CN103913989A - Optimization device and method thereof, control device and method thereof - Google Patents

Abstract

The invention provides an optimization device and method thereof, a control device and a method thereof, capable of optimized control even for integral control subjects. For integral control variables, a predict vale before preset transition state is used by a first predicting portion 121. The first predicting portion 121 predicts the value after the transition state from the output control variables containing integral factors in optimized subject variables becoming optimized subjects in the data collected by a data collecting portion 101. For the unused control variables and operation variables in the first predicting portion 121, the predict value before preset time is obtained in a second predicting portion 122 which predicts value after preset time from the optimized subjects which do not contain control variables output by integral factors.

Description

Optimization apparatus and method and control device and method

Technical field

The present invention relates to a kind of industrial process to the petroleum refinement process, the petrochemistry process etc. that become control object and carry out optimized optimization apparatus and method and control device and method.

Background technology

As the control method of the industrial processs such as petroleum refinement process, petrochemistry process, that knows has a Model Predictive Control.Model Predictive Control is originally in the process that becomes multiple input/output system, keep in the face of being equivalent to the performance variable of input of process and the condition that is equivalent to the control variable defined of the output of process as one, one side grows up the control method that these values tend towards stability towards desired value.In addition, now, the desired value of the steady state to process, determines (referring to patent documentation 1, non-patent literature 1～4) by optimization methods such as linear programming technique (Linear Programming, following brief note are LP) and quadratic programmings (Quadratic Programming, following brief note are QP).

Example to this Model Predictive Control illustrates simply.The system of as shown in Figure 9, carrying out Model Predictive Control comprises steady state optimization portion 501 and carries out the control part 502 of Studies of Multi-variable Model Predictive Control computing.Steady state optimization portion 501 is by input optimization evaluation function, bound conditional value, performance variable, control variable etc., and utilizes process 503 that the optimization method such as LP and QP the calculates control object optimal objective value under steady state.Control part 502 is inputted optimal objective value, control variable and upper lower limit value etc., considers the condition (upper lower limit value) providing while carries out control algorithm, so that the performance variable of process 503 and control variable converge on optimal objective value.For example, control part 502 is controlled, so that performance variable and control variable do not exceed the compass by upper lower limit value defined.In addition,, about the concrete calculating of Model Predictive Control, owing to having detailed description in non-patent literature 4 grades, therefore, in this description will be omitted.

Then, to utilizing LP and QP to determine that the method for desired value describes.In the optimization of actual process, sometimes the difference of the steady state from current is carried out to optimization (hereinafter referred to as difference type optimization).For example, in patent documentation 1, disclose the optimization method about the steady state of the system taking Model Predictive Control as object, but carried out the optimization computing of difference type at this.Again, in non-patent literature 2, record by LP the optimization of process is carried out to definite method, but also used the optimization of difference type at this.

Below, suppose u ₁, u ₂, u _mrepresent performance variable, y ₁, y ₂, y _nrepresent the value of control variable.The number of performance variable is that the number of m, control variable is n.Establishing again k is the index that represents current control cycle.When using synthetic to control variable and performance variable when vector is processed, can represent with following formula (1).And, when control variable and performance variable are a vector while processing, can be as shown in the formula representing with x shown in (2).Again, the optimization problem of difference type can suc as formula (3) illustrated describe.

[formula 1]

u=[u ₁,u ₂,...,u _m] ^T,y=[y ₁,y ₂,...,y _n] ^T...(1)

x=[u ^T，y ^T] ^T...(2)

$u(k-1)={[u_1(k-1),u_2(k-1),...,u_m(k-1)]}^T,y(\∞)={[y_1(\∞),...,y_n(\∞)]}^T$

Δu=[Δu ₁，Δu ₂，...，Δu _m] ^T，Δy=[Δy ₁，Δy ₂，...，Δy _n] ^T

$\mathrm{\Δx}={[\mathrm{\Δ}{u}^T,\mathrm{\Δ}{y}^T]}^T,x_0={[u{(k-1)}^T,y{(\∞)}^T]}^T$

Δx _opt=argminJ(x ₀+Δx)，x _opt=x ₀+Δx _opt

J(x)=x ^THx+c ^Tx

Δy=G ₀Δu

$\underset{\‾}{u_i}-u_i(k-1)\≤\mathrm{\Δ}{u}_i\≤\stackrel{\‾}{u_i}-u_i(k-1),\mathrm{fori}=1,...,m$

$\underset{\‾}{y_i}-y_i(\∞)\≤\mathrm{\Δ}{y}_i\≤\stackrel{\‾}{y_i}-y_i(\∞),\mathrm{fori}=1,...,n...\left(3\right)$

U(k-1) be value, the y(∞ of the performance variable before a control cycle) for by u(k-1) value of control variable convergence while being input to continuously in process.At this, establish continuous u(k-1) and y(∞) group, the steady state of the process in the value situation before a continuous control cycle is optimized initial point x ₀, by the optimum difference value Δ x from initial point _optcarry out optimization as the problem solving.

Δ u is from u(k-1) difference, Δ y for from y(∞) difference.J is that evaluation function, object are that this value is minimized.The non-vanishing words of H are quadratic programmings, are that zero words are linear programming techniquies.The formula of the 6th row of formula (3) means that the variation delta y of control variable will become G if make performance variable only change Δ u from current steady state ₀Δ u.At this, G ₀the transfer function matrix G(S of process (control object)) value when the s=0.In addition " G, ₀=G(0) (4) ".Below, G ₀be called DC current gain matrix.Again, u _i, y _ithe middle variable with underscore represent performance variable and control variable lower limit, bring the variable of line to represent higher limit.All can inequality condition represent.In addition, inequality condition is usually represented, can be write as " A Δ x≤Δ b(5) ".

If solve the problems referred to above and obtain Δ x _opt, by the value obtaining is added to x ₀upper, just can obtain optimal objective value x _opt.By this value being made as to the desired value of Model Predictive Control, just can be by process control at optimum state.

In addition, although the condition of performance variable and control variable is specified to also have method in addition at this by being limited in upper lower limit value.For example, also have the compensation corresponding amount having exceeded from upper lower limit value with control variable is appended to the method above-mentioned evaluation function.The in the situation that of this method, although allow control variable to exceed upper lower limit value, due to the quantitative change exceeding when large evaluation function will become greatly, thereby the amount exceeding will be suppressed.

Again, in this manual, mentioning condition, is to comprise: although do not allow completely depart from upper lower limit value or the condition of setting value and can allow depart from upper lower limit value or setting value by compensation to the condition that exceeds upper lower limit value or setting value and suppress the two.Again, departed from upper lower limit value do not allow or the condition of the type of setting value is called rigid condition, the condition that can allow the type departing from is called to elastic condition.The in the situation that of elastic condition, depart from original upper lower limit value or setting value owing to allowing, so the solution not satisfied condition sometimes.Please note in this manual, also use " satisfying condition " to express although comprise situation as above, what in fact met is the condition being expanded from original upper lower limit value or setting value.

[prior art document]

[patent documentation]

No. 4614536 communique of [patent documentation 1] Jap.P.

[non-patent literature]

[non-patent literature 1] large island is just abundant, " birth of Model Predictive Control-theory launch to develop-", instrumentation and control, the 39th volume, No. 5,321-325 page, 2000.(large Shima is just abundant, " imperial-raw exhibition of Li Theory Birth development exhibition that モ デ Le gives Measuring system-", Meter Measuring と system are imperial, the 39th volume, No. 5,321-325 Page, 2000.）

The clear husband of [non-patent literature 2] Ishikawa, large island are just abundant, paddy wall is prosperously respected, village last week too, " thering is the removal method of the ill-condition in the Model Predictive Control of constant optimization function ", chemical engineering thesis collection, the 24th volume, No. 1,24-29 page, 1998 years.(the clear husband of Ishikawa, large Shima are just abundant, paddy wall is prosperously respected, village last week too, " the Fitnessization Machine of the Chang of Dinging can The hold the imperial In Malignant condition of つ モ デ ル Yu Measuring system remove method ", chemical engineering Theory collected works, the 24th volume, No. 1,24-29 Page, 1998 years.）

[non-patent literature 3] S.Joe Qin, Thomas A.Badgwell, " A survey of industrial model predictive control technology ", Control Engineering Practice, vol.11, pp.733-764,2003.

[non-patent literature 4] Jan M.Maciejowski(foot is vertical repaiies one, wild political affairs are bright translates), " Model Predictive Control-and condition in optimum control-", Tokyo motor university press, 2005.(Jan M.Maciejowski(foot is vertical repair one, wild Zheng Ming Breakdown), " the imperial-”, East Jing Electricity Machine university press offices of the Fitness system of the imperial-Approximately も と で processed of モ デ ル Yu Measuring system, 2005.）

Summary of the invention

The problem that invention will solve

But, in the technology of above-mentioned association, there is the optimization problem of the situation that comprises integral element as the process of control object.

So-called integral element refers to output and the proportional such key element of time integral of inputting.The words that represent by transport function are K/s.As the example of dynamic system (hereinafter referred to as integration class) with integral element, describe with container.If the volume of the fluid in container is y ₁, the flow that flow into container is u ₁, from container flow out flow be u ₂, these relations are represented by following formula (6) so.If using flowing into flow, flow out flow and consider as output as the input of system, volume, this system is integration class.

[formula 2]

$y_1\left(t\right)=\underset{0}{\overset{t}{\∫}}(u_1\left(t^{\′}\right)-u_2\left(t^{\′}\right)){\mathrm{dt}}^{\′}...\left(6\right)$

Integral element is not as long as its input is 0, and output is not just certain.For this reason, the output of integration class is not certain value, changes continuously sometimes.For example, carry out word with the example of said vesse, flow into flow and flow out flow when uneven, the fluid volume in container will continue to change.Flow out if flow into throughput ratio the state continuance that flow is many, the fluid in container just will overflow.On the other hand, if it is many to flow out flow, container will certain time become empty.Certainly, be conventionally necessary to control so that it does not become above-mentioned any state.

In actual process, exist and there is integral element and likely become optimized object.But according to reason described later, above-mentioned technology cannot be applicable to this process at that.

The first is because the variation delta u of performance variable is limited, even in given condition, it is limited that the variation delta y of control variable is also not limited to.In aforementioned, can be by DC current gain matrix G although the description of Δ y ₀calculate with the product of Δ u, but process is while comprising integral element G ₀can not become limited.This is to be K/s, to be become infinity when the s=0 and come by the transport function of integral element.Since G ₀can not become limitedly, Δ y just can not become limited yet.In fact,, if those skilled in the art's words are can easily understand in the example of said vesse, also may there is y ₁can not become limited or overflow or become empty situation.

For the optimization of difference type, cannot calculate y(∞ again) also will become problem.Y(∞) be the input u(k-1 continuing before a control cycle) convergency value of the control variable of steady state in situation., there is y(∞ in above-mentioned optimization method) become prerequisite.But, during for the example of container, if u ₁with u ₂not etc., y ₁(∞) just can not become limited value.For this reason, becoming optimized precondition is false.

According to reason as above, the method that non-patent literature 2 is recorded is due to its distinctive action of process of not considering to comprise integral element, thereby optimization method can not former state be suitable for.The optimization method of the process that for this reason, comprises integral element from common be different.For example, apply in the 3.3.4 item (753 pages) of the non-patent literature 3 of describing in the industry about Model Predictive Control, as the steady state optimization method of process with integral element shown with two kinds of methods.

The first, show the method that the inclination of the control variable that comprises integration class is constrained to 0 such equation restrictive condition (rigid condition) of appending.The second, for example show, by the size of the inclination of the control variable that comprises integration class (tilt square) is joined in optimized evaluation function to carry out by way of compensation the inclination of control variable to provide the method for elastic condition.Like this optimization problem is solved, the inclination of the control variable that comprises integration class will become 0 or very little value, and the optimum solution of performance variable will become the value of the variation that suppresses above-mentioned control variable.

Even if also can carrying out this meaning of optimization from the process of integration class, these methods say excellent.But problem does not solve completely, leaves following problem.

First, cannot carry out energetically optimization to the control variable that comprises integration class.Said method be calculate as the inclination of control variable with integration class be 0 or diminish as much as possible the desired value of performance variable, control variable be object.But, and not shown for making the value of the control variable with integration class itself approach the way of (optimum) value of expectation.

And, the optimization with the related performance variable of control variable with integration class is impacted.For example, consider as shown in figure 10, performance variable is that two (MV1, MV2), control variable are the control object of (CV1).Between from MV1 to CV1, there is integral element, but do not exist between from MV2 to CV1.Be located at while carrying out optimization computing, the inclination of CV1 is 0, and this value is consistent with lower limit.

At this, establish optimized target for MV2 is minimized.Owing to there is long-pending element of taxonomy between from MV1 to CV1, the current inclination of CV1 is 0, adds that when the inclination of CV1 is made as to 0 such restrictive condition and optimization problem is solved, the optimal objective value of MV1 becomes currency.Once this is that CV1 just has inclination owing to departing from currency.

On the other hand, the optimal objective value of MV2 also becomes currency.This is because the value of CV1 is consistent with lower limit, cannot further reduce the cause of the value of MV2.In fact,, if increase and from then on reduce temporary transient the value of MV1, just can realize simultaneously and keep the inclination of CV1 to be 0, the value of CV1 to be remained on to lower limit and the value of MV2 is reduced further compared with currency.But, due in said method, consideration just the inclination of CV1 is remained to 0 or approach 0 value, so cannot make MV1 and MV2 all be changed.Like this, in the technology of above-mentioned association, may occur should be passable the situation that but cannot carry out of optimization.

The present invention makes in order to address the above problem just, even if its object is to have the control object of integration class, also can carry out optimization to the desired value of controlling.

For the means of dealing with problems

The optimization apparatus the present invention relates to, comprising: Data Collection portion, and it collects the data of control object, and the packet of control object is containing the control variable of the performance variable for control object is controlled and control object output, model storage part, its mathematical model to control object is stored, the 1st prediction section, it is just collected in the control variable that comprises integral element output in the middle of the optimization target variable that becomes optimization target in the data of the Data Collection portion value after to the transition state time and predicts, the 2nd prediction section, its with regard to the control variable that does not comprise integral element output in the middle of optimization target variable the value after to the stipulated time predict, inclination prediction section, the variable quantity of its time per unit after to the stipulated time with regard to the control variable that comprises integral element output in the middle of optimization target variable is predicted, steady state variable condition configuration part, its value after to the stipulated time of optimization target variable is carried out condition setting, transition state tilt condition configuration part, its condition of the variable quantity of the time per unit in transition state interval being carried out except 0 with regard to the control variable that comprises integral element output in the middle of optimization target variable is set, transition state variable condition configuration part, it carries out condition setting to the value in the transition state interval of performance variable, and this performance variable is relevant with the control variable that at least comprises integral element output in the middle of optimization target variable, and derivation portion, it is obtained and meets the condition being set by transition state tilt condition configuration part and transition state variable condition configuration part, and given optimized evaluation function is carried out to the optimum solution of optimized optimization target variable, optimization target variable is that the time interval till the near stipulated time is divided into until the more than one transition state of transition state time process is interval and a transition state time warp steady state interval later, and the value of the performance variable in transition state interval and steady state interval and control variable is set as respectively to the variable of optimization target.

In above-mentioned optimization apparatus, also can comprise steady state tilt condition configuration part, the variable quantity of this steady state tilt condition configuration part time per unit after to the transition state time with regard to the control variable that comprises integral element output in the middle of the optimization target variable of setting according to the data that are collected in Data Collection portion carries out condition setting.

In above-mentioned optimization apparatus, also can transition state tilt condition configuration part and steady state tilt condition configuration part, to the control variable that comprises integral element output in the middle of optimization target variable, provide condition according to upper lower limit value, transition state tilt condition configuration part provides the higher limit larger than steady state tilt condition configuration part and less lower limit as condition.

In above-mentioned optimization apparatus, also can transition state tilt condition configuration part and steady state tilt condition configuration part to the control variable that comprises integral element output in the middle of optimization target variable, output compensates accordingly with the size of the variable quantity of time per unit, transition state tilt condition configuration part is for the variable quantity of same time per unit, output is than the little compensation in steady state tilt condition configuration part, derivation portion, carries out optimization to compensation being added in to the function obtaining on evaluation function.

In above-mentioned optimization apparatus, also can, according to evaluation function, the setting of transition state tilt condition configuration part and steady state tilt condition configuration part be adjusted.

In addition, in the control device the present invention relates to, it adopts the desired value of above-mentioned optimization apparatus output, control device comprises: control part, it controls the desired value that performance variable and control variable have been exported towards optimization apparatus, and at least have the control variable of integration class stablized to parameter stabilization time of adjusting to the time till desired value, with stabilization time parameter correlation connection determine the transition state time.

The optimization method the present invention relates to, comprises the steps: data collection step, and it collects the data of control object, and the packet of control object is containing the control variable of the performance variable for control object is controlled and control object output, model storing step, its mathematical model to control object is stored, the 1st prediction steps, its just by the control variable that comprises integral element output in the middle of the optimization target variable that becomes optimized object in the collected data of data collection step the value after to the transition state time predict, the 2nd prediction steps, its with regard to the control variable that does not comprise integral element output in the middle of optimization target variable the value after to the stipulated time predict, inclination prediction steps, the variable quantity of its time per unit after to the stipulated time with regard to the control variable that comprises integral element output in the middle of optimization target variable is predicted, steady state variable condition is set step, and its value after to the stipulated time of optimization target variable is carried out condition setting, transition state tilt condition is set step, and its condition of the variable quantity of the time per unit in transition state interval being carried out except 0 with regard to the control variable that comprises integral element output in the middle of optimization target variable is set, transition state variable condition is set step, and it carries out condition setting to the value in the transition state interval of performance variable, and this performance variable is relevant with the control variable that at least comprises integral element output in the middle of optimization target variable, and derivation step, it is obtained to meet by transition state tilt condition setting step and transition state variable condition and sets the condition that step sets, and given optimized evaluation function is carried out to the optimum solution of optimized optimization target variable, optimization target variable is that the time interval till the near stipulated time is divided into until the more than one transition state of transition state time process is interval and a transition state time warp steady state interval later, and the value of the performance variable in transition state interval and steady state interval and control variable is set as respectively to the variable of optimization target.

In above-mentioned optimization method, also can comprise that steady state tilt condition sets step, the variable quantity that comprises control variable that integral element the exports time per unit after to the transition state time that this steady state tilt condition is set in the middle of the optimization target variable that step just sets according to the collected data of data collection step carries out condition setting.

In above-mentioned optimization method, also can set in step and steady state tilt condition setting step in transition state tilt condition, to the control variable that comprises integral element output in the middle of optimization target variable, provide condition according to upper lower limit value, set in step in transition state tilt condition, provide than steady state tilt condition and set higher limit that step is larger and less lower limit as condition.

In above-mentioned optimization method, also can set in step and steady state tilt condition setting step in transition state tilt condition, to the control variable that comprises integral element output in the middle of optimization target variable, output compensates accordingly with the size of the variable quantity of time per unit, in transition state tilt condition step, for the variable quantity of same time per unit, output is than the little compensation of steady state tilt condition step, in derivation step, carry out optimization to compensation being added in to the function obtaining on evaluation function.

In above-mentioned optimization method, also can be according to evaluation function, the setting that transition state tilt condition is set in step and steady state tilt condition setting step is adjusted.

In control method of the present invention, adopt the desired value of being exported by above-mentioned optimization method, control method comprises control step, this control step is controlled the desired value that performance variable and control variable have been exported towards optimization apparatus, and at least have the control variable of integration class stablized to parameter stabilization time of adjusting to the time till desired value, with stabilization time parameter correlation connection determine the transition state time.

[effect of invention]

By above-mentioned explanation, according to the present invention, even if there is the control object of integration class, also can obtain this excellent effect of optimization of the desired value of control.

Brief description of the drawings

Fig. 1 is the key diagram for principle of the present invention is described.

Fig. 2 is the structural drawing that represents the structure of the optimization apparatus in embodiments of the present invention 1.

Fig. 3 is the process flow diagram that the action case of the steady state prediction section 102 of the 1st prediction section 121 to having the optimization apparatus in embodiments of the present invention 1 and the 2nd prediction section 122 describes.

Fig. 4 is the structural drawing of the structure of the desired value operational part 105 that represents that the optimization apparatus in embodiments of the present invention 1 possesses.

Fig. 5 is the process flow diagram that the action (optimization method) to the optimization apparatus in embodiments of the present invention 1 describes.

Fig. 6 is the structural drawing of the structure of the desired value operational part 105a that represents that the optimization apparatus in embodiments of the present invention 2 possesses.

Fig. 7 is for to providing with respect to the key diagram that example describes of upper lower limit value tilting.

Fig. 8 is the structural drawing of the structure of the desired value operational part 105a that represents that the optimization apparatus in embodiments of the present invention 3 possesses.

Fig. 9 is the structural drawing that represents the structure example of the system of carrying out Model Predictive Control.

Figure 10 is the key diagram for the structure of control object is described.

Embodiment

［ principle ］

First, principle of the present invention is described.

As illustrated in the hurdle of " problem that invention will solve ", in the optimization of desired value of control object with integration class, the inclination that General Requirements has the control variable of integration class after the predetermined stipulated time is 0 or is less value.But in the midvoyage condition controlling towards desired value (transition state), it may not be necessary that this requirement is considered to.Again, the variable quantity of the value to the control variable relevant with integration class carries out optimization, considers the value of the performance variable in transition state, as long as determine the time in an interim state.

In the present invention, the not only desired value after official hour, and also the value of performance variable in transition state and control variable is also considered as the known variables of optimization problem.Append the supposition time (hereinafter referred to as the transition state time) in an interim state as optimized parameter again.And, as shown in Figure 1, for the inclination having in the transition state of control variable of integration class, be not the condition of 0 o'clock (non-zero) as tilting, can allow and there is inclination.By like this, utilize the time in an interim state, to have integration class control variable value taking change to energetically prerequisite be optimized as may.Again, by supposing the time in an interim state, the variable quantity with the control variable of integration class becomes limited, and concrete value is estimated to become possibility.Thus, can list the control variable with integration class in optimization problem clearly.

As the convergency value y(∞ of the control variable under the steady state of another problem described in " problem that invention will solve ") imponderable problem also can by set the transition state time be solved.Be not limited to converge on certain value although there is the control variable of integration class, can predict the value before finite time.Therefore,, for the control variable with integration class, the predicted value before the calculating transition state time is to substitute the convergency value in steady state.Again, will calculate predicted value utilizes in optimization problem.By like this, y(∞) imponderable problem also solved.

By as mentioned above, can make to have the value of the more approaching expectation of desired value of the control variable of integration class and the performance variable relevant with this control variable, can carry out energetically optimization.

Below, with reference to accompanying drawing, embodiments of the present invention are described.

［ embodiment 1 ］

First, with Fig. 2, embodiments of the present invention 1 are described.Fig. 2 is the structural drawing that represents the structure of the optimization apparatus in embodiments of the present invention 1.This optimization apparatus comprises: Data Collection portion 101, steady state prediction section 102, inclination prediction section 103, model storage part 104, and desired value operational part 105.Again, steady state prediction section 102 comprises the 1st prediction section 121 and the 2nd prediction section 122.

Data Collection portion 101 for according to the process 131 as control object, predicts the response in future control variable, performance variable, disturbance variable etc., process 131, collects necessary data.These data are sent to steady state prediction section 102 and inclination prediction section 103.Model storage part 104 is predicted the movement of process 131, again, in order to solve the relation of the variable quantity of performance variable and the variable quantity of control variable, stores necessary mathematical model.As this mathematical model, there are transfer function model, state space schematic model, step response model etc.Certainly, be not limited to these models, can also utilize other mathematical model.

102 supposition of steady state prediction section just continue as the value of the performance variable of the input to process 131 appearance that keeps current, and the value of the control variable after the predetermined stipulated time of process 131 is predicted.This official hour, being preferably for the control variable convergence beyond integration class is the sufficient time.

At this, as previously mentioned, because the control variable of integration class is not limited to converge on finite value, thereby can not get the predicted value of the steady state in original meaning.For this reason, in embodiment 1, in steady state prediction section 102, there is the 1st prediction section 121 and the 2nd prediction section 122.For the control variable of integration class, by the 1st prediction section 121, utilize the predicted value before the transition state time specifying.The control variable of the 1st prediction section 121 to the output that comprises integral element in the middle of the optimization target variable that becomes optimized object in the collected data of data collection unit 101, carries out the prediction of the value after the transition state time.Although the transition state time is now, before the above-mentioned time of predetermining, must be limited.

On the other hand, for untapped control variable and performance variable in the 1st prediction section 121, in the 2nd prediction section 122, obtain the predicted value before official hour.The 2nd prediction section 122 is carried out the prediction of the value after official hour to the control variable of the output that does not comprise integral element in the middle of optimization target variable.This official hour can be also unlimited.Because the object of the 2nd prediction section 122 is not integration class, so the predicted value before Infinite Time converges on certain value.In addition, below, even after supposing and being not Infinite Time, also will comprise after the transition state time till official hour is called steady state.Again, in the middle of the predicted value of exporting in steady state prediction section 102, the predicted value of the control variable of the integration class that the 1st prediction section 121 is exported is the value after finite time, although correctly say it is not the predicted value of steady state, below for simplicity, suppose that also comprising this value is called steady state predicted value.

In addition, optimization target variable is, by interval the more than one transition state of the transition state time process till the time interval of official hour is divided into an and transition state time warp steady state interval later, and the variable that the value of the performance variable in transition state interval and steady state interval and control variable is individually set as optimization target.

Secondly,, to thering is the action case of steady state prediction section 102 of two the 1st prediction section 121 and the 2nd prediction section 122, describe with the process flow diagram of Fig. 3.

First,, in step S101, steady state prediction section 102 is reset to 1 by control variable number i.Again, in step S102, steady state prediction section 102 is reset to 1 by performance variable number j.Secondly,, in step S103, in the model that steady state prediction section 102 is stored model storage part 104, whether exist from performance variable j and inquire about to the model of control variable i.If there is corresponding model ("Yes" of step S103),, in step S104, whether steady state prediction section 102 is that integration class judges to this model.

In the situation that model is integration class ("Yes" of step S104), in step S105, obtained until the time series of the output of model response after the transition state time by the 1st prediction section 121 in steady state prediction section 102, and obtain control variable j from the predicted value to the amount of the variation the termination of prediction at present by performance variable i.

On the other hand, in the situation that model is not integration class ("No" of step S104), in step S106, obtained until the predicted value of the amount (amount that the control variable j producing because of performance variable i changes) of the time series variation of the output response of the model after official hour by the 2nd prediction section 122 in steady state prediction section 102.In the situation that official hour is Infinite Time, obtain until the predicted value of the amount that the value convergence of control variable changes again.

Above-mentioned prediction be in supposition to every couple of performance variable j and control variable i, performance variable j will also maintain from now on and will carry out on the basis of currency.

At this, be there is no the words ("No" of step S103) of model by being judged as of step S103, because it doesn't matter in the prediction of performance variable j and control variable i, so steady state prediction section 102 is skipped prediction computing, whether be last judging (step S107) to the performance variable as current object, in the situation that not being last ("No" of step S107), performance variable number i being added to 1 and carry out lifting operation variable number (step S108), and be back to step S103.

According to the above description, carry out the words of above-mentioned computing ("No" of step S107) to there is all operations variable of model from performance variable j to control variable i, in step S109, steady state prediction section 102 adds up to the predicted value of the variable quantity of trying to achieve, and as the prediction variable quantity of control variable i.This prediction variable quantity is added to the current predicted value of control variable i or current measured value, just become the predicted value under the steady state of control variable i.By carry out above computing (step S110, step S111) with regard to each control variable, steady state prediction section 102 is obtained steady state predicted value the output of all control variable.In addition, steady state predicted value is output together with adopting the value of performance variable of calculating of this value, and is delivered to desired value operational part 105.

In addition, the computing of steady state prediction is not limited to as mentioned above, can be also as follows.For example, in above-mentioned, although for to obtain the predicted value before official hour by the 2nd prediction section 122, this official hour also can be identical with the transition state time.In this case, the 1st prediction section 121 becomes identical with the 2nd prediction section 122.Also official hour can be taken as to infinite point again.In this case, for the control variable that is not integration class, predict the outcome identical with correlation technique will be become.

Again, in the explanation of process flow diagram that adopts Fig. 3, although carried out prediction computing with regard to every pair of performance variable and control variable, be not limited to this, also can gather calculating.For example, adopt the computing of state space schematic model, also can consider side by side to carry out the RESPONSE CALCULATION of multi-variable system entirety.But, for long-pending classification associated part (being equivalent to the 1st prediction section 121), be necessary to note need to stop with the finite time of regulation prediction and calculation.Again, in the case of there is disturbance variable and the response of control variable exerted an influence, preferably, adopt forecast model between disturbance variable and control variable carry out same be predicted as good.In a word, about the computing of steady state prediction, the method that adopts mathematical model the value in the future to control object to predict is to go for various technology.

Then, inclination prediction section 103 is described.The variable quantity (inclination) of inclination prediction section 103 time per unit after with regard to official hour to the control variable that comprises integral element output in the middle of optimization target variable is predicted.Except prediction be not value but inclination this point, all the other are identical with the 2nd prediction section 122, can similarly predict with the 2nd prediction section 122.For example, the mathematical model till the inclination of preparation from performance variable to control variable, obtains the response of inclination by the data of the mathematical model of preparation and the control object of collection, the inclination predicted value using the value of this response convergence under steady state.

Then, illustrate in greater detail with regard to desired value operational part 105 use Fig. 4.Fig. 4 is the structural drawing of the structure of the desired value operational part 105 that represents that the optimization apparatus in embodiments of the present invention 1 possesses.Desired value operational part 105 comprises: transition state tilt condition configuration part 151, steady state tilt condition configuration part 152, transition state variable condition configuration part 153, steady state variable condition configuration part 154 and derivation portion 155.

Condition setting is carried out to the inclination having under the transition state of control variable of integral element output in transition state tilt condition configuration part 151.Set the condition except 0 in the variable quantity (inclination) of the time per unit in transition state interval to comprising the control variable of integral element output in the middle of optimization target variable.In the optimization apparatus of embodiment 1, there is inclination by allowing in transition state, realize the optimization more positive than corresponding technology.Therefore, at least do not provide as inclination being restricted to the condition in the of 0.In addition also can carry out not to the such setting of condition for inclination.

Condition setting, through after having the transition state time of control variable of integral element output, is carried out to the inclination after official hour in steady state tilt condition configuration part 152.In addition steady state tilt condition configuration part 152 inscape not necessarily for purposes of the invention.But the transition state time, it is probably impaired that the stability of control system and desired value are followed characteristic through also allowing that control variable has large inclination later.Therefore, conventionally, with corresponding technology similarly, preferably, by steady state tilt condition configuration part 152 set as after official hour to the condition tilting to suppress.

Transition state variable condition configuration part 153 is set the condition in the transition state of performance variable and control variable.Usually, it is difficult that performance variable physically departs from from bound condition, for at least relevant with the control variable that comprises integral element output performance variable, that is, once for change this performance variable value words,, integral element is exported the performance variable that will change and is just necessary to provide condition.By providing like this condition, due to the inclination of the control variable that comprises integral element output, be just limited in attainable scope from the viewpoint of performance variable, thereby can obtain suitable solution.In addition, can provide condition to control variable, better although it is so, but this is not necessarily.

Condition setting is carried out to the value of (under steady state) after the official hour of the performance variable as optimization target variable and control variable in steady state variable condition configuration part 154.This condition can be both rigid condition, can be again elastic condition.Steady state variable condition configuration part 154 is identical with associated technology, also can similarly set.

The various conditions of above-mentioned transition state tilt condition configuration part 151, steady state tilt condition configuration part 152, transition state variable condition configuration part 153, steady state variable condition configuration part 154 and derivation portion 155, transition state time, the model of control object, the optimization evaluation function of giving are passed to derivation portion 155.The above-mentioned value of derivation portion 155 use is obtained in the scope satisfying condition at all variablees evaluation function is carried out to optimized optimal objective value.

At this, derivation portion 155 is described for the method that provides of the transition state time of computing.The method that provides of transition state time can at random be set in the scope that is no more than official hour.But, be preferably, be greater than the time (after changing the value of performance variable, until the inclination of the control variable of integration class becomes certain time) of the inclination convergence of integration class.

Again, in the case of having the Model Predictive Control portion controlling to by the determined optimal objective value of optimization apparatus (optimization method) of present embodiment, parameter that also can reference model PREDICTIVE CONTROL portion is determined.For example, in the case of provided the stabilization time of control variable of integration class as the parameter of Model Predictive Control portion, can be by this setting-up time as the transition state time.Again, not Model Predictive Control in the control to optimal objective value, have at least in the situation of the parameter of adjusting the stabilization time of the control variable to integration class, if or provided using stabilization time as controlling specification, can associatedly with the stabilization time providing determine the transition state time.

Again, be sent to control device in the determined optimal objective value of the optimization apparatus by present embodiment (optimization method), in the situation that control device is controlled this optimal objective value, stabilization time can control to desired value the control variable of integration class in conjunction with control device time, determine the described transition state time.For example, the control variable that makes integration class at control device with stabilization time TSt follow desired value, the transition state time is determined by " (transition state time)=γ × TSt, (γ is predefined positive constant) ".

γ is preferably the value of 1 left and right, conventionally can be set as 1.Definite method of transition state time is not limited to the method, but is set as and roughly proportional relation stabilization time, if stabilization time elongated, the transition state time is also by elongated.Due to by so associated with stabilization time of control device and transition state time of definite optimization apparatus suitably and automatically, thereby have nothing to do with the setting of stabilization time of control device, can obtain effect of the present invention.

In the situation that determining the transition state time like this, the control device being connected with optimization apparatus is necessary the device of the stabilization time of the control variable that is at least adjustable integration class.Preferably can directly specify stabilization time, but indirectly determine and also can stabilization time by the parameter of controlling.For example, can be also the reference time by pre-determining regulation, and by reference time and described parameter is carried out multiplication or division arithmetic is determined the structure that stabilization time is such.Though as long as meet the method for above-mentioned condition be what control method all can, but consider the given condition of optimization apparatus and can control this point, can say that Model Predictive Control is applicable to the present invention.

In addition, although be illustrated determining the method for the transition state time of optimization apparatus the stabilization time in conjunction with control device at this, but be also the same the stabilization time of determining control device in conjunction with transition state time of optimization apparatus, as long as those skilled in the art's words be can hold intelligible.

Again, can be divided into the interval till the transition state time multiple, using the value of the control variable in each interval of having cut apart and performance variable as optimized object.But, effect of the present invention be utilize be divided into the transition state that allows the control variable of integration class to there is inclination interval and, the steady state interval that tilts to suppress is realized, even transition state interval is cut apart and can't be increased significantly effect.

Above content is gathered, and the action (optimization method) of the optimization apparatus in embodiments of the present invention 1 becomes shown in the process flow diagram of Fig. 5.

First, in step S201, the data of control object are collected by Data Collection portion 101, and the packet of this control object is containing the control variable (data collection step) of the performance variable for control object is controlled and control object output.Then,, in step S202, the mathematical model of control object is stored in to model storage part 104(model storing step).

Then, in step S203, the 1st prediction section 121 is to carrying out the prediction (the 1st prediction steps) of the value after the transition state time by the control variable that comprises integral element output in the middle of the optimization target variable that becomes optimized object in the collected data of Data Collection portion 101 in step S201.Then,, in step S204, the 2nd prediction section 122 is carried out the prediction (the 2nd prediction steps) of the value after official hour to the control variable that does not comprise integral element output in the middle of optimization target variable.

Then,, in step S205, inclination prediction section 103 is carried out the prediction (inclination prediction steps) of the variable quantity (inclination) of the time per unit after official hour to the control variable that comprises integral element output in the middle of optimization target variable.Then,, in step S206, condition setting (steady state variable condition is set step) is carried out to the value after the official hour of optimization target variable in steady state variable condition configuration part 154.Then, in step S207, transition state tilt condition configuration part 151 is carried out the condition except 0 in the variable quantity of the time per unit in transition state interval and is set (transition state tilt condition setting step) to comprising the control variable of integral element output in the middle of optimization target variable.

Then, in step S208, condition setting (transition state variable condition is set step) is carried out to the value in the transition state interval of the performance variable relevant with the control variable that at least comprises integral element output in the middle of optimization target variable in transition state variable condition configuration part 153.After this,, in step S209, derivation portion 155 obtains and meets the condition being set by step S207 and step S208 and the optimum solution (derivation step) of the optimized evaluation function providing being carried out to optimized optimization target variable.

In addition, optimization apparatus is to comprise: CPU(Central Processing Unit; Central operation treating apparatus), the computer equipment such as main storage means, external memory and network connection device, utilize and be deployed in the program of main storage means by the action of CPU, can realize above-mentioned each function.Again, each function also can be scattered in many computer equipments.

As explained above, according to embodiment 1, by the 1st prediction section 121, the control variable that comprises integral element output in the middle of the optimization target variable that becomes optimized object in the collected data of data collection unit 101 is carried out to the prediction of the value after the transition state time, again, due in desired value operational part 105, by transition state tilt condition configuration part 151, the inclination having in the transition state of control variable of integration class is set, making as tilting is not that the condition of 0 o'clock (non-zero) allows to have inclination, even thereby there is the control object of integration class, also the optimization that can control.

［ embodiment 2 ］

Then, with Fig. 6, embodiments of the present invention 2 are described.Fig. 6 is the structural drawing that represents the part-structure of the optimization apparatus in embodiments of the present invention 2.In Fig. 6, show the structure of the desired value operational part 105a that optimization apparatus possesses.Other structure is identical with the embodiment 1 describing with Fig. 2.

In embodiment 2, transition state tilt condition configuration part 151, steady state tilt condition configuration part 152 have the feature providing by being limited in upper lower limit value with respect to the condition of the inclination of the control variable of integration class., the condition of inclination is rigid condition.Can set by transition state tilt condition configuration part 151 bound more relaxing compared with the upper lower limit value being set by steady state tilt condition configuration part 152 again., become that the upper limit is large, lower limit is less.In addition, transition state tilt condition configuration part 151 also can comprise the upper limit is set as to positive infinity, is the infinity of bearing by lower limit set.Under this and transition state, be in fact of equal value to tilting not provide condition.

Again, in embodiment 2, desired value operational part 105a comprises: new pseudo-gain matrix operational part 156 and tilt gain matrix operation portion 157.

Below, the desired value operational part 105a in embodiment 2 is described in detail.In the present invention, as previously mentioned, imagination performance variable transition state under steady state, get different values.Below, in order to distinguish the value in value and the steady state in transition state, suppose and the variate-value of transition state is added to slash symbol " ' in upper right ".For example, u ₁,, u _mrepresent the value of the performance variable in steady state; U ' ₁,, u ' _mrepresent the value in transition state.Distinguish similarly to control variable and by the synthetic vector forming of variable.

Pseudo-gain matrix operational part 156 is obtained the correlation matrix of the variable quantity of performance variable and the variable quantity of control variable.The effect of pseudo-gain matrix is identical with the DC current gain matrix in corresponding technology.The line number of DC current gain matrix is i.e. " the m row " of the quantity quantity that " n is capable ", columns are performance variable of control variable, the key element G of the capable j row of i _ijequate with the amount that the value of performance variable i is changed to control variable j variation in 1 o'clock.Pseudo-gain matrix is also same, but due to the variate-value of transition state is imported in optimization problem, so there is different parts.

First, columns becomes 2 times of quantity of performance variable i.e. " 2m row ".This is to allow and get different values due to performance variable under transition state and steady state.Again, be integration class at the forecast model from performance variable i to control variable j, with being only subject to the impact of value of performance variable of transition state contrary, in the situation that forecast model is not integration class, be only subject to the impact of the performance variable value of steady state.Consider this situation, the calculating of pseudo-gain matrix becomes as follows:

［ when the forecast model from performance variable i to control variable j is integration class ］

The key element of the capable j row of i of matrix only changes at 1 o'clock to performance variable i under transition state, and control variable j sets the amount changing (before becoming steady state) during transition state.As computing method, calculate the unit step response (the output response when unit's of providing step is inputted) of forecast model, and obtain the value after the transition state time.The value and the tilt gain described later that deduct from the transition state time the unhelpful time are multiplied each other to try to achieve, or more merely, have the method for obtaining product of transition state time and tilt gain etc.In addition, the key element of the capable j+m row of i is made as 0.

［ when the forecast model from performance variable i to control variable j is not integration class ］

If the key element of the capable j row of the i of matrix is the DC current gain that the key element of the capable j+M row of 0, i is set as forecast model.

Suppose that by pseudo-gain matrix obtained above be G, the relation of the variable quantity of performance variable and control variable becomes by shown in following formula (7).This is input to derivation portion 155.

[formula 3]

$\mathrm{\Δy}=G\left[\begin{array}{c}\mathrm{\Δ}{u}^{\′}\\ \mathrm{\Δu}\end{array}\right]...\left(7\right)$

Tilt gain matrix operation portion 157 obtains the correlation matrix of the variable quantity of the variable quantity of performance variable and the inclination of control variable.Matrix is the capable m row of n, the key element S of the capable j row of i _ijequate with the amount that the value of performance variable i is changed to the tilt variation of 1 o'clock control variable j.In addition, when the forecast model from performance variable i to control variable j is not integration class, S _ijbe 0.Even if this means that the inclination under the steady state of the value control variable j that changes performance variable i also will converge on 0.Suppose that tilt gain matrix is S, forecast model transfer function matrix is expressed as to G(s), S can be tried to achieve by following formula (8).

[formula 4]

$S=\underset{s\→0}{\lim }\mathrm{sG}\left(s\right)...\left(8\right)$

Steady state variable condition configuration part 154 is set the steady state of performance variable and control variable.The rigid condition providing by restriction upper lower limit value can be represented by following formula (9).Although steady state variable condition configuration part 154 is substantially identical with corresponding technology, but as y(∞) the output of steady state prediction section of value in the middle of, the predicted value of the control variable of integration class becomes by the predicted value before the transition state time of the regulation of the 1st prediction section output, and this point is different from prior art.

[formula 5]

$\underset{\‾}{u_i}-u_i(k-1)\≤\mathrm{\Δ}{u}_i\≤\stackrel{\‾}{u_i}-u_i(k-1),\mathrm{fori}=1,...,m$

$\underset{\‾}{y_i}-y_i(\∞)\≤\mathrm{\Δ}{y}_i\≤\stackrel{\‾}{y_i}-y_i(\∞),\mathrm{fori}=1,...,n...\left(9\right)$

Condition setting is carried out to the value in the transition state interval of performance variable relevant with the control variable that at least comprises integral element output in the middle of optimization target variable in transition state variable condition configuration part 153.For example, the upper lower limit value in the transition state of the performance variable as optimization target variable and control variable is set.Even if general operation variable is also assigned to the upper lower limit value identical with steady state under transition state, be preferably such setting.Also be same for control variable, can not exceed upper lower limit value owing to setting control variable in transition state for, be assigned to the upper lower limit value identical with steady state so be preferably.But, because also there is following viewpoint, so not necessarily.

Action in the transition state of the control variable that comprises integration class is generally and increases progressively from currency to desired value or successively decrease.Thereby, for the control variable that comprises integration class, if be assigned to upper lower limit value under steady state, even if the worry departing from from upper lower limit value under transition state is less.

The inclination of the control variable in transition state is set to upper lower limit value (aftermentioned), and false large inclination if there is not, in transition state, even if the restriction that the value of control variable is not set to upper lower limit value is also sufficient.

According to control variable, also do not affect even if sometimes for the moment exceed upper lower limit value.For this control variable, sometimes use and do not allow the rigid condition departing from from upper lower limit value without exception with it, it would be better to that being suitable for one recompenses in the face of departing from, one side allows that the elastic condition departing from is good.

The variable quantity (inclination) of steady state tilt condition configuration part 152 to the time per unit under the steady state of the control variable that comprises integration class carries out upper lower limit value setting.The most general is also to use in corresponding technology, and the inclination under steady state is restricted to 0.This is of equal value with upper lower limit value is all made as to 0.This restriction is represented by formula, be " S Δ u=0(10) ".This is equality condition, is rigid condition.

If, even if also allow inclination to a certain degree under steady state, make it have upper lower limit value amplitude, become as shown in the formula inequality condition such shown in (11).Because this condition is not yet allowed from upper lower limit value and is exceeded, so become rigid condition.

[formula 6]

$\underset{\‾}{b_{\mathrm{SL}}}\≤\mathrm{S\Δu}\≤b_{\mathrm{SL}}...\left(11\right)$

Describe providing with respect to an example of the upper lower limit value tilting.As shown in Figure 7, although be the inclination of allowing in steady state, if in the situation that control variable approaches the upper limit, the leeway that positive inclination (increase direction) is allowed is less.On the other hand, the inclination till negative inclination (reducing direction) is considered to allow to a certain degree.If in the situation that control variable approaches lower limit, become with contrary as mentioned above.Again, if fully away from coming the upper limit and lower limit, no matter just with negative, even if there have some inclinations immediately control system to be produced to dysgenic possibility to be lower.What make that this idea reflected is as shown in the formula the inequality shown in (12).

[formula 7]

$\frac{\mathrm{\α}(\underset{\‾}{y_i}-{\hat{y}}_i(k-1))}{T_{\mathrm{tr}}}\≤y_i^{\left(\mathrm{SL}\right)}\≤\frac{\mathrm{\β}(\stackrel{\‾}{y_i}-{\hat{y}}_i(k-1))}{T_{\mathrm{tr}}}...\left(12\right)$

the predicted value of an i control variable before control cycle, y _i ^(SL)the variable quantity (inclination) of the time per unit of i control variable under steady state, T _trfor positive arbitrarily constant, the transition state time of imagining while being preferably optimization or the value of same degree.

Again, y _i ^(SL)can calculate according to variable quantity (inclination) predicted value of the time per unit of integration class control variable and the variation delta u of tilt gain matrix S and performance variable.α, β are more than 0 constant.If this constant is set as to 0, become identical with the situation that inclination is restricted to 0.If give the value that is greater than 0, for inclination, set according to the distance till from currency to bound and variable upper lower limit value.In addition, if hypothesis T _trthe transition state time of imagining during for optimization, become according to the transition state time also variable upper lower limit value.

For optimized evaluation function, if this control variable is set for and approached as much as possible specific desired value, also can replace bound by this desired value again.For optimized evaluation function, in the maximized situation that is set with this control variable, if replace lower limit with currency, just become and only allow that making the value of control variable increase or tilt is 0 such setting again.

α, β can determine as shown below.Inclination under steady state is taking as far as possible little scope as good, and α, β are preferably the less value that is no more than 0.1.Again, α, β also can change its value according to control variable.For example, can consider about wanting the control variable that suppresses of tilting with 0 or minimum value; Use relatively large value for the control variable that there is no need to suppress to tilt.

The variable quantity (inclination) of transition state tilt condition configuration part 151 to the time per unit in the transition state of the control variable that comprises integration class set upper lower limit value restriction.Be at least than given will the relaxing in steady state tilt condition configuration part 152 in this given condition.Thereby, be necessary the higher limit of inclination to set for larger than the higher limit of steady state, lower limit is set littlely.For example, under steady state, if set the bound as formula (12), be necessary to make α, β to be greater than the value under steady state.By like this, in transition state, will allow the inclination larger than equilibrium state.By the upper limit of inclination is set as to positive infinity, lower limit set is the infinity of bearing, also have no relations with the state equivalent not imposing a condition again.

The various upper lower limit values that obtain by above-mentioned computing, gain matrix, to optimization evaluation function transferred to optimization operational part 155, all variablees, in condition and range, and can be obtained as made the optimal objective value of evaluation function minimum.Following formula (13) is an example of above-mentioned regularization.

[formula 8]

$u(k-1)={[u_1(k-1),u_2(k-1),...,u_m(k-1)]}^T,y(\∞)={[y_1(\∞),y_2(\∞),...,(\∞)]}^T$

Δu'=[Δu' ₁，Δu' ₂，...，Δu' _m] ^T，Δy'，=[Δy' ₁，Δy' ₂，...，Δy' _n] ^T

Δu=[Δu ₁，Δu ₂，...，Δu _m] ^T，Δy=[Δy ₁，Δy ₂，...，Δy _n] ^T

$\mathrm{\Δ}{x}^{\′}=\left[\begin{array}{c}\mathrm{\Δ}{u}^{\′}\\ \mathrm{\Δ}{y}^{\′}\end{array}\right],\mathrm{\Δx}=\left[\begin{array}{c}\mathrm{\Δu}\\ \mathrm{\Δy}\end{array}\right],x_0=\left[\begin{array}{c}u(k-1)\\ y(\∞)\end{array}\right]$

Δx _opt=argminJ(X ₀+Δx)，x _opt=x ₀+Δx _opt

J(x)=x ^THx+c ^Tx

$\mathrm{\Δy}=G\left[\begin{array}{c}\mathrm{\Δ}{u}^{\′}\\ \mathrm{\Δ}\end{array}\right],\mathrm{\Δ}{y}^{\′}=G\left[\begin{array}{c}0\\ \mathrm{\Δ}{u}^{\′}\end{array}\right]$

S Δ u=0 or S Δ u ≤=b _sL

$\underset{\‾}{u_i}-u_i(k-1)\≤\mathrm{\Δ}{u}_i\≤\stackrel{\‾}{u_i}-u_i(k-1),\mathrm{fori}=1,...,m$

$\underset{\‾}{y_i}-y_i(\∞)\≤\mathrm{\Δ}{y}_i\≤\stackrel{\‾}{y_i}-y_i(\∞),\mathrm{fori}=1,...,n$

$\underset{\‾}{u_i}-u_i(k-1)\≤\mathrm{\Δ}{u}_i\≤\stackrel{\‾}{u_i}-u_i(k-1),\mathrm{fori}=1,...,m$

$\underset{\‾}{y_i}-y_i(\∞)\≤\mathrm{\Δ}{y}_i^{\′}\≤\stackrel{\‾}{y_i}-y_i(\∞),\mathrm{fori}\∈I_1...\left(13\right)$

In formula, I ₁for thering is the set of index entirety of the control variable beyond the control variable of integration class.

In addition, the result of optimization computing, although obtain as Δ u ', the Δ y ' of the solution of transition state with as Δ u, the Δ y of the solution of steady state, but due to the optimal objective value of the multivariable Control as comprising Model Predictive Control, bottom line necessity be only the solution of steady state, thereby also can only Δ u, Δ y be exported as optimal objective value.Certainly, also the solution of transition state can be exported together and utilized by control algorithm.

［ embodiment 3 ］

Then, with Fig. 8, embodiments of the present invention 3 are described.Fig. 8 is the structural drawing that represents the part-structure of the optimization apparatus in embodiments of the present invention 3.In Fig. 8, show the structure of the desired value operational part 105a that optimization apparatus possesses.In embodiment 3, establish the 155a of derivation portion and comprise new function correction portion 158.Other structure is with identical with the embodiment 1 of Fig. 2 explanation.

Embodiment 3 has following feature: in transition state tilt condition configuration part 151 in desired value operational part 105a, steady state tilt condition configuration part 152, the inclination of control variable is set to the elastic condition being compensated.

In embodiment 3, in order to suppress the variable quantity (inclination) of time per unit of the control variable that comprises integration class, the size of inclination is joined in evaluation function by way of compensation.By this compensation, because the value of the inclination less " evaluation function+compensation " of the control variable in steady state and transition state becomes better, thereby can obtain little optimal objective value.This compensation is affixed to the original optimization evaluation function providing from outside by function correction portion 158.Appended the correction evaluation function compensating and be used to the derivation in the 155a of derivation portion again.In following formula (14), the fixed case of evaluation function is shown.

[formula 9]

$J^{\′}\left(x\right)=J\left(x\right)+\underset{i\∈I_2}{\mathrm{\Σ}}{p}_i{\left(y_i^{\left(\mathrm{SL}\right)}\right)}^2+\underset{i\∈I_2}{\mathrm{\Σ}}{p}_i^{\′}{\left(y_i^{\′\left(\mathrm{Sl}\right)}\right)}^2...\left(14\right)$

At this, J ' is (x) revised evaluation function, J(x) be original optimization evaluation function, I ₂the set with the index entirety of the control variable of integration class, p _ithe penalty coefficient to the inclination under the steady state of i control variable, y _i ^(SL)the variable quantity (inclination) of the time per unit of i control variable under steady state, p ' _ithe penalty coefficient to the inclination in the transition state of i control variable, y ' _i ^(SL)it is the variable quantity of the time per unit of i control variable in transition state.

Inclination in transition state is necessary than relaxing in steady state.Thereby, the big or small penalty coefficient p' of the inclination to transition state _ishould set the big or small penalty coefficient p of the inclination of comparison steady state for _ilittle.Again, as p ' _i=0, also can not compensate the inclination in transition state.

Above-mentioned example will compensate in the time tilting for non-zero, but also can not compensated for the inclination till the upper lower limit value of specifying.This example is as shown in the formula shown in (15).

[formula 10]

$\underset{\‾}{y_i^{\left(\mathrm{SL}\right)}}-{\mathrm{\ϵ}}_i\≤y_i^{\left(\mathrm{SL}\right)}\≤\stackrel{\‾}{y_i^{\left(\mathrm{SL}\right)}}+{\mathrm{\ϵ}}_i,{\mathrm{\ϵ}}_i>0$

$J^{\′}\left(x\right)=J\left(x\right)+\underset{{i\∈I}_2}{\mathrm{\Σ}}{p}_i{\left({\mathrm{\ϵ}}_i\right)}^2...\left(15\right)$

In formula, the lower limit of i control variable about the inclination under steady state, the upper limit, ε _iit is the amplitude that can depart from bound.

If the inclination under steady state is limited in bound, because ε _ican be 0, so evaluation function can not append compensation.On the other hand, in the time that the inclination under steady state departs from bound, will be given and square compensation being directly proportional of bias.Due to compensation less revised evaluation function value will become less, thereby, its result, depart from bound measure suppress.

In above-mentioned example, the amplitude that can depart from is set as the value that bound is identical, but also can be set as different values.For example, if want strictly to suppress thering is positive inclination, and slightly do tolerantly to thering is negative inclination, as long as make the variable that is equivalent to ε i respectively as bound, increase penalty coefficient to the upper limit, and reduce the penalty coefficient to lower limit.

As mentioned above, embodiment 3 is characterised in that: the condition for the inclination that sets of tilt condition configuration part for suppressing to tilt not is to provide with rigid conditions such as equality condition or inequality conditions, but so that by the compensation of tilting to carry out is appended to, elastic condition in evaluation function provides.Except this feature, other are identical with embodiment 2.Again, the enforcement that also two kinds of methods can be combined.

In addition, the present invention is not limited to the embodiment of above-mentioned explanation, within the scope of technological thought of the present invention, obviously has this area Chang Shizhe, can implement various deformation and combination.For example, the present invention goes for Model Predictive Control.As long as will be by the determined desired value of optimization of the present invention for Model Predictive Control.Also can, according to control specification or the parameter of stabilization time of determining Model Predictive Control portion, automatically determine the transition state time again.

[explanation of symbol]

101 ... Data Collection portion, 102 ... steady state prediction section, 103 ... inclination prediction section, 104 ... model storage part, 105 ... desired value operational part, 121 ... the 1st prediction section, 122 ... the 2nd prediction section, 151 ... transition state tilt condition configuration part, 152 ... steady state tilt condition configuration part, 153 ... transition state variable condition configuration part, 154 ... steady state variable condition configuration part, 155 ... derivation portion.

Claims (12)

1. an optimization apparatus, is characterized in that, comprising:

Data Collection portion, it collects the data of control object, and the packet of described control object is containing the control variable of the performance variable for control object is controlled and the output of described control object;

Model storage part, its mathematical model to described control object is stored;

The 1st prediction section, it is just collected in the control variable that comprises integral element output in the middle of the optimization target variable that becomes optimization target in the data of the described Data Collection portion value after to the transition state time and predicts;

The 2nd prediction section, its with regard to the control variable that does not comprise integral element output in the middle of described optimization target variable the value after to the stipulated time predict;

Inclination prediction section, the variable quantity of its time per unit after to the stipulated time with regard to the control variable that comprises integral element output in the middle of described optimization target variable is predicted;

Steady state variable condition configuration part, its value after to the stipulated time of optimization target variable is carried out condition setting;

Transition state tilt condition configuration part, its condition of the variable quantity of the time per unit in transition state interval being carried out except 0 with regard to the control variable that comprises integral element output in the middle of described optimization target variable is set;

Transition state variable condition configuration part, it carries out condition setting to the value in the transition state interval of performance variable, and this performance variable is relevant with the control variable that at least comprises integral element output in the middle of described optimization target variable; And

Derivation portion, it is obtained and meets the condition being set by described transition state tilt condition configuration part and described transition state variable condition configuration part and the optimum solution of given optimized evaluation function being carried out to optimized described optimization target variable,

Described optimization target variable is that the time interval till the near stipulated time is divided into until the more than one transition state of transition state time process is interval and a transition state time warp steady state interval later, and the value of the described performance variable in interval described transition state and described steady state interval and described control variable is set as respectively to the variable of optimization target.

2. optimization apparatus according to claim 1, is characterized in that,

Comprise steady state tilt condition configuration part, the variable quantity of this steady state tilt condition configuration part time per unit after to the described transition state time with regard to the control variable that comprises integral element output in the middle of the optimization target variable of setting according to the data that are collected in described Data Collection portion carries out condition setting.

3. optimization apparatus according to claim 2, is characterized in that,

Described transition state tilt condition configuration part and steady state tilt condition configuration part, to the control variable that comprises integral element output in the middle of described optimization target variable, provide condition according to upper lower limit value,

Described transition state tilt condition configuration part provides the higher limit larger than described steady state tilt condition configuration part and less lower limit as condition.

4. optimization apparatus according to claim 2, is characterized in that,

Described transition state tilt condition configuration part and steady state tilt condition configuration part are to the control variable that comprises integral element output in the middle of described optimization target variable, and output compensates accordingly with the size of the variable quantity of time per unit,

, for the variable quantity of same time per unit, export than the little compensation in described steady state tilt condition configuration part described transition state tilt condition configuration part,

Described derivation portion, carries out optimization to described compensation being added in to the function obtaining on described evaluation function.

5. according to the optimization apparatus described in any one in claim 2～4, it is characterized in that,

According to described evaluation function, the setting of described transition state tilt condition configuration part and described steady state tilt condition configuration part is adjusted.

6. a control device, is characterized in that, it adopts the desired value of the optimization apparatus output described in any one in claim 1～5, and described control device comprises:

Control part, it controls the desired value that performance variable and control variable have been exported towards optimization apparatus, and at least has the control variable of integration class is stablized to parameter stabilization time of adjusting to the time till desired value,

With described stabilization time parameter correlation connection determine the described transition state time.

7. an optimization method, is characterized in that, comprises the steps:

Data collection step, it collects the data of control object, and the packet of described control object is containing the control variable of the performance variable for control object is controlled and the output of described control object;

Model storing step, its mathematical model to described control object is stored;

The 1st prediction steps, its just by the control variable that comprises integral element output in the middle of the optimization target variable that becomes optimized object in the collected data of described data collection step the value after to the transition state time predict;

The 2nd prediction steps, its with regard to the control variable that does not comprise integral element output in the middle of described optimization target variable the value after to the stipulated time predict;

Inclination prediction steps, the variable quantity of its time per unit after to the stipulated time with regard to the control variable that comprises integral element output in the middle of described optimization target variable is predicted;

Steady state variable condition is set step, and its value after to the stipulated time of optimization target variable is carried out condition setting;

Transition state tilt condition is set step, and its condition of the variable quantity of the time per unit in transition state interval being carried out except 0 with regard to the control variable that comprises integral element output in the middle of described optimization target variable is set;

Transition state variable condition is set step, and it carries out condition setting to the value in the transition state interval of performance variable, and this performance variable is relevant with the control variable that at least comprises integral element output in the middle of described optimization target variable; And

Derivation step, it is obtained and meets the condition being set by described transition state tilt condition setting step and described transition state variable condition setting step and the optimum solution of given optimized evaluation function being carried out to optimized described optimization target variable

Described optimization target variable is that the time interval till the near stipulated time is divided into until the more than one transition state of transition state time process is interval and a transition state time warp steady state interval later, and the value of the described performance variable in interval described transition state and described steady state interval and described control variable is set as respectively to the variable of optimization target.

8. optimization method according to claim 7, is characterized in that,

Comprise that steady state tilt condition sets step, the variable quantity that comprises control variable that integral element the exports time per unit after to the described transition state time that this steady state tilt condition is set in the middle of the optimization target variable that step just sets according to the collected data of described data collection step carries out condition setting.

9. optimization method according to claim 8, is characterized in that,

Set in step and steady state tilt condition setting step in described transition state tilt condition, to the control variable that comprises integral element output in the middle of described optimization target variable, provide condition according to upper lower limit value,

Set in step in described transition state tilt condition, provide than described steady state tilt condition and set higher limit that step is larger and less lower limit as condition.

10. optimization method according to claim 8, is characterized in that,

Set step and steady state tilt condition is set in step in described transition state tilt condition, to the control variable that comprises integral element output in the middle of described optimization target variable, output compensates accordingly with the size of the variable quantity of time per unit,

In described transition state tilt condition step, for the variable quantity of same time per unit, export than the little compensation of described steady state tilt condition step,

In described derivation step, carry out optimization to described compensation being added in to the function obtaining on described evaluation function.

Optimization method described in any one in 11. according to Claim 8～10, is characterized in that,

According to described evaluation function, the setting that described transition state tilt condition is set in step and described steady state tilt condition setting step is adjusted.

12. 1 kinds of control methods, is characterized in that, it adopts the desired value of being exported by the optimization method described in any one in claim 7～11,

Described control method comprises control step, this control step is controlled the desired value that performance variable and control variable have been exported towards optimization apparatus, and at least have the control variable of integration class is stablized to parameter stabilization time of adjusting to the time till desired value

With described stabilization time parameter correlation connection determine the described transition state time.