JP2009015514A - Pid control device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To automatically adjust PID constants with high precision while operating a device even when device characteristics rapidly change, or a non-linear behavior is shown. <P>SOLUTION: This PID control device is provided with a device; a measurement means; a storage means; a model creation means for creating a prediction model; a calculation means for predicting a control characteristic value a prescribed time ahead by using the prediction model, and for calculating two or more evaluation functions I<SB>m</SB>by using predicted control characteristic value; a determination means for determining a proportional gain K<SB>p</SB>as a PID constant, an integration time T<SB>i</SB>, and a differential time T<SB>d</SB>in order to minimize the square sum E<SB>RNN</SB>of the evaluation function; and a control means for operating the PID control of the control characteristic value based on the PID constant. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a PID control device capable of controlling a control characteristic value of a device with high accuracy while operating the device.

  Conventionally, automation control of industrial devices by PID control has been performed. Generally, this automation control is performed by inputting a PID constant to the transfer function represented by the following formula (1).

  In general, a specific value corresponding to the operating condition is input as the PID constant. However, if each PID constant is kept constant, the accuracy of PID control of the apparatus may be reduced due to disturbances, changes in apparatus characteristics over time, and the like. In such a case, it is necessary to adjust the PID constant that reflects disturbances and changes in device characteristics.

Therefore, in recent years, a PID control device capable of automatically adjusting the PID constant has been proposed.
Patent Document 1 discloses a control device that automatically adjusts a PID constant by forcibly applying an on / off signal (disturbance) to a control system and observing a response waveform of a control target.

  In Patent Documents 2 and 3, the vibration waveform of the deviation observed when the control system is operated under normal conditions is observed, the adjustment coefficient of the PID constant is obtained from the amplitude ratio of the continuous vibration waveform, and this adjustment is made to the predetermined PID constant. A control device that performs automatic adjustment by multiplying by a coefficient is disclosed.

  Patent Document 4 discloses a control device that models a control target based on a step response and adjusts a PID parameter by internal model control approximated based on the control amount, operation amount, gain / time constant ratio, and dead time. Has been.

  In Patent Document 5, the integral gain and the proportional gain are adjusted by measuring the sensitivity and period of the limit vibration when only the integral gain is changed among the PID constants, and when only the proportional gain is changed. A method for adjusting the parameters is disclosed.

Patent Document 6 discloses a PID control device that defines a fitness function by rise time, settling time, damping rate, overshoot amount, etc., and adjusts the PID gain while minimizing this function by a genetic algorithm. .
JP-A-3-3005 Japanese Patent Laid-Open No. 4-138501 JP-A-4-84201 JP-A-8-110802 JP-A-7-281709 JP 10-31503 A

  However, in the method for adjusting the PID constants described in Patent Documents 1 to 6, when the device characteristics to be controlled change abruptly or when the device to be controlled exhibits non-linear behavior, the PID constant is highly accurate. It was difficult to control. Moreover, in the adjustment method of the PID constant described in Patent Documents 1 to 6, there is a method in which a disturbance is applied in the apparatus and its response waveform is used. In such a preparation method, during the adjustment of the PID constant, There was a case where it was necessary to stop the apparatus once or to make an offline adjustment.

Therefore, the present inventor has intensively studied in order to solve the above problems. As a result, the control characteristic value is predicted by the prediction model created by the model creation means, the evaluation function _Im is calculated by the calculation means from the predicted characteristic value, and the square sum E _RNN of the evaluation function is minimized by the decision means. It was discovered that the PID constant should be determined so that And based on the PID constant determined in this way, it discovered that the PID control of an apparatus should just be performed by a control means. That is, the present invention has an object to provide a PID control device capable of performing PID control with high accuracy by automatically adjusting a PID constant even if the device characteristics change suddenly or exhibits nonlinear behavior. And

In order to solve the above problems, the present invention is characterized by having the following configuration.
1. Equipment,
Measuring means for measuring characteristic values and control characteristic values for performing PID control from the apparatus;
Storage means for storing the measured characteristic values and control characteristic values;
Model creation means for creating a prediction model for predicting a control characteristic value ahead of a predetermined time using the measured characteristic value and control characteristic value;
With predicting control characteristic value of the predetermined time later by using the prediction model, and calculating means for using at least predicted control characteristic value, calculating two or more evaluation function I _m,
Determining means for determining a proportional gain K _p , an integration time T _i , and a derivative time T _d that are PID constants so as to minimize the _square sum E _RNN of the evaluation function represented by the following formula (1):

Control means for performing PID control of the control characteristic value of the device based on the proportional gain K _p , the integration time T _i and the differentiation time T _d determined by the determination means;
A PID control device comprising:

  2. 2. The PID control device according to 1 above, wherein the device is an oil refinery plant.

3. The model creation means creates two or more prediction models that predict different types of control characteristic values,
The calculation means predicts a corresponding control characteristic value using each prediction model, calculates the evaluation function using at least the predicted control characteristic value,
The determining means determines the proportional gain K _p , integration time T _i , and derivative time T _d of each control characteristic value,
The control means performs PID control of each control characteristic value based on the proportional gain K _p , the integration time T _i , and the differential time T _d determined by the determination means. The PID control device described.

  4). 4. The PID control device according to any one of 1 to 3, wherein the control characteristic value is at least one selected from the group consisting of temperature, flow rate, and pressure.

  5). 5. The PID control apparatus according to claim 1, wherein the calculation unit calculates four functions represented by the following formulas (2) to (5) as the evaluation function.

6). In the PID control device, every time a predetermined time elapses,
The measuring means newly measures at least one of a characteristic value and a control characteristic value as a new characteristic value,
The storage means stores the new characteristic value,
The calculation means predicts a new control characteristic value ahead of a predetermined time by the prediction model using the new characteristic value, and newly calculates the evaluation function using at least the control characteristic value predicted by the prediction model. Calculate
It said determining means, so as to minimize the square sum E _RNN of the evaluation function I _m which is newly calculated, but a newly updated PID constants proportional gain K _p, integration time T _i, and the derivative time T _d Decide
The control means updates a PID constant to a value of the updated PID constant, and performs PID control of a control characteristic value of the device based on the updated PID constant.
6. The PID control device according to any one of 1 to 5 above.

  7. 7. The PID control apparatus according to 6, wherein the calculation unit further calculates a function represented by the following formulas (6) to (8) as a part of the evaluation function.

  8). 8. The PID control device according to any one of 1 to 7, wherein a transfer function of the PID control by the control means is expressed by the following formula (9).

  In the present specification, the “control characteristic value” refers to a characteristic value that is controlled by the prediction model and is controlled to be a target value based on the PID constant determined by the determining unit. In this respect, the “control characteristic value” is distinguished from the “characteristic value” that is measured by the measuring means but is not controlled so as to be the target value.

  The PID control device of the present invention can automatically adjust the PID constant while operating the device, even if the device characteristics change abruptly or exhibit non-linear behavior. As a result, the PID control of the apparatus can be performed with high accuracy.

  The PID control device of the present invention includes a device, a measurement unit, a storage unit, a model creation unit, a calculation unit, a determination unit, and a control unit. FIG. 2 schematically shows a device configuration of the PID control device of the present invention. In addition, the arrow in FIG. 2 represents the direction of data processing.

  First, a characteristic value and a control characteristic value for performing PID control are measured from the apparatus 1 by the measuring unit 2, and the measured characteristic value and control characteristic value are stored as data in the storage unit 3. Next, the model creation means 4 creates a prediction model that can predict a control characteristic value ahead of the current time by using the measured characteristic value and control characteristic value. The model creation means 4 creates a prediction model using a neural network. As a specific processing method thereof, a back propagation method (Back Propagation Method) is used.

  Then, the calculation means 5 predicts the control characteristic value ahead of a predetermined time by substituting the measured characteristic value and, if necessary, the control characteristic value into the prediction model. At the same time, an evaluation function is calculated using at least the predicted control characteristic value.

Next, the determining means 6 determines the proportional gain K _p , the integration time T _i , and the differential time T _d that are PID constants so that the _square sum E _RNN of the calculated evaluation function is minimized. The PID constant is determined by the determining means 6 using an RNN (Recurrent Neural Network), and a back propagation method is used as a specific processing method. Then, the control means 7 can perform PID control of the control characteristic value of the apparatus based on the PID constant determined in this way.

As described above, in the PID control apparatus of the present invention, the proportional gain K _p , the integration time T _i , and the differentiation time T _d are determined using the control characteristic value predicted by the prediction model. For this reason, the PID constant can be optimized without actually giving a stimulus such as a disturbance to the apparatus. As a result, it is possible to improve the operation time and productivity of the apparatus. Further, since the neural network and the RNN (recurrent neural network) are used for the creation of the prediction model and the determination of the PID constant, respectively, the PID control of the control characteristic value can be performed with high accuracy.

Hereinafter, according to the flowchart of FIG. 3, the processing content by each means of the PID control apparatus of this invention is demonstrated in detail.
Specific examples of the apparatus include a whole plant or a part of a plant, a manufacturing apparatus or a processing apparatus for a specific material / substance, a plant configured by a single apparatus or a plurality of apparatuses. However, it is not particularly limited to these devices. Moreover, as this apparatus, an oil refinery plant can be mentioned, for example.

  Further, this apparatus has one or more characteristic values representing the internal state of the apparatus, the raw material supplied into the apparatus, the state / physical properties of the product / processed product discharged from the apparatus, and the like. As this characteristic value, for example, the flow rate, temperature, pressure, etc. of fluids such as raw materials and products, characteristic values calculated by combining these characteristic values, characteristic values obtained by performing special arithmetic processing on these characteristic values, etc. Can be mentioned.

  First, the measuring means measures the characteristic value and the control characteristic value of this apparatus (S1). There may be one or more types of characteristic values and control characteristic values to be measured. The characteristic values and control characteristic values may be one type or plural types, but typically one type of characteristic value. Measure the value and control characteristic value. Further, the measurement of one kind of characteristic value and control characteristic value by the measuring means may be performed at one place in the apparatus or at a plurality of places. Specific examples of the measuring means include a temperature sensor, a pressure sensor, and a flow meter.

  This measuring means can measure the characteristic value and the control characteristic value under a predetermined condition according to the measurement condition inputted in advance. Specifically, the measuring means inputs the position of measuring the characteristic value and control characteristic value in the apparatus, the measurement time interval, the number of measurement samplings, etc. Measurement can be performed.

  Next, the storage means stores the characteristic value and the control characteristic value measured by the measuring means (S2). The storage means may store only characteristic values and control characteristic values required by the operator, and can delete unnecessary characteristic values and control characteristic values. The storage means may be hardware in which a computer program for executing various processes is stored in advance, for example, a ROM (Read Only Memory), a HDD (Hard Disc Drive), or a storage device that is replaceably mounted. It is possible to implement with a compact disc (CD) -ROM, a flexible disc-cartridge (FD) and a combination thereof.

  The model creating unit creates a prediction model for predicting a control characteristic value ahead of a predetermined time based on the characteristic value and the control characteristic value stored in the storage unit. One or more characteristic values may be used for creating this prediction model, but typically a prediction model is created based on one type of characteristic value and control characteristic value.

  In this model creation means, a prediction model for predicting a control characteristic value ahead of a predetermined time is created by a back propagation method based on the steepest descent method. The steepest descent method is a calculation method for creating an optimum prediction model by iterative calculation for the first time from a certain appropriate initial value. This method will be described in detail below.

First, let x _i ⁽¹⁾ (1 ≦ i ≦ n ₁ ) be the characteristic value used to create the prediction model and the value of the control characteristic value used as necessary. Then, s _j ⁽¹⁾ is defined from x _i ⁽¹⁾ (1 ≦ i ≦ n ₁ ) and x ₀ ^{(1) as} follows. Note that a control characteristic value may or may not be used for x _i ⁽¹⁾ (1 ≦ i ≦ n ₁ ).

In the above equation, n ₁ +1 corresponds to the number of input layers in the back propagation method.

Next, s _k ⁽²⁾ is defined as follows using this s _j ⁽¹⁾ .

In the above equation, n ₂ +1 corresponds to the number of intermediate layers in the back propagation method.
Next, linear processing is performed, and s _k ⁽²⁾ defined as described above is set as a control characteristic value to be predicted as it is.

FIG. 4 schematically shows the processing of the above formulas (11) to (15) in the back propagation method. First, linear processing using a linear function is performed on the measured n ₁ characteristic values x _i ⁽¹⁾ (1 ≦ i ≦ n ₁ ) and one x ₀ ⁽¹⁾ . In this linear processing, the input x _i ⁽¹⁾ is used as it is, and x _i ⁽¹⁾ becomes the input layer. The number of input layers is n ₁ +1.

Next, s _j ⁽¹⁾ is calculated using the input characteristic value x _i ⁽¹⁾ and the coupling load w _ji ⁽¹⁾ . Thereafter, x _j ⁽²⁾ (1 ≦ j ≦ n ₂ ) is calculated by substituting s _j ⁽¹⁾ into the sigmoid function expressed by the equation (15). These x _j ⁽²⁾ (1 ≦ j ≦ n ₂ ) and x ₀ ⁽²⁾ are intermediate layers, and the number of layers is n ₂ +1. Next, s _k ⁽²⁾ is calculated using the calculated x _j ⁽²⁾ and the coupling weight w _kj ⁽²⁾ . This s _k ⁽²⁾ is an output layer, and the number of layers is one. Then, a control characteristic value is finally obtained by performing linear processing using a linear function on s _k ⁽²⁾ . Here, in linear processing using a linear function, s _k ⁽²⁾ is used as it is as a control characteristic value.

Further, from the control characteristic value predicted by the prediction model at time t _{1 and} the control characteristic value actually measured by the measuring means at time t ₁ from y, the error between the predicted value and the actual measurement value is expressed as follows: _It is defined as _NE .

In the back-propagation method, optimum coupling loads w _ji ⁽¹⁾ and w _kj ⁽²⁾ that minimize this error E _NE are obtained by the steepest descent method. That is, with respect to the coupling weights w _ji ⁽¹⁾ and w _kj ⁽²⁾ , if appropriate positive constants η are used as learning coefficients, the correction amounts Δw _ji ⁽¹⁾ and Δw _kj ⁽²⁾ are as follows: Become.

At this time, the new w _ji ⁽¹⁾ and w _kj ⁽²⁾ are modified as follows.

Here, ∂E _NE / ∂w _ji ⁽¹⁾ and ∂E _NE / ∂w _kj ⁽²⁾ are calculated as follows based on the chain rule of differentiation.

Then, initial values are input in advance as w _ji ⁽¹⁾ and w _kj ⁽²⁾ , and ∂E _NE / ∂w _ji ⁽¹⁾ and ∂E _NE / ∂ calculated based on the above formulas (21) and (22). From w _kj ⁽²⁾ , w _ji ⁽¹⁾ and w _kj ⁽²⁾ are determined based on the above equations (17) to (20).

Then, thus determined w _ji ⁽¹⁾ and w _kj ⁽²⁾ as an initial value, at time t ₂ has advanced a predetermined time than the time t ₁ performs the same processing as described above w _ji ^{(1 ) And} w _kj ⁽²⁾ are determined. Thereafter, the same processing is performed while the time is sequentially advanced with t ₃ , t ₄ , t ₅ ..., And initial _values of w _ji ⁽¹⁾ and w _kj ⁽²⁾ when performing the processing of t _(n). As values, w _ji ⁽¹⁾ and w _kj ⁽²⁾ determined by the processing at t _(n−1 ⁾ are used.

In this way, by repeatedly performing a predetermined number of times (S5), finally determining w _ji ⁽¹⁾ and w _kj ⁽²⁾ (S3) and creating a prediction model (S4). For w _ji ⁽¹⁾ and w _kj ⁽²⁾ , after determining w _ji ⁽¹⁾ and w _kj ⁽²⁾ as initial values by a predetermined method in advance, repeat calculation for the same data range. May be determined.

This created prediction model, because w _ji ⁽¹⁾ and w _kj ⁽²⁾ is determined, only the x _i ⁽¹⁾ has become a unknown, by entering the x _i ^(1), a predetermined The control characteristic value ahead of time can be predicted. In addition, the range of the data (x _i ⁽¹⁾ ) to be used is different each time the processes (the processes of equations (17) to (22)) are repeated.

  Further, since the characteristic value and the control characteristic value used for creating the prediction model vary greatly depending on the type, the normalized value may be used for creating the prediction model. Examples of the normalization method include the following methods.

Since the prediction model created by the model creation means uses the back propagation method, it is possible to predict the control characteristic value ahead of a predetermined time with high accuracy.

  Next, the calculation means predicts a control characteristic value ahead of a predetermined time using a prediction model (S6), and calculates an evaluation function using at least the predicted control characteristic value (S7). At this time, when the characteristic value and the control characteristic value used to create the prediction model are normalized, a process of removing the normalization is performed when calculating the evaluation function. Specific examples of this method include the following methods.

  Two or more evaluation functions need to be used, but the specific form of the expression is not particularly limited as long as the control characteristic value can be predicted (approximated). The evaluation function is calculated using at least the control characteristic value predicted by the prediction model. However, other characteristic values or actual measured values of the control characteristic value may be used for this calculation.

  For example, the evaluation functions are preferably four functions represented by the following formulas (2) to (5). The evaluation functions represented by the following formulas (2) to (5) can approximate the behavior of the characteristic value with high accuracy. Moreover, even if the control characteristic value as a device exhibits complicated behavior as in an oil refinery plant, it can be approximated with high accuracy.

Note that T ₀ represents the time when the control characteristic value starts to change because the operating condition of the apparatus that affects the control characteristic value is changed in order to change the target value of the control characteristic value.

FIG. 1 is a diagram showing the significance of the evaluation functions I ₁ , I ₂ , and I ₄ . In FIG. 1, I ₁ is a change portion (area of a gray portion) from the control characteristic value before the change to the target value of the control characteristic value, and I ₂ is a target value after the control characteristic value reaches the target value. The vibration part (the area of the gray part), I ₄ , represents the difference between the maximum value and the minimum value when the control characteristic value oscillates after the control characteristic value reaches the target value. That is, I ₁ represents an evaluation of response responsiveness, and I _{2 to} I ₄ represent an evaluation of response stability. Therefore, when the evaluation functions I _{1 to} I ₄ are used, highly accurate PID control is possible by determining the PID constant so as to minimize the _square sum E _RNN of these evaluation functions.

  Since this evaluation function varies greatly depending on the value of the control characteristic value used for the calculation, a normalized one may be used. Examples of the normalization method include the following methods.

Next, the determining means determines the proportional gain K _p , the integration time T _i , and the differentiation time T _d so that the _square sum E _{RNN of the} evaluation function represented by the following formula (1) becomes the minimum value. .

The proportional gain K _p , the integration time T _i , and the derivative time T _d are determined by the back propagation method based on the steepest descent method, similarly to the creation of the prediction model by the model creation means. However, in the processing by the model creation means, the characteristic value measured in the apparatus as the input value and the control characteristic value as necessary are used. However, in the processing by this determination means, the calculated value by the evaluation function is used as the input value. The point to use is different. In addition, in the processing by the model creation means, a prediction model was created by correcting the connection loads w _ji ⁽¹⁾ and w _kj ⁽²⁾ as needed by performing a predetermined number of iterations. The difference is that the joint loads w _ji ⁽¹⁾ and w _kj ⁽²⁾ are determined once, or w _ji ⁽¹⁾ and w _kj ⁽²⁾ are determined every time the update time elapses.

The method for determining the proportional gain K _p , the integration time T _i , and the differentiation time T _d that minimizes E _RNN by the determination means will be described in detail below.
First, s _j ⁽¹⁾ is defined as follows using the evaluation function I _i calculated by the calculation means.

In the above equation, n ₁ corresponds to the number of input layers in the back propagation method.

Next, s _k ⁽²⁾ is defined as follows.

In the above equation, n ₂ +1 corresponds to the number of intermediate layers in the back propagation method.

The s _k ^{(2) is} linearly processed by a linear function to finally obtain three PID constants (proportional gain K _p , integration time T _i , and differentiation time T _d : the following equation (34) and In (35), g ₁ = K _p (= s ₁ ⁽²⁾ ), g ₂ = T _i (= s ₂ ⁽²⁾ ), and g ₃ = T _d (= s ₃ ⁽²⁾ ). Get. Here, in linear processing using a linear function, s _k ⁽²⁾ is used as it is as the PID constant g _k (proportional gain K _p , integration time T _i , and differentiation time T _d ).

FIG. 5 schematically shows the processing of the above formulas (26) to (29) in the back propagation method. First, linear processing using a linear function is performed on n ₁ evaluation functions I _i calculated by the calculation means. In this linear processing, the input evaluation function is used as it is, and I _i is an input layer, and the number of layers is n ₁ .

Next, s _j ⁽¹⁾ is calculated using the input evaluation function I _i and the coupling weight w _ji ⁽¹⁾ . Thereafter, x _j ⁽²⁾ is calculated by substituting s _j ⁽¹⁾ into the sigmoid function expressed by the equation (29). This x _j ⁽²⁾ is an intermediate layer, and the number of layers is n ₂ +1. Next, s _k ⁽²⁾ (= g _k ) is calculated using the calculated x _j ⁽²⁾ and the coupling weight w _kj ⁽²⁾ . This s _k ⁽²⁾ (= g _k ) is an output layer, and the number of layers is three. Then, linear processing using a linear function is performed on this s _k ⁽²⁾ (= g _k ) to finally obtain a proportional gain K _p , an integration time T _i and a differentiation time T _d which are predetermined PID constants. . Here, in linear processing using a linear function, s _k ⁽²⁾ (= g _k ) is used as it is as a PID constant.

In order to determine the PID constant as described above, first, 同 様 E _RNN / ∂w _ji ⁽¹⁾ and ∂E _RNN / ∂w _kj ^{(2 )}

At this time, the new w _ji ⁽¹⁾ and w _kj ⁽²⁾ are modified as follows.

Here, ∂E _RNN / ∂w _ji ⁽¹⁾ and ∂E _RNN / ∂w _kj ⁽²⁾ are calculated as follows based on the chain rule of differentiation.

In addition, ∂I _m / ∂g ₁ in the above formulas (34) and (35) is obtained from the proportional gains K _p [1] and I [1] previously input to the control means as initial values, ∂I _m / ∂g ₁ = (I−I [1]) / (K _p −K _p [1]). Similarly, ∂I _m / ∂g ₂ = (I−I [1]) / (T _i −T _i [1]), ∂I _m / ∂g ₃ = (I−I [1]) / ( T _d −T _d [1]). Here, T _i [1] and T _d [1] represent the integration time and the differentiation time input to the control means in advance, respectively. In addition, k = 1, 2, and 3, g ₁ = K _p , g ₂ = T _i , and g ₃ = T _d .

Then, w _ji ⁽¹⁾ and w _kj ⁽²⁾ can be finally determined according to the above formulas (30) to (35) (S8). Further, from the w _ji ⁽¹⁾ and w _kj ⁽²⁾ determined in this way, the proportional gain K _p , the integration time T _i , and the differentiation time T _d can be finally obtained (S9).

Then, the control means changes the PID constant to the proportional gain K _p , the integration time T _i and the differentiation time T _d determined as described above, and adjusts the characteristic value based on the PID constant, thereby controlling the control characteristics of the apparatus. PID control is performed so that the target value is reached. Thus, the conditions that are actually adjusted to obtain the target control characteristic value are typically the characteristic values used to create the prediction model.

At this time, when the evaluation function used to determine the PID constant is normalized, a process of denormalizing the obtained PID constant g _k is performed. Specifically, as a method of removing this normalization, for example, the following method can be cited.

  Further, in the PID control device of the present invention, each means only needs to be formed so as to realize its function. For example, dedicated hardware that exhibits a predetermined function, a predetermined function is given by a computer program Or a predetermined function realized by each means by a computer program, a combination thereof, or the like.

  In the PID control apparatus of the present invention, each means does not need to be an individual independent entity, a plurality of means are formed as one member, and a certain means is a part of another means, It is possible that a part of one means and a part of another means overlap.

  In addition, the model creation unit, the calculation device, and the determination unit of the present invention may be hardware that can read a computer program and execute a corresponding processing operation. Specifically, hardware including a CPU (Central Processing Unit) as a main body and various devices such as a ROM, a RAM (Random Access Memory), and an I / F (Interface) unit may be connected thereto.

  In the present invention, the apparatus is preferably an oil refinery plant. An oil refinery plant typically has thousands of control loops to maintain a target production quality. In addition, since multiple control loops interfere with each other, it is difficult to grasp and model the physical behavior of the characteristic values with the conventional method, and it is also difficult to stop the plant halfway. . Therefore, by applying a neural network to an oil refinery plant as an apparatus of the present invention, a prediction model is created, a PID constant is determined using this prediction model, and further, PID control is performed using this PID constant, thereby increasing the The control characteristic value of the plant can be controlled with accuracy.

In the PID control device of the present invention, it is preferable that the measuring means measures a control characteristic value of a type different from the characteristic value. Further, it is preferable that the model creating means creates two or more prediction models for predicting different types of control characteristic values. Preferably, the calculation means predicts a corresponding control characteristic value using each prediction model, and calculates an evaluation function corresponding to each prediction model using at least the predicted control characteristic value. The determining means preferably determines the proportional gain K _p , the integration time T _i , and the derivative time T _d of each control characteristic value. The control means preferably controls each control characteristic value based on the proportional gain K _p , the integration time T _i , and the differential time T _d determined by the determination means. That is, in this case, the prediction model is created for each control characteristic value, and the processing represented by the above formulas (1), (11) to (22), and (26) to (35) is performed. . Moreover, you may perform the process of said Formula (2)-(9), (23)-(25), and (36) for every control characteristic value.

In recent years, devices used have become larger and more complicated, and there are often two or more types of control characteristic values controlled in the device. In the present invention, even in such a large and complicated apparatus, it is possible to determine two or more kinds of control characteristic values K _p , T _i , and T _d as described above. A stable and highly accurate PID control of a plurality of control characteristic values becomes possible.

  In the present invention, it is preferable that the control characteristic value for performing the control so as to be the target value is at least one physical property value selected from the group consisting of temperature, flow rate, and pressure. These control characteristic values are basic quantities that greatly affect the characteristics, raw materials, and products in the equipment. By controlling these control characteristic values, stable control of equipment, raw materials, and products is possible. Become.

In the PID control device of the present invention, it is preferable that each means performs processing of the following flows (A) to (E) each time the update time elapses.
(A) The measuring means newly measures at least one of the characteristic value and the control characteristic value as a new characteristic value.
(B) The storage means stores the new characteristic value.
(C) The calculation means predicts a new control characteristic value ahead of a predetermined time by the prediction model using the new characteristic value, and recalculates an evaluation function from the control characteristic value predicted by the prediction model.
(D) determining means, 2 a square sum E _RNN to minimize a newly updated PID constants proportional gain K _p, integration time T _i, and derivative time of the evaluation function I _m which re newly computed Determine _Td .
(E) The control means updates the PID constant to the value of the updated PID constant, and performs PID control of the control characteristic value of the device based on the value of the updated PID constant.

FIG. 6 is a flowchart showing the process flow of the PID control device that updates the PID constants at predetermined time intervals in this way.
In this apparatus, first, as shown in the flowchart of FIG. 3, a prediction model is created by performing the processing of the above formulas (1), (11) to (22) and (26) to (35). And determine the PID constant (this process is not shown in FIG. 6). Next, when a predetermined update time inputted in advance elapses (S1), the measuring unit newly measures at least one of the characteristic value and the control characteristic value as a new characteristic value (S2). The new characteristic value is stored in the storage means (S3).

  Next, the calculation means calculates a control characteristic value (predicted value) using a prediction model created in advance using the new characteristic value (S4), and then calculates a new evaluation function using the control characteristic value. Correct again (S5). In this case, the evaluation function used at the time of determining the previous PID constant may or may not be used as the evaluation function, but it was used at the time of determining the previous PID constant in order to achieve stable control accuracy. It is preferable to use an evaluation function.

Then, determination unit is newly updated PID constants so as to minimize the square sum E _RNN of the evaluation function I _m which re newly calculated proportional gain K _p, integration time T _i, and derivative time T _d Is determined (S6, S7). At this time, as the updated PID constant, an evaluation function newly recalculated is used as the evaluation function, and in the above formulas (34) and (35), ∂I _m / ∂g ₁ = (I [N] −I [N−1]) / (K _p [N] −K _p [N−1]), ∂I _m / ∂g ₂ = (I [N] −I [N−1]) / (T _i [N ] −T _i [N−1]), ∂I _m / ∂g ₃ = (I [N] −I [N−1]) / (T _d [N] −T _d [N−1]) . Other than this, the updated PID constant is set by the same method as the process for determining the first PID constant (the processes of the above formulas (1), (11) to (22) and (26) to (35)). decide. Here, I [N], _Kp [N], _Ti [N], and _Td [N] are the evaluation function, proportional gain, integration time, and derivative time at the time of updating (Nth time), respectively. Represents. Also, I [N-1], _Kp [N-1], _Ti [N-1], and _Td [N-1] are the evaluation function and proportionality at the time of the previous (N-1) update, respectively. Represents gain, integration time, and derivative time.

Finally, the control means updates the PID constant to the updated PID constant value (S7), and controls the control characteristic value of the apparatus based on the updated PID constant value.
Such an update is repeated until the number of updates input in advance as the update condition is reached (S9).

  By configuring the PID control device of the present invention so that the PID constant can be updated every predetermined time as described above, the control characteristic value can be controlled with high accuracy even when the device is operated for a long time. Is possible. In addition, even when the device characteristics change suddenly due to disturbances or changes in the atmospheric environment of the device, the control characteristic values can be controlled with high accuracy.

The update time of the PID constant may be set as appropriate in accordance with the properties of the device characteristics and control characteristic values. For example, the update time may be shortened for a device whose control characteristic value changes frequently, and the update time may be lengthened for a device whose control characteristic value is stable. Alternatively, the update time may be determined from the amount of change in the control characteristic value per unit time by measuring the control characteristic value in advance for a predetermined time. Further, the number of updates of the PID constant may be set as appropriate according to the operation period of the apparatus.
Further, each time the update time elapses, each means performs the following processes (A) to (E), and the process of determining the update PID constant is performed for the PID constants of a plurality of control characteristic values. You may go.

  Further, when the PID control device updates the PID constant every predetermined time, it is preferable that the calculation means further calculates functions represented by the following formulas (6) to (8) as a part of the evaluation function.

Each of the evaluation functions of the above formulas (6) to (8) includes each update PID constant determined by the first to (N-1) th update. Each update PID constant is multiplied by higher order d ₁ (0 <d ₁ <1) as the update PID constant when the number of updates is small. For this reason, the evaluation functions of the above formulas (6) to (8) represent the influence of the history of the update PID constants up to the previous update, and the update PID constants at the time of update with a larger number of updates are the current update PIDs. The influence on the constant is large. Therefore, by updating the current update PID constant using these evaluation functions, it is possible to determine the update PID constant with higher accuracy reflecting the history of the update PID constants up to now.

FIG. 7 schematically shows the process of the determination unit when determining the update PID constant using the evaluation functions of the above formulas (2) to (5) and (6) to (8). is there. In this process, in addition to the evaluation functions I _{1 to} I ₄ , the history of updated PID constants updated up to the previous time is fed back and the evaluation functions I _{5 to} I ₇ are used as input layers. As described above, the history up to the previous time is fed back as the evaluation functions I _{5 to} I ₇ every time it is updated, thereby enabling highly accurate PID control. In the process of FIG. 7, the other processes are performed by the same method as in FIG.

  The transfer function of PID control of the control characteristic value of the apparatus by the control means of the present invention is preferably represented by the following formula (9).

By using such a transfer function, PID control can be performed with higher accuracy.

(Example 1)
A desulfurization apparatus was incorporated in the PID control apparatus of the present invention as an apparatus for performing PID control, and PID control was performed. FIG. 8 shows an outline of this desulfurization apparatus. The raw materials and products flow in the desulfurization apparatus of FIG. 8 as follows.

  After the raw material oil is supplied from the inlet 11 and the pressure is increased, it is mixed with hydrogen. Then, this mixed composition is supplied to the heating furnace 15 and heated to a predetermined temperature, and then introduced into the reaction tower 16. In this reaction tower 16, after desulfurization by hydrogenation reaction, the gas-liquid mixture is cooled. Thereafter, it is further introduced into the separation tank 17 to separate the gas and oil. Next, the separated oil is guided to the stripper 18 to separate hydrogen sulfide and accompanying gas. Thereafter, after removing moisture, it is introduced into a tank (not shown) as a final product.

  In this embodiment, in the desulfurization apparatus, the opening degree of the valves 12 and 19 is a characteristic value, and the flow rate in the flow meter 13 and the temperature in the thermometer 14 are control characteristic values for performing PID control. These flow rates and temperatures were subjected to PID control by controlling the opening degree of the valves 12 and 19. An orifice flow meter was used as the flow meter 13 (measuring means), and a thermocouple thermometer was used as the thermometer 14 (measuring means).

(A) PID control of flow rate The transfer function of the above equation (9) was used as the transfer function of PID control by the control means. First, as an initial value of the PID constant of the control means, the proportional gain K _p is 0.4, the integration time T _i is 6 minutes, the differential time T _d is 0.5 minutes, and the differential gain a is 0.1. The proportional gain K _p , integration time T _i , and differentiation time T _d were optimized.

  For the apparatus shown in FIG. 8, first, the measurement means measures the opening degree (characteristic value) of the valve 12 and the flow rate (control characteristic value) at the flow meter 13 every minute. Was corrected to data every 0.5 seconds by interpolation with a spline function. And these data were memorize | stored in the memory | storage means. 9 is a diagram showing a part of flow rate (control characteristic value) data in the flow meter 13 actually measured and interpolated in this way.

Next, among the stored data, the opening (characteristic value) of the valve 12 every 1 minute from the predetermined time to 14 minutes before, 15 points, every 1 minute from the 10 minutes before the predetermined time to 1 minute before Using 10 points of flow rate (control characteristic value) in the flow meter 13, a prediction model for predicting the flow rate (control characteristic value) in the flow meter 13 0.5 seconds ahead was created by the model creation means. In this case, the input layer (n ₁ +1 in the above formula (11)) in the model creating means is 25, and the intermediate layer (n ₂ +1 in the above formula (13)) is 15.

Then, w _ji ⁽¹⁾ with respect to the portion of FIG. 9 (A) and w _kj ⁽²⁾ learning the determination of (w _ji ⁽¹⁾ and w _kj ⁽²⁾ a process of determining a) the 3000 times, repeating Finally, w _ji ⁽¹⁾ and w _kj ⁽²⁾ were determined, and the prediction model was confirmed.

  Next, the calculation means uses the opening degree (characteristic value) of the valve 12 and the flow rate (control characteristic value) at the flow meter 13 as data, and substitutes them into the prediction model determined as described above for 0.5 seconds. The flow rate (control characteristic value) at the previous flow meter 13 was predicted. Moreover, the evaluation function represented by following formula (2)-(8) was calculated with this.

  FIG. 10 shows the flow rate (control characteristic value) in the flow meter 13 predicted by the prediction model created in this way, and the flow rate (control characteristic value) in the flow meter 13 actually measured. . From the result of FIG. 10, the curve predicted by the prediction model created by the model creation means of the present invention is in good agreement with the curve of the actual measurement value, and the change in the flow rate at the flow meter 13 can be predicted with high accuracy. I understand that

Next, using the evaluation function determined as described above, the determination means uses a PID constant (proportional gain K) so as to minimize the _square sum E _RNN of the evaluation function represented by the above formula (1). _p , integration time T _i , and derivative time T _d ) were determined. At this time, specifically, the processes of the above formulas (1) to (36) were performed. In this case, the input layer (n _{1 in the} above formula (26)) is 7 and the intermediate layer (n ₂ +1 in the above formula (27)) is 5. And the PID control of the flow volume by the flowmeter 13 was performed by controlling the opening degree of the valve | bulb 12 using the PID constant determined in this way by the control means.

In addition, the following processing was performed every minute.
(1) The opening of the valve 12 and the flow rate at the flow meter 13 were newly measured as new characteristic values by the measuring means.
(2) The new characteristic value is stored by the storage means.
(3) The calculation means predicts the flow rate at the flow meter 13 0.5 seconds ahead by substituting the new characteristic value into the prediction model determined as described above, and also predicts it according to the prediction model. The evaluation function was recalculated from the calculated flow rate.
(4) determination means is a newly updated PID constants so as to minimize the square sum E _RNN of the evaluation function I _m which re newly calculated proportional gain K _p, integration time T _i, and derivative time T _d was determined.
(5) The control means performs PID control of the flow rate at the flow meter 13 by updating the PID constant to the updated PID constant value and controlling the opening of the valve 12 based on the updated PID constant value. It was.
And the process of said (1)-(5) was performed 30 times in total.

FIGS. 11 and 12 show the influence of the number of updates of the PID constant on the evaluation functions I _{1 to} I ₄ and the PID constant (proportional gain K _p , integration time T _i , and derivative time T _d ). In the result of FIG. 11, the number of updates that converges to a constant value varies depending on the evaluation function, but it can be seen that each evaluation function converges to a substantially constant value when the number of updates is about 15. Further, from the results of FIG. 12, it can be seen that each PID constant converges to a constant value after every 15 updates in which the evaluation function converges to a constant value. From these results, it can be seen that the PID control device of the present invention can perform automatic tuning of PID constants with high accuracy.

(B) Temperature control PID control of the temperature in the thermometer 14 is performed in the same manner as in the control of the flow rate (a) except that the opening of the valve 19 is a characteristic value and the temperature in the thermometer 14 is a control characteristic value. went. As a result, like the flow rate, the number of updates was about 15, and each evaluation function and each PID constant converged to a constant value. From these results, it can be seen that the PID control device of the present invention can perform automatic tuning of PID constants with high accuracy.

Is a diagram illustrating the meaning of the evaluation function I _1, I ₂ and I _4. It is a figure showing an example of the apparatus structure of the PID control apparatus of this invention. It is a flowchart showing an example of the process in the PID control apparatus of this invention. It is a figure showing an example of the process by the model preparation means of the PID control apparatus of this invention. It is a figure showing an example of the process by the determination means of the PID control apparatus of this invention. It is a flowchart showing an example of the process in the PID control apparatus of this invention. It is a figure showing an example of the process by the determination means of the PID control apparatus of this invention. 3 is a diagram illustrating an apparatus that performs PID control in Embodiment 1. FIG. It is a figure showing the measurement result of the flow rate of the apparatus of Example 1. It is a figure showing the measurement result of the flow rate of the apparatus of Example 1, and the prediction result by a prediction model. It is a figure showing the process result of Example 1. FIG. It is a figure showing the process result of Example 1. FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Apparatus 2 Measuring means 3 Storage means 4 Model preparation means 5 Calculation means 6 Determination means 7 Control means 11 Inlet part 12 Valve 13 Flowmeter 14 Thermometer 15 Heating furnace 16 Reactor 17 Separation tank 18 Stripper 19 Valve

Claims (8)

Equipment,
Measuring means for measuring characteristic values and control characteristic values for performing PID control from the apparatus;
Storage means for storing the measured characteristic values and control characteristic values;
Model creation means for creating a prediction model for predicting a control characteristic value ahead of a predetermined time using the measured characteristic value and control characteristic value;
With predicting control characteristic value of the predetermined time later by using the prediction model, and calculating means for using at least predicted control characteristic value, calculating two or more evaluation function I _m,
Determining means for determining a proportional gain K _p , an integration time T _i , and a derivative time T _d that are PID constants so as to minimize the _square sum E _RNN of the evaluation function represented by the following formula (1):

Control means for performing PID control of the control characteristic value of the device based on the proportional gain K _p , the integration time T _i and the differentiation time T _d determined by the determination means;
A PID control device comprising:   The PID control device according to claim 1, wherein the device is an oil refinery plant. The model creation means creates two or more prediction models that predict different types of control characteristic values,
The calculation means predicts a corresponding control characteristic value using each prediction model, calculates the evaluation function using at least the predicted control characteristic value,
The determining means determines the proportional gain K _p , integration time T _i , and derivative time T _d of each control characteristic value,
The control means performs PID control of each control characteristic value based on the proportional gain K _p , the integration time T _i , and the differential time T _d determined by the determination means. The PID control device described in 1.   The PID control device according to claim 1, wherein the control characteristic value is at least one selected from the group consisting of temperature, flow rate, and pressure. 5. The PID control apparatus according to claim 1, wherein the calculation unit calculates four functions represented by the following formulas (2) to (5) as the evaluation function. 6. .

In the PID control device, every time a predetermined time elapses,
The measuring means newly measures at least one of a characteristic value and a control characteristic value as a new characteristic value,
The storage means stores the new characteristic value,
The calculation means predicts a new control characteristic value ahead of a predetermined time by the prediction model using the new characteristic value, and newly calculates the evaluation function using at least the control characteristic value predicted by the prediction model. Calculate
It said determining means, so as to minimize the square sum E _RNN of the evaluation function I _m which is newly calculated, but a newly updated PID constants proportional gain K _p, integration time T _i, and the derivative time T _d Decide
The control means updates a PID constant to a value of the updated PID constant, and performs PID control of a control characteristic value of the device based on the updated PID constant.
The PID control apparatus according to any one of claims 1 to 5, wherein The PID control apparatus according to claim 6, wherein the calculation unit further calculates a function represented by the following formulas (6) to (8) as a part of the evaluation function.

The PID control apparatus according to any one of claims 1 to 7, wherein a transfer function of the PID control by the control means is expressed by the following formula (9).

Images