JPH0642162B2 - How to automatically correct a feedforward model - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a method for automatically correcting a feedforward model in response to a change in a disturbance characteristic in a process control device that combines feedback control and feedforward control. It is a thing.

(Technical background of the invention) In process control, feedback control plays an important role, but this control method focuses only on the result, and performs correction control when the result does not match the target value. Therefore, there is no problem in the case of slow fluctuations, but even for sudden disturbances, the result will start to deviate from the target value and correction control will be performed, so the transient response to sudden changes such as load fluctuations and disturbances will be delayed. It has a big drawback. Therefore, the controllability is improved by measuring the disturbance and applying feed-forward control that compensates for the influence of the disturbance in advance before the influence of the disturbance appears in the control amount in combination with the feedback control. . Here, the disturbance refers to an external action that tries to change the state of the control system. In other words, when the control system is in the equilibrium state, that is, when the control deviation en (see FIG. 3) is in a stable state of zero, all of the external factors that change this state and generate the control deviation en Is called disturbance. Disturbances can range from very high to low. For example, it is assumed that a domestic gas water heater controls heating of water to a predetermined temperature. In this case, when the disturbances are listed in descending order of influence, the amount of hot water, the temperature of tap water, the environmental temperature of the gas water heater, the unit calorific value of the gas, the change in the water heater efficiency, and various measurement deviations.

FIG. 1 is a block diagram showing a basic system in which feedback control and feedforward control are combined. In the figure, the comparison unit 1 compares the target value with the control amount, inputs the deviation into the PID adjustment unit 2, performs adjustment calculation, and then guides it to the addition unit 3. On the other hand, the disturbance D influences the control X through the disturbance transfer path 7 having the transfer function GD. Therefore, in order to cancel and compensate for this effect, the disturbance compensation signal is added to the output signal of the adjusting section 2 by the adding section 3 via the feedforward model 8 having the transfer function GF, and this is added as the operation signal to be controlled. In addition to the above, the control quantity X is obtained by adding the output of the block 4 having the transfer function GP to be controlled and the influence of the disturbance (this addition is represented as being performed by the adding section 5).

Therefore, assuming that the output signal of the controller 2 is Y, X = (Y + D * GF) * GF + D * GD = Y * GP + D * (GD + GF * GP) ...
(1) From equation (1), no matter how much the disturbance D changes, the control amount X
GD + GF × GP = 0 must be satisfied in order not to affect Therefore, the transfer function GF of the feedforward model 8 is Need to be defined. GD and GP can generally be approximated by a composite of dead time and first-order lag.

Here, KP and KD are gain constants, TP and TD are time constants, and LP and LD are dead times. Therefore, the transfer function GF of the feedforward model 8 is Becomes In addition, when GP and GD have equal time, And becomes a simple formula. In practice, this simple formula is sufficient. As described above, conventionally, the coefficient KD of the transfer function GD for the disturbance of the controlled object is assumed to be constant, and the coefficient KF = KD / KP of the feedforward model is also assumed to be constant. However, in the actual process, this disturbance coefficient KD is not a constant value, and it is an indirect disturbance factor, a change in characteristics over time, a change in physical quantity inside and outside, a change in the amount of chemical components, a change in ambient temperature, and a disturbance element not measured. Due to the effect of, it changes greatly irregularly. Therefore, there is a problem that the effect of the feedforward control is lost, or conversely, it causes harm.

Especially recently, due to the diversification of raw materials and fuels used in processes, diversification of products, load changes accompanying changes in operating rates due to changes in the economic environment, and versatility, the flexibility of processes, and therefore the flexibility of compulsory devices, has been increased. There is a strong demand for higher abilities.

Hereinafter, the above points will be described in more detail by taking the heat conversion outlet temperature control as an example. In FIG. 2, the raw material 11 is led to the heat exchanger 14 through the pipe 12, where it is heated by steam and taken out. The outlet temperature T ₀ of the heat exchanger 14 is detected by the temperature detector 15 and put in the temperature controller 19, and adjustment calculation is performed so that the outlet temperature T ₀ becomes a predetermined value. on the other hand,
The raw material flow rate Fi is detected by the raw material flow rate detector 13, this is put into the feedforward model 21, the output thereof is added to the output of the controller 19 by the addition section 20, and the addition result is set in the steam flow controller 22. Given as. In the steam flow rate controller 22, the steam flow rate signal detected by the steam flow rate detector 17 is used as a feedback signal to perform adjustment calculation so as to be equal to the steam set value, and the output signal thereof is used to adjust the control valve 18
Control the opening of. In this way, the control for keeping the outlet temperature T ₀ of the heat exchanger 14 constant is performed.

The transfer function G of the feedforward model 22 in the above example
Consider F. First, the heat balance Q in the steady state of the process is obtained.

Where Fs is the steam weight flow rate, Hs is the latent heat of steam, Fi is the raw material weight, Gi is the specific heat of the raw material, Ta is the heat exchanger outlet temperature setting temperature, Ti is the heat exchanger inlet temperature of the raw material, and η is the heat exchanger efficiency. Is.

When the steam flow rate Fs, which is the manipulated variable, is calculated from the equation (5), Becomes From equation (6), the static characteristic compensation component GFS of the transfer function of the feedforward model 22 is And the transfer function GF of the feedforward model 22 loaded with the dynamic characteristic compensation is Becomes Here, TD is a time constant from the raw material flow rate detector 13 to the outlet temperature detector 15, and TP is a time constant from the setting of the vapor flow rate (output of the adding section 20) until this effect reaches the heat exchanger outlet temperature. .

In the above example, the feedforward control is performed with only the change in the raw material flow rate Fi as the disturbance. The coefficient KF is a constant, but in reality, the following factors are used: (i) change in raw material temperature (ii) change in heat exchange efficiency (iii) change in heat quantity of steam held (iv) change in ambient temperature (V) KF changes irregularly and largely due to changes in the specific heat of the raw material. Therefore, the effect of the feedforward control is not sufficiently exerted, the controllability at the time of changing the raw material flow rate is deteriorated, and the product quality is disturbed.

As described above, the conventional control method has a problem that the effect of the feedforward control is not exerted and, in some cases, it is an adverse effect, and the problem becomes more serious as the demand for process flexibility increases. It was supposed to be.

[Object of the Invention]

An object of the present invention is to provide an automatic correction method of a feedforward model, which can always maintain the feedforward compensation in an optimum state and improve the quality of process control.

[Outline of Invention]

The automatic correction method of the feedforward model of the present invention is
Computation for calculating the static characteristic compensation component from the disturbance to cancel the influence of the disturbance in the process control device, which is formed by adding a feedforward control mechanism for suppressing the influence of the disturbance in advance to the feedback control system. An automatic correction method for a static characteristic feedforward model, comprising: a section and a multiplication section for multiplying an output from the calculation section by a coefficient to generate a static characteristic compensation signal, wherein at least one disturbance element is detected. A signal based on the static characteristic compensation signal calculated by this model is added to the adjusted output signal of the feedback control system in the first adder, and the signal based on the addition result and the static characteristic compensation are given to the static characteristic feedforward model. The dynamic characteristic compensation signal obtained from the signal is added by the second adding unit, and the signal based on the signal obtained by the second adding unit is added to the control target, and this control target is added. Is used as a feedback signal in the feedback control system, and a difference signal between the signal based on the output signal from the first addition unit and the static characteristic compensation signal, which is output from the subtraction unit, The coefficient of the multiplication unit in the static characteristic compensation feedforward model is corrected so that the difference signal becomes substantially zero by applying to the feedforward model.

Example of Invention

An embodiment of the present invention will be described below with reference to FIG. In this embodiment, the adjustment output signal of the feedback control system is of the position type.

The process variable PV obtained by measuring the controlled variable X is compared with the target value SV by the comparison unit 31, and the deviation en is adjusted by the adjustment unit 32.
Lead to. The adjusting unit 32 controls P (proportional), I (integral), D
An adjustment operation of any of (differential) operations or a combination is performed. Output signal 32a (first) adder 3 of adjuster 32
It is applied to the controlled object 36 as the operation signal M via the 3, (second) adder 35. In this way, a feedback control system that feeds back and controls the control amount X is configured.

On the other hand, for feed-forward control, the disturbance signal D is input to the calculation unit 39, where the calculation for obtaining the static characteristic compensation amount is performed. The output of the calculation unit 39 is sent to the multiplication unit 40, multiplied by the coefficient in the multiplication unit 40, and the static characteristic compensation signal Dn is obtained.
Becomes In this way, the calculation unit 39 and the multiplication unit 40 form the static characteristic compensation feedforward model 38.

That is, if the formula (4) is transformed, Of these, the first term on the right side is the static characteristic compensation component, and the static characteristic compensation feedforward model 38 obtains this. On the other hand, the second term on the right side is the dynamic characteristic compensation component, It is multiplied by. Therefore, the dynamic characteristic compensation component is obtained by inputting the output of the static characteristic compensation feedforward 38 to the incomplete differentiator 42, which will be described later, having the transfer function represented by the equation (10).

The static characteristic compensation signal Dn is added to the adjusted output signal 32a in the adder 33 to become the signal A. Further, the signal is added to the signal A by the adder 35 via the incomplete differentiator 42.

On the other hand, the subtraction unit 43 obtains the difference between the signal A from the addition unit 33 and the static characteristic compensation component Dn, the integration unit 44 integrates this difference, and the integrated value 44 a is added to the multiplication unit 40. Then, the multiplication unit 4 is set so that the output of the subtraction unit 43 becomes zero.
The coefficient of 0 is adjusted by the output from the integrator 44.
That is, when the disturbance outside the detection target changes, the process variable PV
Changes. Along with this, the deviation en changes, and the change appears as a change in the signal A. As a result, the output from the subtraction unit 43 changes, and the output is added to the multiplication unit 40 via the integration unit 44. In the multiplication unit 40, the coefficient is changed by the output from the integration unit 44 so that the output from the subtraction unit 43 becomes zero. In this way, the feedforward model 38 is corrected so that the difference between the signal A from the adder 33 and the corrected static characteristic compensation amount Dn has a temporal integrated value or an average value of zero.

In terms of conversion, as described above, for example, as shown in FIG. 6, it is described that the difference between the operation signal MV and the feedforward model output signal FF is controlled to be substantially zero. Will be described more appropriately with reference to FIGS. 6 to 8.

In FIG. 6 as well, as in FIG. 1, the disturbance D is added to the control target through the transfer path having the transfer function GD and affects the measured value PV (control amount X). The influence of this disturbance is removed by the combined control of the FB control and the FF control. That is, FIG. 5 shows the relationship between the FB component, the FF component and the operation signal MV when the FB (feedback) control is combined with the FF (feedforward) control. As can be seen from FIG. 6, the operation signal MV is a combined value of the FF component and the FB component as shown in the following equation.

MV = FF + FB (a) FIG. 7 is a diagram showing a basic concept for explaining how the feedforward model is automatically modified. As can be seen from FIG. 7, the FF component is obtained from some of the added disturbances, and the measured value PV
The FF control is executed prior to appearing in the above, and the FF component cannot be obtained from other things, and the FB control is executed by appearing in the measured value PV. As a result of the one control, the measured value PV is set to the set value SV. That is, there are various disturbances. For example, in the case of the domestic gas water heater described above, there are various types such as the amount of hot water. However, in general, not all disturbances are incorporated to obtain FF control. Primarily for economic reasons, we have measured some of the major disturbances that have a high impact and not others. Information is obtained from the disturbance to be measured, and FF control is executed with the FF component based on the obtained information. On the other hand, even a disturbance outside the measurement target is added to the control target, affects the measurement value PV, and changes the measurement value PV. This measured value PV is fed back and subtracted from the set value SV to generate a deviation. That is, the disturbance outside the measurement target appears as the FB control output as a result. As a result, the operation signal MV becomes FF or F.
It is represented by the form B. In this case, if the gain of the FF control model is corrected so that the FB component becomes almost zero, the operation signal MV has only the FF component and only the control in advance. That is, the adaptive FF / FB control is performed. In the present invention, based on such an idea, the FF gain is automatically corrected so that the FB component becomes almost zero, and the adaptive FF / FB control is executed.

In FIG. 6, the FF control based on the disturbance may be performed without advance / delay in time, in some cases with advance in time, or in some cases with delay in time. That is, as can be seen from the equation (4) described in FIG. 1, the FF control model transfer function GF is the disturbance gain constant K
It is represented by D, controlled object gain KP, controlled object time constant TP, and disturbance time constant TD. As can be seen from the equation (4), when TP = TD, there is no time advance / delay, and when TP> TD, time advance compensation is performed and TP
When <TD, time delay compensation is performed. In reality, there are overwhelmingly many cases of time advance compensation.

Further, as can be seen from FIG. 7, when the FF / FB control combining the FF control and the FB control is ideally performed, MVFF (b) is obtained, that is, the equation (b) is obtained. As can be seen from the equation (a), FB0 (c), that is, the operation signal is almost composed of the FF component, and a slight correction is performed in the FB control component. There is.

The above will be described in detail by taking as an example the case of controlling the heating of water to 40 ° C. in the domestic gas water heater described above.

As described above, when the FF / FB control is ideally performed, the MVFF and FB0 are obtained, but the following two cases can be considered as the ideally performed. That is, generally, there are a plurality of factors that may act as a disturbance. For example, the amount of hot water,
These are the tap water temperature, the environmental temperature of the gas water heater, the unit calorific value of the gas, and the water heater efficiency. In the first ideal case,
A case will be described in which only one of these factors acts as a so-called disturbance, and the other factors are constant and do not act as a disturbance. For example, it is assumed that only the amount of hot water acts as a disturbance, and the tap water temperature, the environmental temperature of the gas water heater, the unit calorific value of the gas, the water heater efficiency, etc. are constant. In this case, measure only the amount of hot water,
Calculate how much gas flow is needed to bring the tap water to 40 ° C. If feedforward control is performed based on the calculation, the hot water temperature will be 40 ° C. At this time, the control deviation E becomes zero, and correction by feedback control becomes unnecessary,
After all, MV = FF. This is the first ideal case.

On the other hand, the second ideal case will be described in which all of the above factors act as disturbances. That is, all the factors such as the amount of hot water, the temperature of tap water, the environmental temperature of the gas water heater, the unit calorific value of the gas and the water heater efficiency are accurately measured, and the FF control amount is calculated based on the measurement result. Take control. As a result, the hot water temperature reaches the set value of 40 ° C., no matter which of the disturbances changes. Even in this case, the control deviation E becomes zero, correction by feedback control becomes unnecessary, FB = 0, and eventually MV = FF. This is the second ideal case.

However, as in the first and second ideal cases above,
It is extremely difficult to perform the FF / FB control sufficiently ideally due to various reasons such as economical efficiency and accuracy. That is, FF / FB control is not ideally performed in reality, and FF control is needed. That is, in reality, the FF control model is deviated, the equation (b) is not established, and the FB control component is generated. When the FB control component becomes large, the effect of FF control is attenuated. In order to prevent this, the FB control gain is corrected so that it becomes FB0.

FIG. 8 shows how the influence control characteristic of disturbance changes depending on the composition ratio of the FF component and the FB component in the operation signal MV. It can be seen that when FB = 100% and FF = 0%, that is, when only FB control is performed, the influence of the disturbance is large, but as the FF component approaches 100%, the influence of the disturbance decreases as the FF component approaches.

Therefore, in order to satisfy the equation (b), that is, the operation signal M
If the gain of the feedforward control model is modified so that the difference between the V and FF components becomes substantially zero, the FF / FB control will always be in the optimum state.

If the speed type PID calculation method is adopted,
Since it is easier to use FB0 in the equation (c), this is used.

By doing so, as described above, even if the characteristic of the disturbance process changes economically, the optimum FF control effect can always be obtained.

The reason why the set value (SV) and the measured value (PV) match with each other as described above is as follows. As shown in FIG. 5, the FB control is a closed loop control that corrects the result of control. On the other hand, the FF control only predicts and outputs how much the operation signal should be changed in order to suppress the influence of the disturbance D according to the magnitude of the disturbance D, and outputs the open signal. -It is a loop control. Therefore, as described above, in the FF control, the gain of the FF control model is modified to only slightly change the relationship of the FF control output with respect to the disturbance D, and the original set value SV and the measured value PV are matched. There is no intervention in functional FB control. Therefore, the set value SV and the measured value PV are F
The effect of the B control results in agreement.

In the above embodiment, the adjustment output signal 32a is of the position type, but when it is of the velocity type, a part of the system shown in FIG.
The same correction can be made by modifying as shown in the figure. That is, in the difference portion 51 that constitutes the position-type velocity system signal conversion unit, the difference between the static characteristic compensation amount Dn at each sampling and the static characteristic compensation amount Dn-1 at the previous sampling is obtained, that is, Then, the static characteristic compensation component is converted into a velocity type signal, which is then added to the adjusted output signal 32a by the adding section 33, and then the velocity type / position type signal converting section 5 is added.
In step 2, the signal is returned to the position type signal to form the signal A, which is added to the adding section 35 and is also given to the subtracting section 43. The other points are similar to those of the embodiment shown in FIG.

Further, in the embodiment shown in FIGS. 3 and 4, the correction is made so that the integral value of the difference from the compensation amount of the static characteristic of the signal A becomes zero, but as shown in FIG. May be integrated by the integrating section 44, the component value thereof may be added to the multiplying section 40, and correction may be performed so that the integrated value becomes zero. this is,
This is because, if the feedforward control is optimally performed, the adjustment output signal becomes substantially zero with only minor correction.

In each of the above embodiments, each component of the control system may be of an analog type, a digital type, may be formed of individual parts, or may be formed of a programmed computer. .

Further, in each of the above embodiments, not only the static characteristic compensation but also the dynamic characteristic compensation is performed by the feedforward control, but the anti-invention can be applied to the case where only the static characteristic compensation is performed. In this case, the incomplete differentiator 42, the adder 35, etc. in FIGS. 3 and 6 may be removed.

Further, the bias may be modified instead of modifying the coefficient of the feedforward model.

〔The invention's effect〕

According to the present invention, a feedforward control mechanism is added to the feedback control system in the process control device, and the characteristics of the static characteristic compensation feedforward model in the feedforward control mechanism are corrected by adjusting the feedback control system output signal and the model. The sum signal with the signal based on the output signal from, and the output signal from the model,
Therefore, the influence of the disturbance can be surely controlled in advance and the quick control can be performed without waiting for the feedback control in the feedback control system.

[Brief description of drawings]

FIG. 1 is a block diagram showing a conventional control device, FIG. 2 is a piping diagram showing a heat exchanger equipped with the conventional control device, and FIG. 3 is a control device used for carrying out a correction method according to the present invention. 4 is a block diagram showing a control device used to implement another embodiment of the present invention, and FIGS. 6 to 8 are block diagrams and lines for explaining the principle of the present invention. It is a figure. 32 ... Adjusting unit, 33, 35 ... Addition unit, 36 ... Control object, 38 ... Feedforward model, 39 ... Calculation unit, 40 ... Multiplication unit, 42 ... Incomplete differentiation unit, 43 ...
... subtraction part, 44 ... integration part, 51 ... difference part, 52 ...
Velocity-to-position signal converter.

Claims (3)

[Claims] 1. A static characteristic compensation from a disturbance for canceling the influence of the disturbance in a process control device, wherein a feedforward control mechanism for preventing the influence of the disturbance in advance is added to a feedback control system. An automatic correction method for a static characteristic feedforward model including a calculation section for calculating a minute and a multiplication section for multiplying an output from the calculation section by a coefficient to generate a static characteristic compensation signal, and reducing an element that causes disturbance. One of them is detected and given to the static characteristic feedforward model, and a signal based on the static characteristic compensation signal calculated by this model is added to the adjustment output signal of the feedback control system in the first adding unit, and a signal based on the addition result. And a dynamic characteristic compensation signal obtained from the static characteristic compensation signal are added by a second adding section, and a signal based on the signal obtained by the second adding section is added to a control target. The control amount output from the controlled object is used as a feedback signal in the feedback control system, and the difference between the signal based on the output signal from the first addition unit and the static characteristic compensation signal output from the subtraction unit. A signal is applied to the feedforward model, and the coefficient of the multiplication unit in the static characteristic compensation feedforward model is modified so that the difference signal becomes substantially zero. Automatic correction method. 2. The method according to claim 1, wherein when the adjustment output signal of the feedback control system is a position type, the static characteristic compensation signal is directly added to the first adding section, and a feedback control system is provided. When the adjustment output signal of is a velocity type signal, the velocity type signal obtained by converting the static characteristic compensation signal to the velocity type signal is added to the first adding unit, and the output signal from the first adding unit is added to the position type signal. The method of adding the position-type signal converted to the above to the second adder and the subtractor. 3. The method according to claim 1, wherein the dynamic characteristic compensation signal is obtained by incompletely differentiating an output of the static characteristic compensation feedforward model.