JPH0415706A - Model estimation controller - Google Patents

Abstract

PURPOSE:To eliminate the interruption of the estimation control by reducing gradually the control request conditions until a second optimum manipulated variable when the solution of an optimum manipulated variable that satisfies temporarily the limit conditions can not be obtained. CONSTITUTION:A future value estimating means 1 obtains the estimation formulas for the future values of a controlled variable (y) and a manipulated variable (u) based on a model obtained by approximating the dynamic characteristic of a control subject and sets the limit conditions related to the future values of both variables (y) and (u) together with the evaluation functions of the secondary forms related to the future values of the variable (y) estimated by the estimation formula, the deviation against the target value (r), and the future value of the variable (u) respectively. Then the means 1 calculates the future value of the variable (u) by a secondary plan method in order to satisfy those set limit conditions and also minimizes the evaluation function. If the evaluation function includes no enable solution, the conditions concerning the control request are least changed for the time being only at that time point in order to obtain a second optimum solution. Therefore the conditions are gradually reduced until the second optimum solution (manipulate valiable) is obtained in regard of such control requests for the limit conditions, the evaluation function, the target value, etc. Thus the estimation control is never interrupted.

Description

[Detailed Description of the Invention] [Object of the Invention] (Industrial Application Field) The present invention relates to a model predictive control device for optimally controlling a controlled object using a model that approximates the controlled object, and particularly relates to a model predictive control device for optimally controlling a controlled object using a model that approximates the controlled object. The present invention relates to a model predictive control system suitable for optimally controlling and operating a plant that is subject to a large number of constraints regarding conditions.

(Prior art) In recent years, in order to achieve optimal control and operation while satisfying the many constraints imposed on plants, a linear discrete-time model is constructed based on the impulse response or step response of the plant, and from this model Model predictive control is often used to sequentially calculate an optimal manipulated variable that minimizes a quadratic evaluation function regarding the deviation of the future value of the controlled variable from the target value and the future value of the manipulated variable from the derived predictive equation.

That is, it is intended to determine the amount of operation to be applied at the present time so as to obtain a control future value that moves as close to the target value as possible under a constant amount of operation.

The prediction formula for determining future values of controlled variables and manipulated variables is expressed as a function related to past controlled variables and manipulated variables, so past controlled variables and manipulated variables are A prediction is made each time based on the , and the amount of operation to be applied at that time is determined.

To find the manipulated variable that satisfies the constraints on the controlled and manipulated variables, use quadratic programming (QP) to find a solution that minimizes the evaluation function.
).

(Problem to be solved by the invention) However, if there are too many constraints imposed on the plant, if the time required to obtain a prediction formula is too long, or if an attempt is made to make the controlled variable follow an impossible target trajectory, the controlled variable , there is no solution to the optimal manipulated variable such that all predicted values of the manipulated variable simultaneously satisfy the constraint conditions, and the manipulated variable at that point in time cannot be determined, making it impossible to perform predictive control.

The present invention was made in order to solve the above problems, and its purpose is to change the parameters of the evaluation function, constraints, target value trajectory, etc. to a minimum even when there is no solution such that all predicted values satisfy the constraints. It is an object of the present invention to provide a model predictive control device capable of determining a quasi-optimal operation amount at that point in time and continuing predictive control.

[Structure of the Invention] (Means for Solving the Problem) In order to achieve the above object, the invention of claim (1) uses a model to predict the future value of the control amount, and predicts the future value of the control amount while satisfying the conditions regarding the control request. In addition to calculating the optimal manipulated variable that minimizes the evaluation function, if there is no possible solution for the manipulated variable that satisfies the conditions related to the control request, the conditions related to the control request are temporarily minimized only at that point in order to obtain a semi-optimal solution. The present invention is characterized in that a future value prediction means is provided with a condition processing means for changing.

(Operation) According to the invention of claim (1) having the above configuration, the future value prediction means uses a model that approximates the dynamic characteristics of the controlled object to
I! A prediction formula for the future value of the I quantity and the manipulated variable is determined, a constraint condition regarding the future value of the controlled variable and the manipulated variable is set, and the deviation between the future value of the control variable predicted by the prediction formula and the target value is calculated. A quadratic evaluation function regarding the future value of the manipulated variable is set, and a future value of the manipulated variable that satisfies the set constraints and minimizes the evaluation function is calculated using quadratic programming.

If there is no possible solution to this evaluation function, the conditions related to the control request are temporarily changed to the minimum level only at that point in time in order to obtain a quasi-optimal solution.

Therefore, conditions related to control requests such as constraints, evaluation functions, and target values are gradually relaxed until a quasi-optimal solution (quasi-optimal manipulated variable) is obtained.

Thereby, predictive control can be continued without being interrupted.

(Example) Next, an example of the model predictive control device according to the present invention will be described. FIG. 1 is a block diagram showing a process control system 3 to which the model predictive control device 1 is applied, and FIG. 2 is a block diagram showing the configuration of the model predictive control device 1. As shown in FIG.

The model predictive control device 1 inputs the target value r and the control amount y of the plant 5, which is the object to be controlled, and outputs the optimal operation amount U. In plant 5, which produces product p from raw material S, when the production target amount of product p (target value) changes, raw material S
Consider the case where the production amount y (controlled amount) of product p is made to follow the target value by manipulating the input amount U (operated amount).

As shown in FIG. 2, the model predictive control device 1 includes:
A human power device 7 for inputting constraint conditions regarding future values of controlled variables and manipulated variables of the plant 5, parameters of evaluation functions, and controlled variable target values, and a sensor 9 for observing actual controlled variables of the plant 5 are provided. There is.

Further, the model predictive control device 1 is provided with a model 11 that approximates the dynamic characteristics of the plant 5 and a future value prediction means 2.

The future value prediction means 2 includes a prediction means 13 that uses a model 11 to obtain a prediction formula for the future value of the controlled variable, and a prediction means 13 that uses a model 11 to obtain a prediction formula for the future value of the controlled variable, and a prediction means 13 that uses a model 11 to transform the constraint conditions manually input from the human power device 7 to generate a prediction formula for quadratic programming (QP). Constraint condition setting means 15 for setting constraint conditions, and evaluation function setting means 17 for setting a quadratic-form evaluation function regarding the deviation of the controlled variable future value from the target value and the manipulated variable future value are provided.

Further, the future value prediction means 2 includes a constraint setting means 15.
The evaluation function setting means 17 satisfies the constraints set by
Optimal operation amount calculation means 19 that sequentially calculates the future value of the operation amount that minimizes the evaluation function set by using quadratic programming; condition processing means 21 for temporarily changing the conditions related to the control request only at that point in time in order to obtain a semi-optimal operation amount;
and are provided.

Further, the model predictive control device 1 is provided with an actuator 23 that applies the determined optimal operation amount to the plant 5.

In order to determine the manipulated variable uk at the current moment k (the 1st stage), the manipulated variable change rate ΔU may be determined as follows. First, from the relationship between the manipulated variable U and the controlled variable y, the future series of the controlled variable and the rate of change of the manipulated variable over several steps in the future V-[Vm+L・'°°・y*
+L...o-1 co゛(1) Δu-[Δ' k l ”'
A prediction formula of I Δu −+−−−+] −(2) is determined. Here, np and n are time lengths (extra measurement length, control length) for predicting future values of control quantities and manipulated variables. Next, k + L from +L at the time extracted by the human-powered device 1
+ n p The target value r up to 1 is the target value future value series 3” ” [Y” o + ...+ Y” IIp
-1] ...Set as (3). Deviation between the predicted control value and the future target value (y” +3' k+L
The rate of change ΔU from the manipulated variable u k− at the previous point in time is determined so that the predicted manipulated variable change value Δu k+1 (h+) is as small as possible.

This calculation is performed according to the flowchart shown in FIG.

As shown in FIG. 3, in step 101, the control amount y,
is read in, and in step 103, the prediction means 13 uses the model 11 to find a prediction formula for the controlled variable and the rate of change in the manipulated variable. In step 105, it is determined whether there are any constraint conditions. If there are constraint conditions, the constraint conditions are set in step 106.

Next, in step 107, the evaluation function setting means 17 sets the evaluation function J shown below.

J ″Σ (y” + y −・L・1)
+λΣ (Δu,,,) (λ is the weighting coefficient)...
(4) If there are no constraint conditions in step 105, express the controlled variable y by the formula of the predicted manipulated variable change rate ΔU, and then proceed to step 105.
9, optimal operation amount calculation means 19, δJ/δΔu-0...
By solving (5), the optimal operation amount change rate ΔU that minimizes J can be obtained, and Δu is determined.

However, due to equipment reasons, the manipulated variable U, the controlled variable y,
Alternatively, these rates of change Δu1Δy are often provided with upper and lower limits, such as a law.

u 1゜ ≦ U ≦ U □8 Δ umlm ≦ Δ U ≦ Δ U □8Y m
l a ≦ y ≦ yllat
... (6)Δyl. ≦Δy≦
Δy□8 In order to determine the manipulated variable change rate ΔU so as to minimize the evaluation function J between the upper and lower limits, quadratic programming (QP) is used in step 1°7. By the constraint setting means 15,
Using model 11, u, y, Δy1 are expressed by the formula ΔU, and the constraint condition is b ffi,, ≦A−ΔU≦b, , −(
Rewrite them in the form of 7).

In step 109, the optimum operation amount calculation means 19 calculates
Solve this using quadratic programming to find the optimal manipulated variable change rate Δ that minimizes the evaluation function J while satisfying all constraints.
You can get U.

In step 111, it is determined whether an optimal solution exists for the evaluation function J ((formula 4)). If the optimal solution does not exist, conditional processing is executed in step 113.

In this condition processing, the range between the upper and lower limits is small,
Alternatively, if the varata n in the evaluation function (4) is large (that is, the predicted length and control length are long), there is no possible solution, that is, ΔU that satisfies all the constraints, and the quadratic Sometimes a solution cannot be obtained using the planning method. If you can't decide on the current value, you won't be able to decide on the future value. Therefore, predictive control is interrupted. In order to avoid such a situation, in this embodiment, when it is determined in step 113 that the optimal operation amount calculating means 19 cannot obtain a solution to the evaluation function, the following condition processing is performed in step 113.

(1) Reduce the prediction length by one step (n, by 1) and re-solve the quadratic programming. This is because the last element of the predicted value of a series of control variables for a series of manipulated variables predicted in the current situation is in the distant future and has low priority for constraints and evaluation functions. This is based on the reason that it does not matter if the value deviates from the target value.

(2) Resolve the quadratic programming method by changing the weighting coefficient λ regarding the manipulated variable of the evaluation function J. This is to change the degree to which the stationarity of ΔU is evaluated.

For example, reducing λ means allowing greater variation in ΔU.

(3) Relax the upper and lower limits (thresholds) of the controlled variable and solve the quadratic programming again.

(4) Change the target value of the controlled variable and solve the quadratic programming again. This is done by making the change in the target value of the controlled variable gradual, alleviating unreasonable control demands, and allowing a quasi-optimal solution to be obtained by quadratic programming.

For example, in the model predictive control device 1 of this embodiment, yl is newly set using a target value filter as shown in the following equation.

1+σS+α2 (σS)2 +α3 (σ5)3(α
1 σ is a parameter) ...(8) The time constant of this filter increases as the value of the parameter σ increases, and the target value r applied to the filter changes more slowly than before, as shown in Figure 4. become.

One or more of the methods (1) to (4) are repeated until a solution is obtained (step 113). Of the semi-optimal operation amount ΔU calculated in this way, the first element Δu is used to calculate the operation amount u * - u to H at that point.
+Δuk is determined, and the manipulated variable at the next point in time is newly calculated using the conditions before the change.

As described above, even if a solution to the evaluation function that satisfies the temporal constraints cannot be obtained, predictive control can be performed without interruption.

[Effects of the Invention] As explained above, in the model predictive control device according to the present invention, - Even if a solution to the optimal manipulated variable that satisfies the temporal constraints cannot be obtained, sequential control is performed until a quasi-optimal manipulated variable is found. Since there is a means for relaxing the requirements, a sub-optimal operator (sub-optimal solution) can be obtained in any case, and predictive control can be performed without interruption, which is an excellent effect.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing a process control system to which the model predictive control device according to the present invention is applied, Fig. 2 is a block diagram showing the configuration of an embodiment of the future value prediction means, and Fig. 3 is the model predictive control device. FIG. 4 is an explanatory diagram of a target value filter that moderates the controlled variable target value. 1...Model predictive control device 11...Model 13...Prediction means]5...Constraint condition setting means 17...Evaluation function setting means 19...Optimum operation amount calculation means 21...Conditions processing means

Claims (1)

[Claims] A model predictive control device that optimally controls a controlled object using a model that approximates the dynamic characteristics of the controlled object, which uses the model to predict the future value of a controlled variable to satisfy conditions regarding control requests. While calculating the optimal manipulated variable that minimizes the evaluation function, if there is no possible solution for the manipulated variable that satisfies the conditions related to the control request, the conditions related to the control request are temporarily determined only at that point in order to obtain a semi-optimal solution. 1. A model predictive control device comprising future value predicting means equipped with condition processing means for minimally changing conditions.