JPH103302A - Method for controlling model prediction - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize model prediction control where desired responsiveness is provided with few calculation loads by dividing controlled variable into the one with quick change and the one with slow change and executing the thinning of a prediction time concerning the latter so as to incorporate it in an evaluation function. SOLUTION: The manipulate variable of a controlled object 1 is adopted as u1 and u2 and its controlled variable is as yF and yS. High-speed response is required in yF but convergence is executed to a set value by slow response in yS. A controlled variable predicting means 2 calculates the past values of yF and yS and the prediction value pre-fixed portions yF*' and yS*' of yF and yS based on the past values of change quantities >=u1 and Δu2 in manipulated variable. The dimension of yF' is the number of sampling times within a prediction section. Since yS' is thinned, its dimension becomes aboust 1/K as compared with a case without thinning. A manipulated variable change quantity optimizing means 3 substitutes the obtained prediction value pre-fixed portions of yF and yS for the evaluation function so as to obtain the future values of the manipulated variable change quantities by applying a min. square method algorithm.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to the field of control of industrial systems. The present invention relates to a model for a control system in which it is difficult to realize control using available controller resources due to a large calculation load in conventional model predictive control. Related to the application of predictive control.

[0002]

2. Description of the Related Art Conventionally, model predictive control has been put to practical use mainly in the field of process control as a means for realizing high control performance which cannot be achieved by a simple control method such as PID control. In the model predictive control, an evaluation function J is determined as shown in Expression 1 with respect to future values of input / output variables based on a dynamic characteristic model of a control target, and an operation amount for minimizing the evaluation function J is obtained by performing an optimization calculation. In the following, a time section in which the control amount is predicted and evaluated is referred to as a prediction section, and a time section in which the operation amount is optimized is referred to as an operation section.

[0003]

(Equation 1) Here, r _i ′ is a vector configured by arranging the future values of the set values in the prediction section for each sampling cycle, y _i ′ is a vector configured by arranging the predicted values of the control amount y _i for each sampling cycle in the prediction section, Δu _i 'is a vector constructed by arranging the future value of each sampling period in the operation amount of the change amount of the operation section, W _ei weighting factor imposed on the magnitude of the control deviation, W _ui the weight imposed on the magnitude of the operating amount change amount The coefficient, i is each variable number, _Ny is the number of control amounts, and _Nu is the number of operation amounts.

[0004] The prediction section is set in a time region after the current time. The start point of the operation section is the current time, and the operation amount is held after the end point of the operation section. The symbol | ... | represents a norm (that is, size), and a Euclidean norm or the like is used. For the response of the control system, the prediction section, the operation section, and the weight coefficient are set to appropriate values, or the weight coefficient of the manipulated variable change amount is set to 0, and the set value given from the outside and the value of the control amount at the current time are used. The adjustment is performed by appropriately setting an internal setting value (referred to as a reference trajectory) to be set on the basis of the setting value and appropriately setting a weighting coefficient of a prediction section, an operation section, and a control deviation.

[0005]

In the conventional model predictive control, a control deviation at all sampling times in a prediction section of all control amounts, that is, a difference between a set value and a control amount is incorporated into an evaluation function to perform optimization calculation. It is carried out. As the number of control variables or the number of sampling steps in the prediction section increases, the scale of the optimization calculation increases, and the resources (memory area, processing capacity, etc.) of the controller required for realization are increased. This requires expensive hardware or is virtually impossible to implement with available resources, limiting its scope.

The present invention has been made to solve these problems, and has as its object to realize a model predictive control having a desired responsiveness with a small calculation load.

[0007]

According to the method of the present invention,
The control amount is divided into those that change quickly and those that change slowly. The former is incorporated into the evaluation function as in the conventional method, and the latter is thinned out at the predicted time and incorporated into the evaluation function. That is, the evaluation function is set as shown in Expression 2.

[0008]

(Equation 2) Here, r _Fi ′ is a vector configured by arranging scheduled values for each sampling period in the prediction section of the set value given to the fast-changing control amount y _Fi , and y _Fi ′ is a prediction of the fast-changing control amount y _Fi A vector configured by arranging predicted values for each sampling cycle in the section, r _Si ′ is a vector configured by arranging scheduled values for k sampling cycles in a prediction section of a set value given for a control amount y _Si that changes slowly, y _Si ′ is a vector configured by arranging predicted values for each k sampling periods in the prediction section of the slowly changing control amount y _Si , and Δu _i ′ is a vector arranging future values for each sampling cycle in the operation section of the operation amount change amount The constituent vectors,
W _Fei weighting factor imposed on the size of the fast control amount of the control deviation of change, W _Sei weighting factor imposed on the magnitude of the control deviation of the slow control the amount of change, the magnitude of the Wu _i is MV change amount weighting factors impose, i is the variable number, N _yf is the number of fast control amount of change, N _yS is the number of slow control the amount of change, N _u is the number of manipulated variables, k is y _Fi, response y _Si Is an integer appropriately determined according to the speed of

[0009] The prediction section is set in a time region after the current time. The start point of the operation section is the current time, and the operation amount is held after the end point of the operation section. Symbols... | Represent norms, such as Euclidean norms. Other norms may be employed as long as they are suitable for the control purpose of converging the control amount to the set value. The subsequent development is the same as that of the conventional model predictive control. The characteristic of the control system is the weight coefficient W
Adjust by changing _Fei , W _Sei and W _ui .

[0010] From Equation 2, the control deviation of all the sampling times in the prediction section is evaluated for the control amount y _Fi, so that the control deviation is suppressed by the same operation as the conventional model prediction control.
It converges to the set value. On the other hand, y _Si is k since only values for each step not subject to evaluation, the effect that suppresses this if the control deviation at the time decimated such increase is not obtained. However, since the change in y _Si is slow, the difference between the control deviation at the thinned time and the control deviation at the time incorporated in the evaluation function for each k steps is small. Therefore, k
If the control deviation at the time for each step is small, the control deviation at the thinned time is also small. For this reason, the control deviation can be evaluated with respect to the control amount by representing the value of the time at every k steps. Thus, the responsiveness of the variable can be adjusted in the same manner as the other control amounts. on the other hand,
As described above, since the number of variables to be incorporated in the evaluation function is reduced by thinning, the scale of the optimization calculation is reduced,
The calculation load for realizing control is reduced.

[0011]

FIG. 1 shows an embodiment of the present invention. The operation amounts of the control target 1 are u ₁ and u ₂ , and the control amounts are y _F and y
_S. Although the y _F fast response is required, y
_S only needs to converge to the set value with a gentle response, and suddenly changing this will result in an excessive operation,
This is undesirable for economic or safety reasons.
It is also assumed that y _S itself has only slow dynamic characteristics.

For such a controlled object 1, control amount predicting means 2, operation amount change amount optimizing means 3, and operation amount calculating means 4
Is implemented on the control device to constitute a model predictive control system.
The control amount predicting means 2, y _F, past values and the y _S, the operation amount of the variation Δu _1, y _F based on historical values of Delta] u _2, y
Predicted value of _S default content _{y F * ', y S *} ' is calculated. This calculation is performed based on the second term of Expression 3.

[0013]

(Equation 3) Here, y _F ^* is a vector configured by arranging the current time value and the past value of y _F necessary for predicting the future value, and y _S ^* is the current time value and the past value of y _S required for predicting the future value. The vectors, Δu ₁ ^* and Δu ₂ ^* , arranged by arranging the values are the change amounts of u ₁ and u ₂ required for predicting the future value (difference between the value of the relevant time and the value of the previous time).
Of vectors constituting side by side past values, y _F 'is a vector which constitutes side by side a prediction value of the prediction interval in y _F, y _S' is a vector constructed by arranging a prediction value of the prediction interval in y _S, Delta] u
₁ ′ and Δu ₂ ′ are vectors configured by arranging future (including the current time) values in the operation section of the change amounts of u ₁ and u ₂ .

The dimension of y _F ′ is the number of sampling times in the prediction section, but since y _S ′ is thinned out, its dimension is about 1 / k of that when no thinning is performed.
G ₁ , G ₂ , F ₁ , and F ₂ are constant matrices determined based on the model of the control target 1. G ₁ and F ₁ are obtained by the same method as the conventional model predictive control, and G ₂ and F ₂ are the conventional ones. It is obtained by extracting rows every k rows from a matrix obtained in the same manner as in model predictive control. Therefore, y _S in the control amount prediction means 2
Calculation amount for obtaining a predicted value default content y _{S *} 'is about 1 / k of the conventional model predictive control.

The manipulated variable change amount optimizing means 3 substitutes the predetermined values of the predicted values of y _F and y _S obtained above into the evaluation function of Equation 4, and applies an algorithm of the least squares method to change the manipulated variable. The future value of the quantity ΔU ′ = [Δu ₁ ′ ^T Δu ₂ ′ ^T ] ^T is obtained.

[0016]

(Equation 4) Here, r _F 'is a set value future value in the prediction interval for y _F y _F' constituting side by side in correspondence with the vector,
r _S 'is a set value future value in the prediction interval for y _S y _S' vectors constituting side by side so as to correspond to, y F _* '
y _{S *} ′ is a vector configured by arranging predetermined values of predicted values in the prediction section obtained by the control amount predicting means 2 by the second term of Expression 3;
It is.

Since the dimension in the parentheses of the second term in Equation 4 is about 1 / k of the conventional model predictive control, the scale of calculation by the least squares method is correspondingly reduced. Weights W _Fe , W _Se ,
W _u1 and W _u2 are adjusted by performing a computer simulation. The operation amount calculating means 4 adds the current time in ΔU ′ to u ₁ and u ₂ of the previous step, and calculates u at the current time.
_1, to calculate the u _2. This is given to the control target 1.

The control amount predicting means 2 and the manipulated variable change amount optimizing means 3 may be integrally realized.

[0019]

As described above, by dividing the control amount into a fast change and a slow change, and by thinning out the latter prediction time, it is possible to reduce the scale of the prediction and optimization calculation of the model prediction control. The calculation load on the controller can be reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Control target, 2 ... Control amount prediction means, 3 ... Operation amount change amount optimization means, 4 ... Operation amount calculation means.

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Tomoji Kanno 1 Fujimachi, Hino-shi, Tokyo Fujifacom Control Co., Ltd.

Claims (1)

[Claims] In a model predictive control method for a multivariable control system having a control amount having only a mode slower than the response of another variable, a control deviation of each control amount other than the control amount for each sampling cycle is obtained. The weighted sum of the magnitude of the value, the magnitude of the control deviation future value for each of a plurality of sampling cycles of the control quantity, and the magnitude of the change amount of the manipulated variable after the current time is used as an evaluation function, and the manipulated variable that minimizes this is the Model predictive control method characterized by calculation by optimization calculation based on prediction formula derived from dynamic characteristic model of controlled object