses a statistical method called "variance", a large amount of time-series data is required, and since a large amount of time-series data is required, the change occurs when there is a change in the plant. It takes a considerable amount of time for the data to be reflected in the index.

In addition, the Harris index is highly dependent on wasted time, which makes it impossible to measure wasted time accurately, or simply makes it difficult to function in long plants, and a large amount of time series data in inexpensive equipment. The application itself becomes difficult because it cannot be maintained.

Therefore, as disclosed in Non-Patent Document 3, instead of using the entire frequency band of noise, only low-frequency components related to the response of the system are extracted through an LPF (low-pass filter) to reflect the reflection speed. And a method of compressing the amount of data using the forget-me-not coefficient have been devised.

However, the above-mentioned Non-Patent Document 3 is only a mathematical simulation, and in order to actually perform this, specific problems such as where to cut the cutoff frequency of the LPF and where to set the performance evaluation standard are solved. It has not been.

An object of the present invention is to provide a plant control adjustment device capable of appropriately adjusting a controller in response to plant fluctuations in view of the above technical problems.

The plant control adjustment device according to the first invention for solving the above problems is
It is provided with a PID controller that inputs an operation amount MV to the plant based on the set value SV, and a processing unit that processes the output y in which the disturbance D is added to the output of the plant. A plant control adjustment device connected to a programmable controller that performs feedback control on the plant based on the measured value PV in which the processing is performed.
A data storage unit that outputs an SV trigger signal when the set value SV is constant, and
An operation of creating an identification model and a normative model of the plant based on the manipulated variable MV, the measured value PV, and the setting data of the programmable controller, and adjusting the control parameters of the PID controller using the normative model. Parameter optimization section to be performed and
A parameter transmission unit that inputs control parameters adjusted by the parameter optimization unit to the PID controller, and a parameter transmission unit.
A frequency determination unit that determines the _{LPF cutoff frequency f cf} and order as frequency data based on the time constant and the setting data included in the identification model or the reference model.
It is provided with a control performance evaluation unit that outputs an adjustment trigger signal when it is determined that the control performance of the PID controller is deteriorated by using the measured value PV, the SV trigger signal, and the frequency data.
The parameter optimization unit
When the adjustment trigger signal is input, an operation for adjusting the control parameters of the PID controller is performed.
The control performance evaluation unit
When the SV trigger signal is input, an average value calculation unit that obtains the average value of the measured value PV from the measured value PV and subtracts the average value of the measured value PV from the measured value PV, and
A low-pass filter in which filter parameters are designed based on the frequency data and the values calculated by the average value calculation unit are filtered.
When the SV trigger signal is input, a variance calculation unit that calculates the _{output variance σ y} ² based on the measured value PV filtered by the low-pass filter, and
When the SV trigger signal is input, the disturbance D is estimated based on the measured value PV, and the minimum variance calculation unit for _{obtaining the minimum variance σ MV} ^{2 from the estimated disturbance D, and the minimum variance calculation unit.}
The Harris index calculation unit for obtaining the Harris index η based on the output variance σ _y ² and the minimum variance σ _MV ^2.
It is characterized by including a data evaluation unit that outputs the adjustment trigger signal when the value related to the Harris index η exceeds a preset threshold value.

The plant control adjustment device according to the second invention for solving the above problems is
In the plant control adjustment device according to the first invention.
The frequency determination unit
The first LPF cutoff frequency candidate calculation unit that calculates the first LPF cutoff _{frequency candidate f ca} based on the time constant, and
Based on the above setting data, the second LPF cutoff frequency candidate f _cb and the second LPF cutoff frequency candidate / lower limit frequency calculation unit for calculating the lower limit frequency f _{min, and}
Of the first LPF cutoff frequency candidate f _ca and the second LPF cutoff frequency candidate f _cb , whichever is higher than n ₁ times the lower limit frequency f _min , or both are n ₁ times the _{lower limit frequency f min.} When the frequency is higher than the frequency of, the lower frequency is provided as the LPF cutoff frequency f _{cf with} a frequency comparison determination unit to be input to the control performance evaluation unit, and 1 ≦ n ₁ ≦ 100. It is characterized by that.

The plant control adjustment device according to the third invention for solving the above problems is
In the plant control adjustment device according to the second invention.
The first LPF cutoff frequency candidate calculation unit is
Assuming that the time constant is T, the first LPF cutoff frequency candidate f _ca is calculated by the following equation (A).
The second LPF cutoff frequency candidate / lower limit frequency calculation unit is
From the set data, the moving average score m and the sampling period ΔT when the processing by the processing unit is the moving average processing are read, and 1/10 of the cutoff frequency of the moving average filter used for the moving average processing is set. As the second LPF cutoff frequency candidate f _{cb , the second LPF cutoff frequency candidate f cb} is calculated as shown in the following equation (B).

According to the plant control adjustment device according to the present invention, the controller can be appropriately adjusted in response to plant fluctuations.

It is a block diagram which shows the state which the plant control adjustment apparatus which concerns on Example 1 of this invention is provided in the feedback control system. It is a block diagram which shows the structure of the plant control adjustment apparatus which concerns on Example 1 of this invention. It is a block diagram which shows the structure of the control performance evaluation part in Example 1 of this invention. It is a block diagram which shows the structure of the frequency determination part in Example 1 of this invention. It is a graph which shows an example of a power spectral density (PSD) distribution. It is a graph which shows an example of the change of PSD distribution when a plant change occurs. It is a graph which shows the state which cut the high region component in the PSD distribution, and extracted only the PSD of a necessary frequency region. It is a block diagram which shows the structure of the plant control adjustment apparatus which concerns on Example 2 of this invention. It is a block diagram which shows the structure of the frequency determination part in Example 2 of this invention. It is a block diagram which shows the structure of the plant control adjustment apparatus which concerns on Example 3 of this invention. It is a block diagram which shows the structure of the control performance evaluation part in Example 3 of this invention. It is a block diagram which shows the structure of the frequency determination part in Example 3 of this invention. It is the graph corresponding to the said FIG. 5C in Example 3 of this invention, and shows the state of calculating only the line of the frequency component of each selected order. It is a graph which shows the state which distributed the weighting coefficient with respect to the frequency of each selected order in Example 3 of this invention. It is a block diagram which shows the structure of the plant control adjustment apparatus which concerns on Example 4 of this invention. It is a block diagram which shows the structure of the control performance evaluation part in Example 4 of this invention. It is a block diagram which shows the structure of the frequency determination part in Example 4 of this invention. It is a graph which shows the output dispersion in the frequency domain of Example 4 of this invention. It is a graph corresponding to the above FIGS. 5C and 11A in the fourth embodiment of the present invention, and the solid line portion shows the value of pv (n) when n is large. It is a graph which plotted the PV signal to which white noise was added, and the pv at that time obtained by the forgetting coefficient M = 0.99 to the PV signal up to 1000 times of calculation in Example 4 of this invention. It is a graph which showed the ratio of pv and the actual average value of PV in Example 4 of this invention.

Hereinafter, the plant control and adjustment device according to the present invention will be described with reference to the drawings in Examples.

[Example 1]
First, the configuration of the plant control and adjustment device according to this embodiment will be described. FIG. 1 is a block diagram showing a state in which the plant control adjustment device (plant control adjustment device 6) according to the present embodiment is provided in the feedback control system.

The feedback control system shown in FIG. 1 includes a programmable controller (PLC) 1, a distributed control device (DCS) 2, and a plant 3. Further, the programmable controller 1 includes a PID controller 4 and a processing unit 5.

The PID controller 4 inputs an input u to the plant 3 based on the output target r of the plant P. The value of the input u at this time is defined as the manipulated variable MV. The plant 3 outputs the output y when the input u is input. The set value SV, which is the value of the target r, is input from the distributed control device 2 which is a higher-level control device of the programmable controller 1, or is input by the operator.

The processing unit 5 adds the processing H to the output y obtained by adding the disturbance Dw to the output of the plant 3. This process H is a moving average process that is often inserted to see the data trend on the programmable controller 1 side. The value after the processing H is added to the output y is set as the measured value PV, the value obtained by adding the measured value PV to the target r (set value SV) is re-input to the PID controller 4, and feedback control is performed to the plant 3. Do.

Here, the disturbance Dw applied to the output of the plant 3 collectively represents an external element such as a change in room temperature and measurement noise, for example, in the case of a temperature controller, and the white noise w passes through the transmission characteristic D. Express it as if it came.

The plant control adjustment device 6 is provided in such a feedback control system. At that time, it may be realized on a personal computer (so-called loader PC) connected to the distributed control device 2 or the programmable controller 1. Further, some of the functions (configurations) of the plant control and adjustment device 6 having a light load may be performed on the programmable controller 1.

FIG. 2 is a block diagram showing the configuration of the plant control and adjustment device 6. As shown in FIG. 2, the plant control adjustment device 6 includes a data storage unit 11, a parameter optimization unit 12, a parameter transmission unit 13, a frequency determination unit (FD) 14, a control performance evaluation unit (ACP) 15, and an evaluation. A data storage unit 15a is provided.

The data storage unit 11 stores each data of the operation amount MV and the measured value PV (temporary storage (recording) by the volatile cyclic memory and long-term storage (storage) by the non-volatile memory or the like). Further, the data storage unit 11 inputs the stored measured value PV to the parameter optimization unit 12 and the control performance evaluation unit 15, and inputs the stored operation amount PV to the control performance evaluation unit 15. Further, the data storage unit 11 determines whether or not the value of the set value SV has changed, and controls the SV trigger signal when the change of the set value SV stops (when the set value SV is constant). Input to 15 (SV trigger in FIG. 2).

The parameter optimization unit 12 creates an identification model of the plant 3 based on each data of the operation amount MV and the measured value PV, and the setting data of the programmable controller 1, and controls the PID controller 4 using this identification model. The parameters are optimized, and the identification model and the control parameters are stored (the parameter optimization unit 12 creates not only the identification model but also the normative model described in the second embodiment, but omits it in this embodiment). Further, the parameter optimization unit 12 performs an operation for adjusting the control parameters of the PID controller 4 when an adjustment trigger signal described later is input.

The parameter transmission unit 13 inputs the control parameters optimized by the parameter optimization unit 12 to the PID controller 4.

The control performance evaluation unit 15 evaluates the control performance using the measured value PV and the SV trigger signal, and outputs the adjustment trigger signal. Further, the control performance evaluation unit 15 improves the accuracy of the control performance evaluation by using the frequency data input from the frequency determination unit 14 described later.

The control performance evaluation unit 15 having a relatively light load and the data storage unit 11 when the storage area is secured may be provided on the programmable controller 1 side.

FIG. 3 is a block diagram showing the configuration of the control performance evaluation unit 15. As shown in FIG. 3, the control performance evaluation unit 15 includes an average value calculation unit 21, an LPF (low-pass filter) 22, a dispersion calculation unit 23, a minimum dispersion calculation unit 24, a Harris index calculation unit 25, and a data evaluation unit 26. I have.

When the SV trigger signal is input from the data storage unit 11, the average value calculation unit 21 obtains the average value of the measured value PV from the measured value PV input from the data storage unit 11. Then, the average value of the measured value PV is subtracted from the measured value PV.

The LPF 22 has filter parameters designed based on frequency data, and filters the values calculated by the average value calculation unit 21. The filter parameters of the LPF 22 are fixed until the frequency data itself is updated due to re-identification or the like.

When the SV trigger signal is input from the data storage unit 11, the variance calculation unit 23 calculates the variance (output variance) σ _y ² based on the measured value PV filtered by the LPF 22.

When the SV trigger signal is input from the data storage unit 11, the minimum variance calculation unit 24 estimates the disturbance D from the measured value PV, obtains the minimum variance σ _MV ² from the estimated disturbance D, and uses it as a benchmark for evaluation.

The Harris index calculation unit 25 obtains _{the Harris index η based on the output variance σ y} ² _{input from the variance calculation unit 23 and the minimum variance σ MV} ² input from the minimum variance calculation unit 24, and evaluates the data as evaluation data. Output to unit 26 and evaluation data storage unit 15a. The evaluation data storage unit 15a stores this evaluation data.

The data evaluation unit 26 creates a benchmark based on the past evaluation data input from the evaluation data storage unit 15a, obtains the difference between the benchmark and the latest evaluation data input from the Harris index calculation unit 25, and obtains the difference. When the difference exceeds the threshold value, an adjustment trigger signal is generated.

Further, the frequency determination unit 14 shown in FIG. 2 refers to the identification model, which is the data obtained by the parameter optimization unit 12, and the setting data of the programmable controller 1, and obtains the cutoff frequency and the order data. It is determined and these frequency data are input to the control performance evaluation unit 15. The control performance evaluation unit 15 can improve the accuracy by using these frequency data.

FIG. 4 is a block diagram showing the configuration of the frequency determination unit 14. The frequency determination unit 14 includes a first LPF cutoff frequency candidate calculation unit 27, a second LPF cutoff frequency candidate / lower limit frequency calculation unit 28, and a frequency comparison determination unit 29.

The first LPF cutoff frequency candidate calculation unit 27 calculates the first LPF cutoff frequency candidate f _ca based on the identification model input from the parameter optimization unit 12.

The second LPF cutoff frequency candidate / lower limit frequency calculation unit 28 calculates the second LPF cutoff frequency candidate f _cb and the lower limit frequency f _min based on the setting data of the programmable controller 1.

The frequency comparison determination unit 29 sets _{the lower frequency of the LPF cutoff frequency candidates f ca and} f _cb as the LPF cutoff frequency f _cf, and inputs this LPF cutoff frequency f _cf to the control performance evaluation unit 15. To do. Higher than n ₁ times the frequency of the lower frequency f _min, or, if both higher than n ₁ times the frequency of the lower frequency f _min is the lower frequency therein, the LPF cutoff frequency f _cf To do. For n ₁ , 1 ≦ n ₁ ≦ 100.

In the following, the feedback control system provided with the plant control adjustment device 6 will be described in more detail.

In the case of the discrete system system configuration shown in FIG. 1, the output y is represented by the following equation (1).

When the variance σ _y ² of the output y is taken in the control performance evaluation unit 15, it is expressed in FIGS. 1 to 3 that the measured value PV is input to the control performance evaluation unit 15 instead of the output y. If it is 1, the measured value PV and the output y are equal), and if the set value SV is constant by the constant value control, the term on the right side r becomes 0. That is, _{it can be considered that the variance σ y} ² of the output y is entirely caused by the disturbance. Therefore, if there is no trend in the disturbance, the disturbance transfer function D can be estimated from the output y by the correlation analysis by the ARIMA model.

The Harris index calculation unit 25 uses the output variance σ _y ² and obtains the Harris index η from the following equation (2) _{using the variance σ MV} ² when the minimum variance control is performed as a benchmark.

The minimum variance σ _MV ² is the minimum value of the variance that can be achieved by suppressing the disturbance component as much as possible, and when σ _y ² = σ _MV ² is achieved, the Harris index η becomes the maximum 1. Further, since the minimum variance σ _MV ² is a theoretical value that does not depend on the control parameters of the PID controller 4, it is possible to evaluate the control system using this as a benchmark.

The minimum variance σ _MV ² estimates the disturbance D from the measured value PV in the minimum variance calculation unit 24 when the set value SV is constant, that is, when the SV trigger signal is input from the data storage unit 11, and the disturbance It is required based on D, and this is used as a benchmark for evaluation. The minimum variance σ _MV ² does not need to be calculated every time, and is updated by presetting a sufficiently long cycle for the system, such as every adjustment, every day, or every week.

Here, consider when the transmission characteristics of the plant 3 instead of the PID controller 4 fluctuate (for example, when the outside air temperature decreases and the cooling effect of the intake fan increases).

When the transmission characteristic of the plant 3 fluctuates, the PID controller 4 responds to the fluctuation, so that the average value of the output y is maintained so as to be the set value SV, but the transmission of the disturbance is changed. The variance fluctuates from.

When the plant 3 is an intake fan, the gain becomes excessive, the output y becomes oscillating, the dispersion becomes larger, and the average value is maintained, but the power consumption increases. On the contrary, if the gain is too small, the control performance deteriorates so that the average value cannot be maintained in the end.

That is, the change in the output variance σ _y ² , that is, the change in the Harris index η indicates the timing when the control parameter should be readjusted. Utilizing this, the control performance evaluation unit 15 outputs an adjustment trigger signal to the parameter optimization unit 12 when the Harris index η exceeds a predetermined threshold value. As a result, the plant control and adjustment device 6 executes self-tuning.

_{The variance σ y} ² of N data y _n (n = 1, 2, ..., N) is obtained by the following equation (3) in the variance calculation unit 23.

That is, if the average value calculation unit 21 obtains the average value and _{subtracts it from the data y n in} advance, the variance σ _y ² is obtained by multiplying the sum of squares by 1 / N as in the above equation (3). ..

FIG. 5A is a graph showing an example of the power spectral density (PSD) distribution, in which the horizontal axis represents frequency and the vertical axis represents PSD. The variance σ _y ² represents the area of the PSD in the frequency domain as shown in FIG. 5A.

Further, as described above, if the set value SV is constant, the variance σ _y ² of the output y is only the disturbance element, so this PSD is equivalent to the frequency analysis of the disturbance. Therefore, if there is only steady disturbance (white noise), the PSD area will be constant (no change over time) unless plant fluctuations occur.

FIG. 5B is a graph showing an example of changes in the PSD distribution when plant fluctuations occur, with the horizontal axis representing frequency and the vertical axis representing PSD. As shown in FIG. 5B, the portion of PSD shown in gray is increasing due to plant fluctuations (the gray portion in FIG. 5C below has the same meaning).

In order to utilize the change in the PSD area due to plant fluctuations _{for detecting changes in the variance σ y} ² , it is empirical that the changes in PSD cannot reach a level that can be detected unless the fluctuation range is considerably large. I know.

This is because, as shown in FIG. 5B, the ratio of plant fluctuations to the dispersion of high-frequency components, which is particularly difficult to see, is rather large, and the increase / decrease due to plant fluctuations is included in the increase / decrease due to changes in high-frequency components ( The gray increase portion in FIG. 5B) is likely to be buried.

In reality, the processing H (moving average) by the processing unit 5 removes the high frequency components from the system, but this moving average may be placed for the purpose of visual smoothness for trend data. , The filter effect required here is weak. Therefore, as shown by the broken line in FIG. 5C, it is necessary to cut the high frequency component larger and extract only the necessary frequency region.

Therefore, the identification model shown in FIG. 2 is the following equation (4a) for the first-order lag system, the following equation (4b) for the second-order lag system obtained by synthesizing the first-order lag system and the integral system, and the following equation for the generalized second-order lag system. When expressed by either the equation (4c) or the following equation (4d) of the third-order lag system obtained by synthesizing the second-order lag system and the integral system, the first LPF cutoff frequency candidate calculation unit 27 in FIG. The reciprocal of the time constant of the identification model or 1/10 of the natural angular velocity is used as the first LPF cutoff frequency candidate, and is calculated as the following equation (5a) or the following equation (5b).

Further, in parallel with the operation of obtaining the first LPF cutoff frequency candidate f _ca described above, in the second LPF cutoff frequency candidate / lower limit frequency calculation unit 28 of FIG. 4, the moving average of the process H is obtained from the setting data of the programmable controller 1. The score m and the sampling period ΔT are read, the cutoff frequency f _cm of the moving average filter is obtained by the following equation (6), and 1/10 of the moving average cutoff frequency f _cm is set as the second LPF cutoff frequency candidate f _cb. As a result, it is calculated as shown in the following equation (7).

However, the sampling frequency f _s is obtained from the setting data of the programmable controller 1. If there is no processing H (moving average) for the second LPF cutoff frequency candidate f _{cb , 1/2 times the sampling frequency f s} is used and f _cb = (1/2) f _s .

Further, in the frequency comparison determination unit 29, _{the lower frequency of the LPF cutoff frequency candidates f ca and} f _cb is set as the LPF cutoff frequency f _cf, and this LPF cutoff frequency f _cf is used as the control performance evaluation unit 15. Enter in.

_{Based on the LPF cutoff frequency f cf} , the control performance evaluation unit 15 _{forms an LPF 22 having a primary LPF 1} value as shown in the following equation (8), whereby only the low frequency component of the measured value PV is extracted. Extract. It is also conceivable to use a higher-order filter or FIR (finite impulse response) filter so that a filter having desired characteristics can be obtained.

However, for the low frequency component, the lower limit frequency f _min is determined by the data length (sampling period ΔT × number of data points N _{s) as shown in the following equation (9).} The LPF cutoff frequency f _cf is at least 10 times or more the lower limit frequency f _min.

Therefore, the second LPF cutoff frequency candidate / lower limit frequency calculation unit 28 calculates the lower limit frequency f _{min together with the second LPF cutoff frequency candidate f cb} based on the setting data of the programmable controller 1, and the LPF cutoff frequency f _cf. Consider when making a decision.

According to this embodiment, in the extraction of low-frequency components performed in the control performance evaluation, the controller can be adjusted at an appropriate timing for plant fluctuations by determining the cutoff frequency of the LPF using the identification model. it can.

[Example 2]
Plant control adjustment apparatus according to the present embodiment, in the selection of the LPF cutoff frequency f _cf described in Example 1, using a reference model R instead of identifying the model, the reciprocal of the constant T _R when the reference model R Chapter It is used for one cutoff frequency candidate f _ca. Hereinafter, the parts different from those of the first embodiment will be mainly described.

FIG. 6 is a block diagram showing a configuration of a plant control adjustment device (plant control adjustment device 7) according to the present embodiment. Further, FIG. 7 is a block diagram showing the configuration of the frequency determination unit 34.

As shown in FIG. 6, the plant control adjustment device 7 includes a data storage unit 11, a parameter optimization unit 32, a parameter transmission unit 13, a frequency determination unit (FD) 34, a control performance evaluation unit (ACP) 15, and an evaluation. A data storage unit 15a is provided.

The parameter optimization unit 32 further generates a normative model R based on the identification model obtained as described in the first embodiment.

The frequency determination unit 34 includes a first LPF cutoff frequency candidate calculation unit 37, a second LPF cutoff frequency candidate / lower limit frequency calculation unit 28, and a frequency comparison determination unit 29.

The first LPF cutoff frequency candidate calculation unit 37 calculates the first LPF cutoff frequency candidate f _ca based on the norm model R generated by the parameter optimization unit 32.

The second LPF cutoff frequency candidate / lower limit frequency calculation unit 28 calculates the second LPF cutoff frequency candidate f _cb and the lower limit frequency f _min based on the setting data of the programmable controller 1 as in the first embodiment.

_{The frequency comparison determination unit 29 includes a first LPF cutoff frequency candidate f ca} calculated by the first LPF cutoff frequency candidate calculation unit 37 and a second LPF cutoff frequency candidate calculated by the second LPF cutoff frequency candidate / lower limit frequency calculation unit 28. If f _cb is higher than 10 times the lower limit frequency f _min , or both are higher than 10 times the lower limit frequency f _min , the lower frequency is the LPF cutoff frequency. Let f _cf.

The parameter optimization unit 32 creates a norm model R as an approximation of the Nth-order plant model of the plant 3 + the PID controller 4 in the form of the Nth power of the first-order lag shown in the following equation (10), and creates this norm model R. It is used to optimize the control parameters of the PID controller 4. The order N of the normative model R in the following equation (10) is determined by the order of the plant 3 and the order of the PID controller 4.

The first LPF cutoff frequency candidate calculation unit 37 calculates the first LPF cutoff frequency candidate f _ca as shown in the following equation (5c).

In the first embodiment, the order of the plant 3 is a quadratic system up to a quadratic system + an integral system, but even if the order of the plant 3 is higher, it is expressed by one time constant by using the above equation (10). Therefore, the first LPF cutoff frequency candidate f _ca can be uniquely determined.

However, in the normative model R, a programmable controller is further provided above the user or the host control system (distributed controller 2 or the target programmable controller 1) depending on whether the system (plant 3 + PID controller 4) is high speed or low speed. If so, the response of the normative model R does not necessarily represent the characteristics of the system because it may be determined based on the setting of the time constant determined by the programmable controller).

However, although it is possible to set the norm model R late for a fast system, it is common to set the norm model R so fast that it exceeds the limit of the system because the stability of the system is lost. There is nothing to do with adjustment.

Therefore, although the original system frequency, that is, the fluctuation region of the frequency component due to the fluctuation of the plant 3 may be underestimated, it cannot be overestimated. Therefore, the LPF cutoff frequency f _cf created based on the norm model R is It functions as a removal of unnecessary high frequency components in the index calculation.

In this way, in this embodiment, the identification model in Example 1 can be substituted for the normative model.

In Examples 1 and 2, the first LPF cutoff frequency candidate f _ca is determined by the time constant included in the identification model or the normative model. Then, Example 2 is one in which a portion of the constant T _P when the (5a) equation described in Example 1 was replaced with constant T _R when the equation (10).

The explanation is supplemented on this point. First, the time constant is a coefficient included in the equations of the first-order lag and the second-order lag, and the time constant can be expressed only under limited circumstances in the equation of the third order or higher.

The identification model expresses an equation based on actual data. _{Then, the time constant T P} can be expressed from the above equations (5a) or (5b) only when the equation is described in any of the above equations (4a) to (4d).

Reference model contrast, using a constant T _R when determined by the user or the higher-level control system is intended to create an expression of first-order lag or second-order lag time constant T _R is the previously determined There is.

Then, regardless of whether the identification model of Example 1 is used or the normative model of Example 2 is used, a time constant is set for the parameter optimization units 12 and 32 to the frequency determination units 14 and 34. The model is input without extraction, and the first LPF cutoff frequency candidate calculation units 27 and 37 in the frequency determination units 14 and 34 determine the first LPF cutoff frequency candidate f _{ca using the time constant included in the model.} .. If the time constants of the identification model and the normative model are collectively T, the first LPF cutoff frequency candidate f _ca is expressed by the following equation (5d).

[Example 3]
Examples 1 and 2 include a control performance evaluation unit and a frequency determination unit, and the frequency components are calculated in "planes" using dispersion. However, the plant control adjustment device according to this embodiment is a control performance evaluation unit. A 45 and a frequency determination unit 44 are provided, and this is calculated by a "line" of a specific frequency component. Hereinafter, the parts different from those of Examples 1 and 2 will be mainly described.

FIG. 8 is a block diagram showing the configuration of the plant control and adjustment device 8. As shown in FIG. 8, the plant control adjustment device 8 includes a data storage unit 11, a parameter optimization unit 12, a parameter transmission unit 13, a frequency determination unit (FD) 44, a control performance evaluation unit (ACP) 45, and an evaluation. A data storage unit 45a is provided.

The control performance evaluation unit 45 outputs an adjustment trigger signal when it is determined that the control performance of the PID controller 4 is deteriorated by using the measured value PV, the SV trigger signal, and the base frequency f _base. More specifically, a sine wave and a chord wave having a base frequency f _base and an frequency n times the base frequency are generated, and the product-sum calculation of the generated sine wave and the chord wave and the measured value PV is performed to obtain only a specific frequency component. Extract from the measured value PV. Then, the performance index is calculated from the extracted frequency component, evaluated, and the adjustment trigger signal is output. Incidentally, a wave of n times a frequency referred to as the n-order harmonic of the base frequency f _base, base frequency f _base and n th harmonic is determined by the frequency determining unit 44 which will be described later.

FIG. 9 is a block diagram showing the configuration of the control performance evaluation unit 45 in this embodiment. As shown in FIG. 9, the control performance evaluation unit 45 includes a sine wave / cosine wave generator 51, a steady power spectrum calculation unit 54, an index calculation unit 55, and a data evaluation unit 56.

The sine wave / cosine wave generator 51 generates a sine wave and a cosine wave having a base frequency f _{base and an n times frequency thereof based on the base frequency f base and the order n.} After that, the product-sum calculation with the measured value PV is performed. As a result, only the base frequency f _base and its nth harmonic component can be extracted.

When the SV trigger signal is input, the stationary power spectrum calculation unit 54 measures the stationary power spectrum of the measured value PV based on the measured value PV, and uses this as a benchmark to obtain the minimum variance σ _{MV in Example 1.} Use instead of ^2. Let the steady power spectrum of this measured value PV be σ _yst ² . Similar to the minimum variance σ _MV ² , this is also updated at a sufficiently long cycle with respect to the system time constant after confirming the SV trigger signal.

The index calculation unit 55 _regards the base frequency f base and its nth-order tuning component multiplied by weight distribution as the power spectrum σ _y ² of the measured value PV, and regards the power spectrum σ _y ² and the steady state. _{Based on the steady power spectrum σ yst} ² input from the power spectrum calculation unit 54, the index ξ is obtained from the following equation (2c) and output as evaluation data. The output evaluation data is stored in a storage unit (not shown).

The data evaluation unit 56 compares the latest evaluation data with the past evaluation data, and generates an adjustment trigger signal when the preset threshold value is exceeded.

_{On the other hand, the frequency determination unit 44 calculates the cutoff frequency f cm} and the lower limit frequency f _min of the moving average processing from the setting data of the programmable controller 1 as in the first and second embodiments, and based on the cutoff frequency f min, the base frequency f. _{The base} and the order n are determined. At the same time, the weight is determined. Then, these are input to the control performance evaluation unit 45.

FIG. 10 is a block diagram showing the configuration of the frequency determination unit 44 in this embodiment. As shown in FIG. 10, the frequency determination unit 44 includes a maximum frequency / lower limit frequency calculation unit 58, a degree determination unit 59, and a weight distribution calculation unit 60.

Maximum frequency and lower frequency calculating unit 58, the moving average of the cut-off frequency f _cm, and reads the sampling frequency f _s, the frequency with rounded down from the moving average of the cut-off frequency f _cm (eg f _cm = 4.47Hz If so, 4 Hz), or the frequency obtained by rounding off the fraction from the frequency (1/2) f _{s which is 1/2 of the sampling frequency f s} is calculated as _{the maximum frequency f max.} It should be noted that the process of truncating the fraction is for improving the convenience of calculation, and is not an essential process in this embodiment.

Further, the maximum frequency / lower limit frequency calculation unit 58 lowers the frequency by 0.1 times (10th order) with respect to _{the maximum frequency f max, such as 0.1 times, 0.01 times, and so on.} Select a frequency that does not fall below the lower limit frequency f _min. At this time, the smallest frequency selected is set as the base frequency f _base (that is, the base frequency f _base is determined between the maximum frequency f _max and the lower limit frequency f _min). The lower limit frequency f _min is obtained by the equation (9) in the first embodiment based on the setting data.

The order determination unit 59 determines _{n times the base} frequency f base, that is, the order n of the nth harmonic. Other frequencies including the maximum frequency f _max (frequency other than the _{base frequency f base ) are n times the base} frequency f base, that is, nth harmonic harmonics.

However, depending on the characteristics of the sine wave / cosine wave generator 51 provided in the control performance evaluation unit 45, it is not possible to generate an accurate and sharp frequency, and it is not possible to simply round down or perform 0.1 times. There is.

_{Therefore, the base frequency f base} and the order n determined by the maximum frequency / lower limit frequency calculation unit 58 and the order determination unit 59 are rounded and approximated to the closest frequencies that can be generated by the sine wave / cosine wave generator 51. You may.

Incidentally, when the maximum frequency f _{max is not separated from the base} frequency f base by 100 times or more, the order determination unit 59 changes the order width to be lowered with respect to _{the maximum frequency f max from the 10th order.} As a result, three or more frequencies that do not fall below the _{lower limit frequency f min are always selected.}

The weight distribution calculation unit 60 distributes weight coefficients to frequencies of each order selected by the order determination unit 59. _{By doing so, the lower limit frequency f min} input from the maximum frequency / lower limit frequency calculation unit 58 of the extracted frequencies is the same as using a window function such as a Gaussian window in the spectrum analysis by the general Fourier transform. , and is half the sampling frequency (1/2) to lower the weight as the frequency close to f _s, it is possible to increase the dynamic range.

The base frequency f _base , the order n, and the weight distribution determined in this way are input to the control performance evaluation unit 45 from the frequency determination unit 44 as frequency data.

The control performance evaluation unit 45 generates a sine wave and a cosine wave based on the frequency data determined by the frequency determination unit 44, and performs a product-sum calculation with the measured value PV to obtain the base frequency f _base and its nth harmonic. Only the ingredients can be taken out.

The performance is evaluated by regarding the extracted component multiplied by the weight distribution determined by the frequency determination unit 44 as the power spectrum σ _y ^{2 of the measured value PV.}

However, since the power spectrum σ _y ² is extremely lower than that of Examples 1 and 2, when the minimum variance σ _MV ² is used, the power spectrum σ _y ² _{becomes smaller than the minimum variance σ MV} ² , and the above (2) There is a possibility that η represented by the formula will exceed 1. Therefore, the steady-state power spectrum calculation unit 54 obtains the steady-state power spectrum σ _yst ² , and uses this instead of the minimum variance σ _MV ^2.

FIG. 11A is a graph corresponding to FIG. 5C, and shows a state in which only the lines of the frequency components of each selected order are calculated. Further, FIG. 11B is a graph showing a state in which weighting coefficients are distributed to frequencies of each selected order. For reference, FIGS. 11A and 11B show the values of Fmin, Fbase, Fmax, Fcm, and (1/2) Fs (particularly, an example in the case of Fcm <(1/2) Fs).

The sum of the solid lines in FIG. 11B is the power spectrum σ _y ² in this embodiment. Incidentally, the total area including the broken line portion in FIGS. 11A and 11B is the variance σ _y ² described in Examples 1 and 2. That is, in Examples 1 and 2, the frequency component was calculated in terms of the surface using dispersion, but in this example, this is calculated by the line of the frequency component of each selected order. As described above, in this embodiment, the response is fast because the dispersion is not used.

Therefore, according to this embodiment, a high-speed evaluation response can be realized by extracting only a specific frequency component without using model data and variance.

[Example 4]
In the plant control adjustment device according to this embodiment, the control performance evaluation unit passes the measured value PV through the BPF (bandpass filter) 72 instead of the LPF22 used in the first and second embodiments, and compresses the output data of the BPF 72. , The index calculation is performed as the evaluation data, and the amount of data can be suppressed as compared with Examples 1 to 3. Hereinafter, the parts different from those of Examples 1 to 3 will be mainly described.

FIG. 12 is a block diagram showing the configuration of the plant control and adjustment device 8. As shown in FIG. 12, the plant control adjustment device 8 includes a data storage unit 11, a parameter optimization unit 12, a parameter transmission unit 13, a frequency determination unit (FD) 64, a control performance evaluation unit (ACP) 65, and an evaluation. A data storage unit 65a is provided.

FIG. 13 is a block diagram showing the configuration of the control performance evaluation unit 65 in this embodiment. The control performance evaluation unit 65 outputs an adjustment trigger signal when it is determined that the control performance of the PID controller is deteriorated by using the measured value PV, the SV trigger signal, the cutoff frequency, and the forgetting coefficient M. As shown in FIG. 13, the control performance evaluation unit 65 includes a BPF 72, a steady power spectrum / compression calculation unit 74, an index calculation unit 75, and a data evaluation unit 76.

The BPF 72 is configured based on the cutoff frequency designed by the frequency determination unit 64, and filters the measured value PV.

When the SV trigger signal is input, the steady-state power spectrum / compression calculation unit 74 measures the steady-state power spectrum σ _yst ² , which is the power spectrum at the steady state, based on the measured value PV, and uses the oblivion coefficient M to PV. _{The data is compressed, and σ y} ² corresponding to the output variance (or the deemed amount thereof) in Examples 1 to 3 is calculated.

The index calculation unit 75 calculates the control performance index ζ using the following equation (2d) based on _{the steady power spectrum σ yst} ² and the deemed quantity σ _y ^2.

The data evaluation unit 76 compares the latest evaluation data with the past evaluation data, and generates an adjustment trigger signal when the preset threshold value is exceeded. This point is the same as the description of the data evaluation unit 26 in the first embodiment. That is, when the value related to the evaluation data (control performance index ζ) exceeds the preset threshold value, the adjustment trigger signal is output.

FIG. 14 is a block diagram showing the configuration of the frequency determination unit 64 in this embodiment. The frequency determination unit 64 determines the cutoff frequency for the measured value PV as frequency data based on the setting data of the programmable controller 1, and determines the cutoff frequency for the measured value PV based on the internal data (data type) of the programmable controller 1. It determines the oblivion coefficient M. As shown in FIG. 14, the frequency determination unit 64 includes an upper cutoff frequency / lower cutoff frequency, a BPF design unit 78, and a forgetting coefficient calculation unit 79.

The upper cutoff frequency / lower cutoff frequency calculation and the BPF design unit 78 include the time constant included in the identification model or the normative model (only the identification model is described in FIGS. 12 and 14), and the setting data of the programmable controller 1. Based on, the cutoff frequency for forming the BPF 72 is calculated. The cutoff frequency consists of an upper cutoff frequency f _c1 and a lower cutoff frequency f _c2 .

The forgetting coefficient calculation unit 79 calculates an effective forgetting coefficient M for compressing the data based on the internal data (data type) of the programmable controller 1 based on the restrictions on the hardware.

In Examples 1-3, both using the above equation (9) when determining the lower limit frequency f _min, but the lower limit frequency f _min is a value determined by hardware constraints of the data points N _s.

FIG. 15 is a graph showing the _{output variance σ y} ^{2 in the frequency domain.} On the horizontal axis of FIG. 15, the lower limit frequency f _{min , (1/2) f s} which is 1/2 of the sampling frequency, or the cutoff frequency f _{cm of the} moving average, and the LPF cutoff frequency f _c are described. The LPF cutoff frequency f _cf indicates one of the nth harmonic frequencies in the third embodiment.

_{Considering that f min} is the lower limit and (1/2) f _s or f _cm is the upper limit in FIG. 15, the LPF cutoff frequency f _cf (or nth harmonic frequency) is set between the upper and lower limits. It will be. Moreover, the output variance σ _y ² itself is equivalent to the calculation of the frequency band between the upper and lower limits.

The upper limit is the right end, because it is determined by the sampling frequency f _s, largely depends on the processing speed system equipment performance, readily but can not be extended to higher frequency bands, the lower limit is the measured value PV Since it is determined by the number of data points N _s , if the data storage capacity can be increased, it can be further extended.

However, since the storage capacity is also limited by hardware, we would like to obtain the same effect by compressing the stored data. Therefore, the steady-state power spectrum / compression calculation unit 74 performs the processing of the following equation (11) on the measured value PV (PV after passing through the BPF 72).

The forgetting coefficient M is a number close to 1 but smaller than 1 (for example, 0.99, etc.). The value obtained by multiplying the previous calculation result pv (n-1) by the forgetting coefficient M and the current measured value PV are added to obtain pv (n).

Here, considering the geometric progression S of the first term a and the common ratio r, it converges to a / (1-r) if r <1 from the formula of the sum of the geometric progressions in the following equation (12). Therefore, the first term a can be obtained by multiplying S by (1-r) as shown in the following equation (13).

Considering the same as a constant a = PV _c of this geometric progression S, it converges as shown in the following equation (14). Similar to the following equation (13), the following equation (14) becomes the following equation (15) by multiplying both sides by (1-M), so that the converged measured value PV _c can be taken out.

In reality, PV is not constant, but in this way, while referring to past PV data, it is possible to integrate the older PV data in a form that has less influence.

FIG. 17 is a graph in which the PV signal to which white noise is added and the pv at that time obtained by the forgetting coefficient M = 0.99 on the PV signal are plotted up to 1000 times of calculation. Further, FIG. 18 is a graph showing the ratio of pv to the actual average value of PV.

As the actual data, only the current value and the previous calculation result are used, and the error is about 3% at the 400th point (time) calculation and about 1% at the 500th point (time) calculation.

In this embodiment, the stationary power spectrum / compression calculation unit 74 does not perform the calculation for obtaining the variance and power spectrum as in Examples 1 to 3, but by performing the calculation using the above equation (11), If n is large, the value of pv (n) becomes the _{value σ y} ^{2 corresponding to the dispersion and power spectra in Examples 1 to 3.} FIG. 16 is a graph corresponding to FIGS. 5C and 11A. The solid line portion in FIG. 16 shows the value of this pv.

In this way, in this embodiment, the point that it takes time to converge the error does not change, but it is not necessary to hold a large amount of data.

Further, as the oblivion coefficient M is closer to 1, the term of the previous calculation result pv (n-1) becomes 1 / (1-M) times larger than that of the above equation (11), so that the internal data (data type) of the programmable controller 1 is increased. The oblivion coefficient M is determined according to the number of bits of), the upper and lower limits of the measured value PV, and the resolution. The resolution is the smallest unit of internal data, and the upper and lower limits are limits imposed by a measuring instrument, a converter, or the like.

Below (16), k _M of (17) represents the ratio of the upper limit that can retain data, Bit is the number of bits, PV _peak represents the operation area.

For example, when the internal data of the programmable controller 1 is 16 bits, the operating region is ± 2000 if the resolution of the measured value PV is 0.1% and the upper and lower limits are 200% with respect to ^{2 16 = 65536.} Is 65536/4000 = 16.384, which is 16 times when rounded down. As a result, since the forgetting coefficient M is a value smaller than 15/16 = 0.9375 according to the above equation (17), it becomes 0.93 by rounding down the second decimal place according to the resolution of 0.1%.

_{However, when the output variance σ y} ² is used as in the methods of Examples 1 and 2, the average value calculation is required, and a certain amount of data is required (in the average value calculation unit 21 of FIG. 3, the average value is calculated. Keeps track of past PV data to find out).

Therefore, in this embodiment, BPF72 is used instead of LPF22 in order to suppress the amount of data to be held. Since the BPF 72 passes only the frequency component of a specific band, the DC component is removed and it is not necessary to take a difference from the average value.

The upper cutoff frequency / lower cutoff frequency calculation and the BPF design unit 78 set the upper cutoff frequency f _{c1 of the BPF 72 to the same cutoff frequency f cf} as the LPF 22 of Examples 1 and 2 (time constant included in the identification model or the normative model). and, it said determined based on the setting data, using a cut-off frequency f _cf) as frequency data, the lower limit cutoff frequency f _c2 is the forgetting factor M, and, among the setting data (setting data, the measurement value PV Based on the sampling period ΔT, the resolution reso, and the upper limit or the lower limit frequency), the time when the internal data becomes less than 1 is defined as the lower limit cutoff cycle and is obtained as the reciprocal thereof.

The final nth term on the right side, that is, the oldest data component pv _n when the limit is not taken in the above equation (14) is expressed by the following equation (18). Here, when pv (1) is at the upper limit value limp, how many data points n will cause the oldest data component pv _{n in} Eq. (18) to be below the resolution, that is, 0, by taking the logarithm as follows ( 19) Expressed in equation.

For example, according to the above example, when the internal data is ΔT: 1s, reso: 0.1%, limup: 200%, M: 0.93, when n = 105, the internal data is 0. It becomes 98, which is less than 1, that is, it is treated as zero because it is smaller than the minimum unit. Since the oldest data component pv _n disappears at n × ΔT = 105s, this is the lower cutoff cycle. As shown in the above equation (20), the reciprocal is taken and rounded up to 0.01 Hz. The cutoff frequency is f _c2 (however, rounding up is not an essential configuration in this embodiment).

Using the above, the upper cutoff frequency / lower cutoff frequency calculation and the BPF design unit 78 design the value of the BPF 72. The frequency determination unit 64 sends the frequency data and the forgetting coefficient M to the control performance evaluation unit 65.

The control performance evaluation unit 65 configures the BPF 72 based on the frequency data, filters the measured value PV, and compresses the filter output using the forgetting coefficient M. The compressed data has already been stripped of DC and averaging, so it is converted to the _{output variance σ y} ^{2 by square.}

However, since this output variance σ _y ² is small as a variance value, the minimum variance σ _MV ² cannot be used as an index. Therefore, the stationary power spectrum σ _yst ² is stored in the same manner as in Example 3. In the index calculation, the stationary power spectrum σ _yst ² is used as a benchmark.

The steady-state power spectrum σ _yst ² is updated at a sufficiently long cycle with respect to the system time constant while confirming the SV trigger signal, as in the third embodiment.

In this embodiment, the upper cutoff frequency / lower cutoff frequency calculation and the BPF design unit 78 will use the _{same cutoff frequency f cf} as the LPF 22 of Examples 1 and 2 as _{the upper cutoff frequency f c1 of the BPF 72.} That is, the upper cutoff frequency / lower cutoff frequency calculation and the BPF design unit 78 calculate the first cutoff frequency candidate f _ca based on the identification model or the normative model, and based on the setting data of the programmable controller 1. The second cutoff frequency candidate f _cb and the lower limit frequency f _min are calculated, and of the first cutoff frequency candidate f _ca and the second cutoff frequency candidate f _cb , whichever is higher than n _{3 times the lower cutoff frequency f min.} Or, if both are _{higher than n 3 times the lower limit frequency f min} , the lower frequency is obtained as the cutoff frequency f _cf. At this time, 1 ≦ n ₃ ≦ 100.

Further, assuming that the time constant is T, the upper cutoff frequency / lower cutoff frequency calculation and the BPF design unit 78 calculate the first cutoff frequency candidate f _ca by the following equation (21), and the processing unit from the setting data. The moving average score m and the sampling period ΔT when the processing according to 5 is the moving average processing are read, and 1/10 of the cutoff frequency of the moving average filter used for the moving average processing is set as the second cutoff frequency candidate f _{cb. , The second cutoff frequency candidate f cb} is calculated as shown in the following equation (22).

According to this embodiment, it is possible to suppress the amount of data stored due to the low frequency component in Examples 1 and 2.

The present invention is suitable as a plant control and adjustment device.

1 Programmable controller (PLC)
2 Distributed Control System (DCS)
3 Plant 4 PID controller 5 Processing unit 6 (corresponding to Example 1 of the present invention) Plant control coordinator 7 (corresponding to Example 2 of the present invention) Plant control coordinator 8 (corresponding to Example 3 of the present invention) Plant Control regulator 9 (corresponding to the fourth embodiment of the present invention) Plant control regulator 11 Data storage unit 12 (in the first embodiment) Parameter optimization unit 13 Parameter transmission unit 14 (in the first embodiment) Frequency determination unit (FD)
15 Control Performance Evaluation Unit (ACP) (in Example 1)
15a Evaluation data storage unit 21 Average value calculation unit 22 LPF (low-pass filter)
23 Variance calculation unit 24 Minimum dispersion calculation unit 25 Harris index calculation unit 26 (in Example 1) Data evaluation unit 27 1st LPF cutoff frequency candidate calculation unit 28 2nd LPF cutoff frequency candidate / lower limit frequency calculation unit 29 Frequency comparison determination unit 32 Parameter optimization unit (in Example 2) 34 Frequency determination unit (FD) (in Example 2)
37 First LPF cutoff frequency candidate calculation unit 44 Frequency determination unit (FD) (in Example 3)
45 Control Performance Evaluation Unit (ACP) (in Example 3)
45a Evaluation data storage unit 51 Sine wave / cosine wave generator 54 Steady power spectrum calculation unit 55 (in Example 3) Index calculation unit 56 (in Example 3) Data evaluation unit 58 Maximum frequency / lower limit frequency calculation unit 59 Order determination Unit 60 Weight distribution calculation unit 64 Frequency determination unit (FD) (in the fourth embodiment)
65 Control Performance Evaluation Unit (ACP) (in Example 4)
65a Evaluation data storage unit 72 BPF (Bandpass filter)
74 Steady power spectrum / compression calculation unit 75 (in Example 4) Index calculation unit 76 (in Example 4) Data evaluation unit 78 Upper cutoff frequency / lower cutoff frequency and BPF design unit 79 Oblivion coefficient calculation unit

Claims (3)

It is provided with a PID controller that inputs an operation amount MV to the plant based on the set value SV, and a processing unit that processes the output y in which the disturbance D is added to the output of the plant. A plant control adjustment device connected to a programmable controller that performs feedback control on the plant based on the measured value PV in which the processing is performed.
A data storage unit that outputs an SV trigger signal when the set value SV is constant, and
An operation of creating an identification model and a normative model of the plant based on the manipulated variable MV, the measured value PV, and the setting data of the programmable controller, and adjusting the control parameters of the PID controller using the normative model. Parameter optimization section to be performed and
A parameter transmission unit that inputs control parameters adjusted by the parameter optimization unit to the PID controller, and a parameter transmission unit.
A frequency determination unit that determines the _{LPF cutoff frequency f cf} and order as frequency data based on the time constant and the setting data included in the identification model or the reference model.
It is provided with a control performance evaluation unit that outputs an adjustment trigger signal when it is determined that the control performance of the PID controller is deteriorated by using the measured value PV, the SV trigger signal, and the frequency data.
The parameter optimization unit
When the adjustment trigger signal is input, an operation for adjusting the control parameters of the PID controller is performed.
The control performance evaluation unit
When the SV trigger signal is input, an average value calculation unit that obtains the average value of the measured value PV from the measured value PV and subtracts the average value of the measured value PV from the measured value PV, and
A low-pass filter in which filter parameters are designed based on the frequency data and the values calculated by the average value calculation unit are filtered.
When the SV trigger signal is input, a variance calculation unit that calculates the _{output variance σ y} ² based on the measured value PV filtered by the low-pass filter, and
When the SV trigger signal is input, the disturbance D is estimated based on the measured value PV, and the minimum variance calculation unit for _{obtaining the minimum variance σ MV} ^{2 from the estimated disturbance D, and the minimum variance calculation unit.}
The Harris index calculation unit for obtaining the Harris index η based on the output variance σ _y ² and the minimum variance σ _MV ^2.
A plant control adjustment device including a data evaluation unit that outputs the adjustment trigger signal when a value related to the Harris index η exceeds a preset threshold value. The frequency determination unit
The first LPF cutoff frequency candidate calculation unit that calculates the first LPF cutoff _{frequency candidate f ca} based on the time constant, and
Based on the above setting data, the second LPF cutoff frequency candidate f _cb and the second LPF cutoff frequency candidate / lower limit frequency calculation unit for calculating the lower limit frequency f _{min, and}
Of the first LPF cutoff frequency candidate f _ca and the second LPF cutoff frequency candidate f _cb , whichever is higher than n ₁ times the lower limit frequency f _min , or both are n ₁ times the _{lower limit frequency f min.} When the frequency is higher than the frequency of, the lower frequency is provided as the LPF cutoff frequency f _{cf with} a frequency comparison determination unit to be input to the control performance evaluation unit, and 1 ≦ n ₁ ≦ 100. The plant control adjustment device according to claim 1, wherein the plant control and adjustment device is characterized. The first LPF cutoff frequency candidate calculation unit is
Assuming that the time constant is T, the first LPF cutoff frequency candidate f _ca is calculated by the following equation (A).
The second LPF cutoff frequency candidate / lower limit frequency calculation unit is
From the set data, the moving average score m and the sampling period ΔT when the processing by the processing unit is the moving average processing are read, and 1/10 of the cutoff frequency of the moving average filter used for the moving average processing is set. The plant control adjustment device according to claim 2, _{wherein the second LPF cutoff frequency candidate f cb} is calculated as the second LPF cutoff frequency candidate f _cb as shown in the following equation (B).

Images