CN1231436A - Universal multi-variable quantity model pre-estimating coordinating control method - Google Patents

Abstract

The universal multi-variable model pre-estimation coordination control method, belonging to the field of automatic control technology, can use actually-obtained any control model to make pre-estimation, including universal robust pre-estimation controller based on structure, steady-state characteristics and response time. Said control method possesses the single-value pre-estimation for correcting time domain, and multiloop, multiple pre-estimation time domain and multicycle control utilizing all measurable variables to make dynamic feedback. It can be adapted to structural change of controlled object, can automatically select and use the different pairs of controlled variable and operation variable, and optimize these variables to implement real-time coordination under the condition that these variables are not transfinite. It also can be adapted to the characretistic change of the controlled object, and can automatically modify control algorithm when the pure lag time is changed.

Description

Universal multi-variable quantity model pre-estimating coordinating control method

The present invention relates to a kind of autocontrol method, belong to automation field based on the controlled device mathematical model.

In the existing model pre-estimating control method by the digital machine realization, use the input that maximum classes is based on the controlled device of actual measurement, typical case representative has: based on model algorithm control (Model Algorithm Control) and corresponding software I DCOM, dynamic matrix control (Dynamic Matrix Control) and the corresponding software DMCplus (Aspen Tech Corp.) of discrete-time convolution model and on-line correction and RMPCT (Hi Spec Solution, Honeywell).Broad sense Prediction Control (GeneralPredictive Control) based on time series models and the knowledge of online pigtail.The another kind of discrete state spatial model that is based on feeds back Prediction Control (State Feedback Predictive Control), retreats time domain control (Receding Horizon Control) etc. as total state.Prior art all can be used for multivariable controlled process, and control and coordination optimization strategy when having considered that constraint, performance variable dimension and controlled variable dimension are inequality, adopt weighted sum that priority approach etc. is set among linear programming method, the RMPCT as adopting among the DMCplus.

The prior art weak point is: the model that can not adapt to various available controlled devices.The actual measurement input and output will be disturbed the even running of controlled device to set up model, or are difficult to use because of controlled device has the input of can not surveying; And inputoutput data can not reflect the state of controlled process comprehensively, and the state variable information of some actual measurements can not be fully used, and loses the chance of further improvement control performance.All feedback of status makes its application limited because of state may not be available entirely.Owing to adopt the multistep discreet value, do not utilize the measured state variable again, calculated amount is bigger in real time, especially to large scale system, implements comparatively difficulty.In the constraint coordination strategy, optimization aim or priority orders and linear programming method have only been considered the stable state requirement, lack the dynamic perfromance of controlled device is required coordination mutually, coordination strategy when performance variable is less than controlled variable and the two dimension real-time change with optimizing.

The objective of the invention is to provide a kind of available various models (or have only structure, steady-state characteristic and response time and do not have the controlled device of strict model) general-purpose algorithm, can make full use of all measurable variables, calculate the multi-variable quantity model pre-estimating control method simple, that robustness is high; On this basis, the quantity that provides adaptation controlled device structure and characteristic variations, multivariate constraint, performance variable (hereinafter to be referred as MV) and controlled variable (hereinafter to be referred as CV) makes dynamic control and optimizes to require to coordinate mutually with the pre-estimating coordinating control strategy that running status changes.

It is as follows to reach the technology that these purposes adopt:

Universal multi-variable quantity model pre-estimating coordinating control method is characterized in that:

The universal method that the discrete time model of available any controlled process is estimated the controlled variable and the state variable in the moment in future comprises Multi-Nominal Matrix model, convolution model (pulse or step response), pulsed transfer function, state-space model or time series models (CARMA or ARMAX); On being difficult, during strict mathematical model, estimate by structure, response time and the steady-state characteristic of controlled process.

Have the correction time domain, the future value of estimating is proofreaied and correct with the weighted mean of the difference of the state variable (containing controlled variable) of each sampling instant actual measurement in this time domain and model pre-estimating value;

The SINGLE PREDICTION PREDICTIVE CONTROL algorithm: the weighting quadratic performance index of deviate will be minimum between the discreet value after calibrated and its set-point in a certain moment in future (time domain estimated in title) is target to each controlled variable, and the calculating operation variable is at the adjustment amount of each control cycle;

(controlled variable reaches the multiple of its steady-state value to the response of performance variable step with estimating horizontal β to set priority by the requirement of controlled variable, 0＜β＜1), according to Model Calculation response index and correlation index, the pairing of selected controlled variable and performance variable, according to dynamic responding speed be set the different sampling periods, and then determine to estimate time domain and estimate step number accordingly, if estimate step number, should increase the sampling period greater than 25.Form and estimate time domain and multicycle Prediction Control strategy more;

Unified control system structure with three kinds of variable dynamic feedbacks: it is characterized in that:

Utilize the currency and the history value of all available real measured datas, feed back and feedover, comprise the feedback of three kinds of variablees: the feedback of controlled variable (surveying beyond the controlled variable) state variable, performance variable; With the feedforward of estimating that can survey interference, form multiloop dynamic feedback control system structure; See figure one for details.

Based on this architectural feature, can determine to estimate time domain and estimate step number accordingly by the steady-state characteristic and the response time of controlled device, the feedback matrix of selected each backfeed loop does not need the mathematical model of controlled device accurately, constitutes the robust predictor controller.

Be adapted to the control and the coordination of the object of multivariate, multiple goal requirement, it is characterized in that:

Be applicable to that two kinds of controls require: controlled variable is maintained the set point control on the set-point and controlled variable maintained Region control in the given area;

The quantity that is applicable to controlled variable and performance variable changes with object running environment, is less than, equals at performance variable or the control when transforming between the various situations of controlled variable; For this reason, press the size of controlled variable and its set-point or given area deviation, a plurality of zones are set, when the controlled variable discreet value is in zones of different, adopt different control strategies or weighting coefficient (nonlinear Control or nonlinear weight).

When performance variable is less than controlled variable, three kinds of Real-time and Dynamic coordination approachs are arranged: controlled variable is estimated the deviation weighting, selects satisfied difference to control the intermediate value of the performance variable that requires or implemented step control.

When performance variable during more than controlled variable, consistent when the Optimizing operation variable with the action direction of control controlled variable, control by the priority orders selection operation variable of optimizing; When control and requiring of optimizing are contradictory,, select for use the fast performance variable of response to reduce deviation rapidly if the controlled variable discreet value is in the large deviation district; If the controlled variable discreet value is in little deviation district, select for use the slower performance variable of response to keep little deviation, simultaneously, push its optimal value to the performance variable that given pace will need to optimize.

Situation according to controlled device, the variation of residing zone of controlled variable discreet value and controlled variable quantity, the variation of validity of performance variable (whether limited or do not allow to use) and respective numbers, automatically select different controlled variables and performance variable for use, coordinate in the following order: each variable is not transfinited, make controlled variable reach given, make performance variable reach optimization;

Adapt to the method that the controlled device characteristic changes

Online in real time is revised control algolithm, adapts to the performance variable variation of pure retardation time.

Adjustment amount to each performance variable multiply by attenuation coefficient α, the main adjustment parameter during as on-line operation.

Size and absolute value thereof to the each adjustment amount of each performance variable are provided with the bound constraint, and the adjustment amount of performance variable can not make relative variable transfinite simultaneously.

Figure one has three kinds of variable dynamic feedbacks and can survey the general Prediction Control system chart that feedforward is estimated in interference;

Figure two is catalytic cracking riser reactor multivariable control system block diagrams;

Figure three is petroleum fractionating tower multivariate pre-estimating coordinating control system block diagrams;

Figure four is debutanizer control system block diagrams.

Below in conjunction with accompanying drawing, by embodiment, the present invention is described in further detail.

One, available various model, have the SINGLE PREDICTION PREDICTIVE CONTROL algorithm of proofreading and correct time domain: the present invention describes controlled device take following polynomial matrix model as the basis: $\left[\begin{array}{ccc}A\left(q^{-1}\right)& D\left(q^{-1}\right)& B\left(q^{-1}\right)\\ -C& 0& 0\end{array}\right]\·\left[\begin{array}{c}X\left(k\right)\\ -v(k-1)\\ -u(k-1)\end{array}\right]=\left[\begin{array}{c}0\\ -Y\left(k\right)\end{array}\right]----\left(1\right)$ Wherein: X (k) is the value of n dimension measured state variable (being called for short SV) k sampling instant;

U (k) is the value of m dimension performance variable (being called for short MV) k sampling instant;

Y (k) is the value of r dimension controlled variable (being called for short CV) k sampling instant;

V (k) is the value of g dimension measurable disturbances variable (being called for short DV) k sampling instant;

  q ^-1Be hysteresis operator, that is: q^-NX(k)=X(k-N)； $A\left(q^{-1}\right)=I+A_1{q}^{-1}+A+A_{N_{.1}}{q}^{{-N}_{.1}}$ $B\left(q^{-1}\right)=B_0+B_1{q}^{-1}+\mathrm{\Λ}+B_{N_R}{q}^{{-N}_R}$ $D\left(q^{-1}\right)=D_0+D_1{q}^{-1}+\mathrm{\Λ}+D_{N_d}{q}^{N_D}$

Be respectively n * n, n * m, n * g polynomial matrix; C is r * n constant matrices. Consider that each MV has different time lags, can be expressed as: $u(k-1)=\left[\begin{array}{c}{u}_1(k-1-{\mathrm{\τ}}_1)\\ u_2(k-1-{\mathrm{\τ}}_2)\\ M\\ u_m(k-1-{\mathrm{\τ}}_m)\end{array}\right]$

By model (1), can by each variate-value in the current and former moment, calculate the following constantly discreet value [containing the discreet value of controlled variable] of controlled device state variable; Also can be calculated by the variable in past the discreet value of current state variable. General Prediction Control algorithm is poor with the currency of surveying state variable and discreet value, and the discreet value take revised discreet value as practical application is revised in discreet value constantly in future. Characteristics of the present invention are with the correction time domain (N that sets₀To N_CSampling instant) weighted average of difference carries out on-line correction between interior each sampling instant measured value and the discreet value. To j controlled variable at following P_jDiscreet value is after the correction of individual sampling instant: Y_j(k+P _j)=Y _j ⁰(k+P _j)+S _j(P _j)·Δu(k)                          (2) $Y_j^0(k+P_j)=Y_j\left(k\right)+\underset{n=N_0}{\overset{N_C}{\mathrm{\Σ}}}{\mathrm{\λ}}_n\{F_{\mathrm{xjn}}\left(q^{-1}\right)\[X\left(k\right)-X(k-n)\]$

    +F _ujn(q ^-1)Δu(k-1)+F _vjn(q ^-1) Δ v (k) (3) wherein: Δ u (k)=u (k)-u (k-l)

  Δv(k)=v(k)-v(k-l)

  S _j(P _j)=[S _j1(P _j)S _j2(P _j)ΛS _jm(p _j)]

S _ji(P _j) be j CV to the step response of i MV at P_jThe value of individual sampling instant.

F _xjn(q ^-1),F _ujn(q ^-1) and F_vjn(q ^-1) being the multinomial vector of q-1, can be calculated by Diophantine equation mode by (1) formula.

λ is the weighting weight coefficient to the difference of each sampling instant actual measurement state variable and discreet value, and: λ_n0+λ _n1+Λ+λ _nc=1

The discrete time model that another characteristics of the present invention are available following any actual controlled devices that obtain carries out the calculating of above-mentioned discreet value: 1. polynomial matrix or pulsed transfer function shown in (1) formula (can be converted into the polynomial matrix model) input and output (discrete convolution) model of 2. surveying: $Y\left(k\right)=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}H\left(i\right)u(K-i)$ Wherein H (i) is impulse response coefficient. Then: $A\left(q^{-1}\right)=I,B\left(q^{-1}\right)=\underset{l=1}{\overset{N}{\mathrm{\Σ}}}H\left(i\right)q^{-1},C=I,S_j\left(i\right)=\underset{l=1}{\overset{i}{\mathrm{\Σ}}}{H}_j\left(l\right),F_{\mathrm{xj}}\left(q^{-1}\right)=0$ 3. input and output time series (CARMA or ARMAX) model: A (q^-1)Y(k)=B(q ^-1) u (k-1) then: C=I, X=Y (only having MV and CV dynamical feedback) is the discrete time state-space model 4., all states can be surveyed: X (k+1)=GX (k)+Hu (k) Y (k)=CX (k)

Then: A (q^-1)=I-Gq ^-1, B=H (constant matrices), M (q^-1The simple model in)=0 5.: controlled device structure, response time and steady-state characteristic, see second section for details.

On the basis of above model pre-estimating, implement SINGLE PREDICTION PREDICTIVE CONTROL (on-line optimization) algorithm, be another characteristics of the present invention. Its characteristics are: make each CV at following P_jThe discreet value Y of individual sampling instant (thus much moment)_j(k+P _j) and its set-point Y_j ^S(k+P _j) between the quadratic performance index of deviation minimum, calculate MV at the adjustment amount of each control cycle:

When m＜r, obtain following control algolithm:

Δu(k)=[S ^T(P)·W·S(P)] ^-1S ^T(P)·W·(Y ^S-Y ^O)       (5a)

Wherein: W=diag[Wj] Wj is the weight coefficient to j CV.

When m=r, obtain following control algolithm:

Δu(k)=S ^-1(P)·(Y ^S-Y ^O)???????????????????????????(5b) $S\left(P\right)=\left[\begin{array}{c}{S}_1\left(P_1\right)\\ S_2\left(P_2\right)\\ M\\ S_r\left(P_r\right)\end{array}\right]Y^S=\left[\begin{array}{c}{Y}_1^S\\ Y_2^S\\ M\\ Y_r^S\end{array}\right]Y^0=\left[\begin{array}{c}{Y}_1^0(k+P_1)\\ {\mathrm{<figure-callout\; id="2"\; label="Y"\; filenames="A9910554600261.png,A9910554600271.png"\; state="\{\{state\}\}">Y</figure-callout>}}_2^0(k+P_2)\\ M\\ Y_r^0(k+P_r)\end{array}\right]$

Δ u (k) is the adjustment amount of each control cycle MV.Two, the design of controller: the territory is estimated with the multicycle and is controlled for a long time

To multivariable controlled device, satisfy dynamic response and the interdependent characteristics of requirement, adaptation CV and MV to each CV and MV, guarantee the stable and performance of system, form and estimate time domain and multicycle control more, be another characteristics of the present invention.The method of determining to estimate time domain and control cycle is as follows: 1. the demanding CV of pair control performance is provided with higher priority.Weigh priority with " estimating horizontal β " [the corresponding time domain of estimating is that step response reaches its steady-state value β time doubly first, generally selects β=0.2-0.8], the high person of priority should select less β value for use.To r CV, the corresponding level of estimating is a vector: β=[β ₁, β ₂, Λ β _r] 2. determine the CV-MV pairing and estimate time domain by " response index " and " correlation index ":

To j CV, selecting and estimating level is β _j, this CV reaches this to the step response of each MV and estimates level institute and a corresponding sampling period be respectively R _J1, R _J2, Λ, R _Jm

To all r CV with estimate level accordingly, all can obtain the corresponding sampling period, be called " response index "; Constitute response index battle array RA by these response indexs: $\mathrm{RA}=\left[\begin{array}{cccc}{R}_{11}& {\mathrm{<figure-callout\; id="12"\; label="R"\; filenames="A9910554600261.png"\; state="\{\{state\}\}">R</figure-callout>}}_{12}& \mathrm{\&Lambda;}& R_{1m}\\ R_{21}& R_{22}& \mathrm{\&Lambda;}& R_{2m}\\ M& M& O& M\\ R_{r1}& \mathrm{\&Lambda;}& \mathrm{\&Lambda;}& R_m\end{array}\right]$ Definition " correlation index " is: ${\mathrm{\&mu;}}_{\mathrm{ij}}=\frac{\underset{l=1,l\&NotEqual;i}{\overset{P}{\mathrm{\&Sigma;}}}\left|S_{\mathrm{ji}}\left(\mathrm{Rij}\right)\right|}{\left|S_{\mathrm{ij}}\left(R_{\mathrm{ij}}\right)\right|}$ Wherein: S _Ij(R _Ij) be corresponding to R _IjStep-response coefficients.

i=1,2,Λ,r；???j=1,2,Λ,m

The response index value is more little, illustrates that this CV-MV pairing response is fast more.Correlation index is more little, illustrates that this CV-MV pairing is more little to the influence of its dependent variable.The principle of pairing is to make response the fastest, and is related minimum.But the two contradiction can occur, for this reason, i controlled variable is established weighting coefficient W _i, and ask for: $\underset{i}{\mathrm{Min}}\&lsqb;W_i{\mathrm{\&mu;}}_i+R_{\mathrm{ij}}\&rsqb;----\left(6\right)$ The corresponding j of institute, thus determine pairing selection CV _i-MV _j, correspondingly can determine estimating time domain, estimating step number and corresponding step response battle array S (P) of each CV.3. whether the check pairing satisfies the necessary condition of system stability: $\frac{\mathrm{Det}\&lsqb;S\left(P\right)\&rsqb;}{\mathrm{Det}\&lsqb;G_s\&rsqb;}>0----\left(7\right)$ Wherein: S (P) be select estimate the corresponding step response matrix of step number institute;

G _sIt is the steady-state gain matrix of controlled device;

Det represents to get determinant.

If selectedly estimate the requirement that step number can not satisfy system stability, then can increase the level of estimating, make it stable.

To larger controlled device, can be divided into several subsystems by relevance, the pairing selection of CV-MV is identical with said method in each subsystem.4. estimate time domain and many sampling periods more:

When the dimension (quantity) of CV and MV is identical, then has only a kind of CV-MV combinations of pairs.To r CV, then there be r to estimate time domain.

When the dimension (quantity) of MV during more than CV, multiple combinations of pairs (being made as the L kind) then can be arranged, correspondingly, there be r * L to estimate time domain.

By above method, form territory Prediction Control for a long time naturally.

Estimate step number P _IjShould not surpass 25, to reduce workload in line computation; If to estimate step number excessive, should increase control cycle, make and estimate step number and reduce, form multicycle control, to adapt to large scale system or the bigger system of different MV response time difference.Three, multiloop control and general robust predictor controller

(5) formula is the basic control method that the present invention provides, because X=[Y Z] ^T, Y is a controlled variable, Z is the measured state variable beyond the controlled variable.The control corresponding system architecture is shown in figure one.Figure one expresses another characteristics of the present invention: have dynamic feedback, the performance variable MV of controlled variable CV (Y) dynamic feedback, measured state variable (Z) dynamic feedback multiloop control and can survey the dynamic Feedforward of interference (V).

F is passed through in feedback and feedforward respectively _y(q ^-1), F _u(q ^-1), F _x(q ^-1) and F _v(q ^-1) realize that its essence is the measured value that not only utilizes current time, also utilizing in the past, the measured value in the moment feeds back and feedovers.

In the time can not accurately obtaining the controlled device mathematical model, can utilize the architectural feature of the Prediction Control system that has three kinds of variable feedbacks shown in the figure one, selected estimate level after, by the steady-state characteristic and the response time of controlled device, determine to estimate time domain and estimate step number accordingly, the feedback matrix of selected each backfeed loop does not need the mathematical model of controlled device accurately, constituting the robust predictor controller, is another characteristics of the present invention.Its algorithm is as follows: $\mathrm{\&Delta;u}\left(k\right)=\frac{\mathrm{\&alpha;}}{A_0}\{Y^S-Y\left(k\right)-F_x\&lsqb;X\left(k\right)-X(k-P)\&rsqb;-\underset{i=1}{\overset{P}{\mathrm{\&Sigma;}}}\&lsqb;S\left(P\right)-S\left(i\right)\&rsqb;\mathrm{\&Delta;u}(k-i)\}---\left(8\right)$ Wherein: Y ^S, Y (k), X (k), X (k-P), Δ u (k), S (P) meaning is ditto known: G ₀={ g _Ij}: controlled variable is to the steady-state gain (enlargement factor) of MV

G _x={ k _Ij}: controlled variable is to the measured state variable steady-state gain of (containing controlled variable)

t _Ij: i CV is to the response time (reaching for 95% time) of j MV

τ _Ij: i CV be to the requirement to controlled variable of several certificates of pure retardation time in corresponding sampling period of j MV, selected estimate horizontal β i (i=1,2, Λ, r)

Selected CV-MV pairing with same index

The step number of estimating to i CV is P _i, then: $S_{\mathrm{ii}}\left(P_i\right)={\mathrm{\&beta;}}_i\&CenterDot;g_{\mathrm{ii}}\underset{i+j}{S_{\mathrm{ij}}}\left(P_i\right)=(1-\frac{t_{\mathrm{ij}}-{\mathrm{\&tau;}}_{\mathrm{ij}}}{p_i})g_{\mathrm{ij}}---\left(9\right)$

(i=1,2,Λr???j=1,2,Λm) $S_{\mathrm{ij}}\left(l\right)=\frac{S_{\mathrm{ij}}\left(p_i\right)}{p_i}\mathrm{l\; l}=\mathrm{1,2},\mathrm{\&Lambda;},P,$

S _Ij(l) be equivalent to the controlled device rank and get over response coefficient.In case of necessity, can revise this result of calculation: suitably reduce hour S of l _Ij(l) numerical value is to strengthen the robustness of control.

A ₀=det[S(P)]

α=[α ₁, α ₂, Λ α _m] ^T, be the control attenuation coefficient.General optional α=0.1-1.0 β _iNumerical value is big more, α _iNumerical value is more little, and robustness is good more, and the control action increment is more little, responds slow more.

F _xBe m * n dimension feedback of status battle array:

F _x=diag[δ _i]G _x

δ wherein _i(i=1,2, Λ r) is adjustable parameter, suggested range 0.2-0.8.

By above method, only need have structure, response time and the steady-state characteristic of controlled device, get final product controlled performance and adjustable " the robust predictor controller " of robustness, and do not rely on strict controlled device mathematical model.Four, dynamically control and the real time coordination of optimizing requirement

The present invention considers that two kinds of controls of the object that multivariate, multiple goal require require: controlled variable CV is maintained the set point control (being called for short SCV) on the set-point and controlled variable maintained Region control (being called for short ZCV) in the given area.Require by the optimization of considering two aspects: make CV reach optimal value and make part MV reach optimal value.

The present invention also considers the structural change that controlled device causes because of environmental change, the quantity that is controlled variable is because of whether satisfied control requires to change, performance variable is tied because of itself, correlated variables (hereinafter to be referred as RV) is tied) or do not allow to use the number change that takes place, form controlled device and be less than, equal at performance variable or more than the conversion between the various situations of controlled variable.Satisfy two kinds of control requirements and two kinds of optimization requirements in the case.

By dynamic control, change MV, be to make CV reach the method for optimal value.Obviously, this with make MV reach optimal value may be contradictory.Optimization to dynamic control and MV realistic (with) time coordinates, and comprises the coordination of (overall situation) between the coordination and subsystem of sub-control system self.Be characteristics of the present invention.

The principle of coordinating is: guarantee that at first each variable does not transfinite or not super constraint; Secondly, guaranteeing that CV reaches under the condition of optimization (given) value substantially, make MV reach optimal value.

Coordinate requirement for adapting to, the present invention adopts following method: 1. couple CV carries out the multizone setting

1. little deviation district: in the time of in the SCV discreet value is in little deviation district, MV is regulated, make it progressively reach optimal value.In the time of in the ZCV discreet value is in little deviation district, can it not controlled, the quantity of CV is changed.

2. large deviation district: in the time of in the SCV discreet value is in the large deviation district, select response MV fast, make SCV get back to set-point as early as possible; In the time of in the ZCV discreet value is in the large deviation district, tackles it and control.

3. bound district: when the CV discreet value is transfinited, select strong control action, it is not transfinited.

In case of necessity more zone can be set.2. dynamic coordinate optimization:

As effective MV during more than CV:

1. when optimization is consistent with the action direction of control, select MV to control by the priority orders of optimizing;

2. when control and requiring of optimizing are contradictory,, select for use the fast performance variable (usually can make this MV depart from its optimal value) of response to reduce deviation rapidly if when the CV discreet value is in the large deviation district; When the CV discreet value is in little deviation district, keep little deviation with responding slower MV, simultaneously, will need the MV that optimizes to push its optimal value to given pace.

When effective MV is equal with CV quantity, calculate the adjustment amount of required MV with (5) formula.

When effective MV is less than the CV that needs control, provide following three kinds of coordination strategies:

1. to (CV discreet value and its set-point) deviation nonlinear weight, calculate the adjustment amount of MV by (5) formula; Weighting coefficient changes with the deviation of CV, in the time of in the CV discreet value is in little deviation district, and weighting coefficient minimum (can be zero); In the time of in the CV discreet value is in the large deviation district, weighting coefficient increases with deviation.

2. select the intermediate value of MV: selected Cv with effective MV equal number is set point control, calculates the adjustment amount of required MV by (5) formula; All the other CV by ZCV control requirement, are limited to set-point with its zone (up and down), calculate the adjustment amount of required MV, thus, can calculate three adjustment amounts, select its intermediate value to implement control each MV by (5) formula.

3. control is coordinated in classification: set point control as subordinate's sub-control system, and Region control is as higher level's sub-control system, when estimating the Region control controlled variable and transfiniting, by the set-point of higher level's subsystem adjustment subordinate subsystem.The higher level of the also a plurality of subordinates of higher level's subsystem subsystem constitutes the global coordination system, if effectively the quantity of MV is zero, provides subsystem failure information, carries out global coordination.

3. state is judged and decision-making in real time: in each controlling of sampling constantly, calculate the variation of residing zone of controlled variable discreet value and controlled variable quantity, the variation of validity of detecting operation variable (whether limited or do not allow to use) and respective numbers in real time, automatically select the pairing of different controlled variables and performance variable for use, coordinate in the following order: each variable is not transfinited, make controlled variable reach given, make performance variable reach optimization.Five, to the self-adaptation of controlled device characteristic variations

1. adapt to the variation of controlled device state: select the CV-MV pairing automatically, estimate time domain and estimate step number, corresponding control strategies accordingly, satisfy different CV control and require and different effective MV, and may push optimal value to the MV of necessity.

2. the adjustment amount of each MV that (5) formula is calculated multiply by control attenuation coefficient α, the main adjustment parameter during as on-line operation; Generally speaking, 0＜α＜1, α is more little, and the robustness of control system is high more, and the actual adjustment amount of MV is more little, to adapt to the variation of controlled device characteristic.

3. the size (absolute value) to the adjustment amount of each MV is provided with bound and speed limit, and adjustment amount must not surpass limit value.

4. the variation of online adaptive MV retardation time: the MV adjustment amount that (5) formula provides is to be foundation with the model under the controlled device rated condition.As the pure retardation time τ of controlled device CV to each MV _iWhen changing, to discreet value Y in (5) formula _j ⁰(j=1,2, Λ r) carries out online correction.Method is as follows:

The state-space model that 1. ought have controlled device, and when all state can be surveyed, $Y_j^0=Y_j\left(k\right)+C_j{A}^{P_{j0}}{A}^{{\mathrm{\&tau;}}_j}\&lsqb;X\left(k\right)-X(k-P_{j0}-{\mathrm{\&tau;}}_j)\&rsqb;$ $+\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{L=1}{\overset{{\mathrm{\&tau;}}_i}{\mathrm{\&Sigma;}}}{S}_{\mathrm{ji}}(P_j+L)\mathrm{\&Delta;}{u}_i(k-L)$ $+\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{L=1}{\overset{{\mathrm{\&tau;}}_i}{\mathrm{\&Sigma;}}}\&lsqb;S_{\mathrm{ji}}(P_j+{\mathrm{\&tau;}}_i)-S_{\mathrm{ji}}(L+{\mathrm{\&tau;}}_i)\&rsqb;\mathrm{\&Delta;}{u}_i(k-{\mathrm{\&tau;}}_i-L)$ $+\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{L=1}{\overset{{\mathrm{\&tau;}}_i}{\mathrm{\&Sigma;}}}\&lsqb;S_{\mathrm{ji}}(P_j+{\mathrm{\&tau;}}_i)-S_{\mathrm{ji}}(P_j+L)\&rsqb;\mathrm{\&Delta;}{u}_i(k-P_j-L)\}---\left(10\right)$ Wherein: P _J0=P _j-τ _j

2. when not possessing above-mentioned condition, only to F _Ujn(q ^-1)=[F _Ujn1(q ^-1) F _Ujn2(q ¹) Λ F _Ujnm(q ^-1)]] $F_{\mathrm{ujni}}\left(q^{-1}\right)=+\underset{L=1}{\overset{P_j-{\mathrm{\&tau;}}_i}{\mathrm{\&Sigma;}}}\&lsqb;S_{\mathrm{ji}}(P_j+{\mathrm{\&tau;}}_i)-S_{\mathrm{ji}}(L+{\mathrm{\&tau;}}_i)\&rsqb;\&CenterDot;\mathrm{\&Delta;}{u}_i(k-L-{\mathrm{\&tau;}}_i)$ $+\mathrm{\&Sigma;}{S}_{\mathrm{ji}}(P_j+L)\&CenterDot;\mathrm{\&Delta;}{u}_i(k-L)+\underset{L=1}{\overset{N_{\mathrm{Bji}}}{\mathrm{\&Sigma;}}}{M}_{\mathrm{ji}}\left(L\right)\mathrm{\&Delta;}{u}_i(k-L-{\mathrm{\&tau;}}_{i0})$ Carry out on-line correction.Be without loss of generality, establishing and proofreading and correct time domain is N ₀=N _c=P _j, then:

Binomial changed with the actual hysteretic time before this equation the right, and the 3rd is (the τ that does not change with the actual hysteretic time _J0Being used retardation time of model, is constant).

Embodiment one: the control of catalytic cracking riser reactor

Catalytic cracking riser reactor is under the effect of high temperature and catalyzer, is the chemical reactor of lightweight oil (petroleum gas, liquefied petroleum gas (LPG), gasoline, diesel oil etc.) with the mink cell focus cracking.For reactor is operated steadily, be typically provided with riser outlet temperature pi controller TC, terminator flow proportional integral controller FC.

The block scheme of the control system of riser reactor pre-estimating coordinating shown in the figure two, wherein: T ₁, T ₂Be respectively riser reactor middle part and outlet temperature,

F ₀, T ₀Material flow and temperature, F _C, T _CCatalyst flow and temperature,

F ₃, T ₃Terminator flow and temperature, V: variable valve (guiding valve) valve position

FC: terminator flow proportional integral controller,

TC1: riser outlet temperature pi controller,

TC2: the raw material preheating temperature pi controller,

This control system is to be controlled variable in the reaction depth of line computation and actual measurement temperature of reaction, to be given as performance variable (MV1 and MV2) with two pi controllers, use universal model pre-estimating coordinating controller provided by the invention, the control of realization response device.1. controlled device and variable are divided

The fundamental purpose that chemical reactor is controlled is to keep reaction depth steadily optimizing on the definite value, here (the unit charging is in reaction time institute's heat requirement with the reaction heat that calculates in real time, kJ/kg) weigh reaction depth and (see the patent of having declared for details, 90108193.0), with reaction heat as main controlled variable (set point control), with temperature of reaction as another controlled variable.

The variable of native system is divided as follows:

Sequence number	???CV	????MV	????RV	?FBV	?FFV
						????1	Reaction heat (S)	Temperature of reaction PI is given	The sliding position cunning to be generated of cutting down of regeneration is cut down the position	Riser middle part temperature	The feedstock oil flow
????2	Temperature of reaction (Z)	PI is given for the terminator flow	The position is cut down in adjusting	Raw material preheating temperature	Regenerated catalyst temperature
						????3		Preheat temperature PI is given	The position is cut down in threeway

RV is the variable relevant with MV, and MV and RV are equipped with bound and speed limit;

FBV (for surveying beyond the CV) state variable feedback, FFV is a feed forward variable.2.CV-MV pairing, control cycle and estimate the step number native system may and reasonably CV-MV be paired into:

1. reaction heat (CV1)-temperature of reaction PI given (MV1)

Temperature of reaction (CV2)-terminator flow (MV2)

2. reaction heat (CV1)-temperature of reaction PI given (MV1)

Temperature of reaction (CV2)-preheat temperature PI given (MV3) 3. controls and coordination approach

CV1-MV1 and CV2-MV2 pairing have response faster, are first-selected controlling schemes, can adopt identical control cycle, and the level of estimating accordingly all adopts β=0.5.

The response of CV1-MV1 and CV2-MV3 is very slow, only in the limited use of MV2, or CV2 when being in the large deviation district and the scheme of 1. planting use simultaneously; Adopt long control cycle and the bigger level of estimating.

Above CV is identical with MV quantity, directly calculates the adjustment amount of MV with (5) formula.

When MV2 and MV3 were all limited, MV was less than CV, and a kind of coordination of selecting in following three kinds of methods arranged: calculate MV1 by following three kinds of requirements 1.: 1. reach its set-point and calculate by CV1, MV1 (1); 2. reach its large deviation higher limit by CV2 and calculate, get MV1 (2); 3. reach its large deviation lower limit by CV2 and calculate, get MV1 (3); Select MV1 (1), MV1 (2), the intermediate value of MV1 (3) is controlled.2. to estimating the deviation nonlinear weight: when the CV2 discreet value is in little deviation district, W1=1, W2=0; When the CV2 discreet value is in the large deviation district, W1=1-ABS (E2)/B, W2=ABS (E2)/B; E2 is a deviate, and B is the deviation sector width.When ABS (E2)＞1, W1=0, W2=1.3. step control: with the reaction heat Prediction Control is subordinate, and the temperature of reaction Prediction Control is a upper level, when estimating temperature of reaction and transfinite, adjusts the reaction heat set-point and coordinates.Embodiment two: the control of petroleum fractionating tower

The petroleum fractionating tower is the process units that various products or semi-manufacture in the petroleum fraction is separated into specification product, it utilizes each product boiling spread difference in the petroleum fraction, is heated vaporization earlier, utilizes the backflow heat-obtaining then, in tower, form sufficient mass-and heat-transfer, product is separated.General heated raw material at the bottom of tower or the bottom of tower enter fractionator, the gas phase petroleum fraction moves upward, and carries out mass-and heat-transfer through the multilayer column plate, heavier product distillates from the tower middle and lower part, lighter product distillates from top and each side line of middle and upper part of fractionator.Consumed energy is all wanted in heating and backflow, and the heat-obtaining recoverable energy reduces energy consumption; Recover energy, obviously be a controlled target of petroleum fractionating tower more; But to the control of petroleum fractionating tower, topmost target is to guarantee that product is qualified, and steadily edge can bring bigger benefit.Need the two is carried out real time coordination.

Figure three is certain petroleum fractionating Tower System flow process and control system block diagram thereof.Wherein:

TC: temperature PI controller

FC: flow PI controller

LC: liquid level PI controller

MV: performance variable, T: temperature point

E: heat interchanger or reboiler

: variable valve or T-valve

: process equipment or pipeline

The instrument line : conventional PI controller

The correlation analysis and the measured state variable that utilize the present invention to provide, the control of fractionator can be divided into subsystems such as control at the bottom of gasoline endpoint control, the control of diesel oil solidifying point, temperature of lower control, the tower, absorption tower control, stabilizer control, the measured state variable that indicates # number in the utilization in the table is as feedforward (FF) variable, can reduce interrelated between each subsystem, reach the same effect of controlling with a big system.

Native system is implemented on the basis of conventional PI control system, and MV is the set-point of conventional PI controller.

1. variable and subsystem are divided as follows:

Sequence number	Subsystem	????CV	????MV	?????RV	?????FBV	????FFV
							????1	The gasoline endpoint control subsystem	Gasoline endpoint (S)	It is given that flow FC201 is followed on the top	A * is cut down in threeway	The top is followed and is returned the tower temperature	Middle part temperature #
????2	Tower top temperature (Z)	Push up the given TC201 of warm PI	The top is followed flow and is cut down the position	The top is followed and is extracted temperature under the plate out	Rich absorbent oil heat-obtaining amount
						????3			Cold flow FC202 is given on the top	Push up the cold position of cutting down
????4	The control of diesel oil solidifying point	Diesel oil solidifies (90%) point	In warm PI TC202 given	A * is cut down in threeway	Return the tower temperature in one								In one after boiling again temperature #
						????5	Middle part temperature (S)	Heat-obtaining flow FC203 in one	Circular flow cuts down the position in one	Extract temperature under the plate in one out	Tower temperature of lower #
????6		The control of tower temperature of lower	Temperature of lower (S)	Heat-obtaining circular flow FC204 is given in two	A * is cut down in threeway							Return the tower temperature in two	In two after # boils again temperature
	????7						Temperature of lower PI TC203 is given	Circular flow cuts down the position in two	Extract temperature under the plate in two out	Temperature # on the herringbone baffle plate
????8			Control and global coordination at the bottom of the tower	Column bottom temperature (Z)	Heat-obtaining circular flow FC207 is given at the bottom of the tower						The position is cut down in adjusting	The tower temperature is returned in circulation at the bottom of the tower	Material temperature
	????9	Liquid level at the bottom of the tower (Z)				The heat-obtaining threeway is cut down at the bottom of the tower	Circular flow cuts down the position at the bottom of the tower	Temperature on the herringbone baffle plate
???10				The freshening oil tank level	FC205 is given for the recycle stock flow					Recycle stock flow valve position
	???11	The main fractionating tower constraint				Extract flow FC204 at the bottom of the tower out	Extract the flow valve position at the bottom of the tower out
???11				The constraint of absorption steady component	FC101 is given for the feedstock oil flow					Feedstock oil flow valve position
	???12	Desorber control				The desorber column bottom temperature	Column bottom temperature PI TC302 is given	The position is cut down in threeway	The reboiler outlet temperature				Heating load # in one
Tower middle part, temperature of lower			The absorption tower feed rate
				???13	Stabilizer control				Stablize tower top temperature (C5)	Cat head cold reflux flow is given	Flow cuts down the position	The reboiler outlet temperature	Heating load # in two
???14	The stabilized gasoline vapour pressure	Column bottom temperature PI TC301 is given	The position is cut down in threeway			Tower middle part, temperature of lower	The stabilizer feed rate

Illustrate: * number expression MV has the requirement of optimization; For ease of saying something, be example with the gasoline endpoint control subsystem below; 2, general robust predictor controller

With the gasoline endpoint control subsystem is example, if can not obtain plant model accurately, can utilize the general robust predictor controller of following information structuring: the structure of (1) controlled device:

CV ₁: gasoline endpoint Y, CV ₂: tower top temperature T ₁

MV:MV ₁Top heat-obtaining circular flow, MV ₂Top cold reflux amount, MV ₃Circulation heat-obtaining T-valve; To MV ₁And MV ₃The optimization requirement is arranged, and optimal value is its lower limit; Optimization is in proper order: MV ₃Preferentially, MV ₁Take second place.

The feedback of status variable: the top is followed and is returned the tower temperature T ₂, the top follows and extracts temperature T under the plate out ₄(2) steady-state characteristic:

CV ₁To T ₂Steady-state gain be A ₁₂=0.8, to T ₄Steady-state gain be A ₁₃=0.9,

To MV ₂The steady-state gain of (T-valve aperture) is B ₁=0.1 ℃/%;

CV ₁To MV ₁Steady-state gain be 0.5 ℃/ton/time, so S (P ₁)=0.3 ℃/ton/time;

CV ₁To MV ₃Steady-state gain be 1.4 ℃/ton/time.So S (P ₃)=0.84 ℃/ton/time.(3) dynamic response:

CV ₁(gasoline endpoint) and CV ₂(tower top temperature) has identical dynamic perfromance

CV ₁To MV ₂(pushing up the given or T-valve of warm PI) response is slower, and about 12 minutes, other had pure-time-delay τ=1 minute, is N corresponding to the hysteresis step number in 30 second sampling period _d, estimate step number and be: P ₀=6, P ₂=N _d+ P ₀=8;

CV ₁To MV ₁(flow is followed on the top) and MV ₃(top cold reflux flow) response is very fast, about 8 minutes; Corresponding to 10 seconds sampling weeks, estimate time domain P ₁=P ₃=11; (4) robust predictor controller: with CV and MV ₂Be paired into example, referring to (7) formula

Select β=0.6, A ₀=S (P ₂)=β B ₁, following robust predictor controller is arranged:

ΔMV ₂(k)=α·S ^-1(P ₂){Y ^S-Y-[F ₁·(T ₁(k)-T ₁(k-p)+F ₂·(T ₂(k)-T ₂(k-p)) $+{\mathrm{<figure-callout\; id="3"\; label="F"\; filenames="A9910554600261.png,A9910554600271.png"\; state="\{\{state\}\}">F</figure-callout>}}_3\&CenterDot;(T_4\left(k\right)-T_4(k-p))\&rsqb;+\underset{i=1}{\overset{P}{\mathrm{\&Sigma;}}}\&lsqb;S\left(P_2\right)-S\left(i\right)\&rsqb;\mathrm{\&Delta;M}{V}_2(k-i)$ $+\underset{L=1}{\overset{\mathrm{\&tau;}}{\mathrm{\&Sigma;}}}\&lsqb;S({\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A9910554600261.png,A9910554600271.png"\; state="\{\{state\}\}">P</figure-callout>}}_2+L)\mathrm{\&Delta;M}{V}_2(k-L)+\underset{L=1}{\overset{\mathrm{\&tau;}}{\mathrm{\&Sigma;}}}\&lsqb;S({\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A9910554600261.png,A9910554600271.png"\; state="\{\{state\}\}">P</figure-callout>}}_2+\mathrm{\&tau;})-S({\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A9910554600261.png,A9910554600271.png"\; state="\{\{state\}\}">P</figure-callout>}}_2+L)\&rsqb;\mathrm{\&Delta;M}{V}_2(k-L-P)$ Wherein: F ₁=(0.2-1.0), F ₂=(0.2-1.0) A ₁₂, F ₃=(0.2-1.0) A ₁₃3.MV online in real time select and coordination optimization

Three possible CV-MV pairings are arranged, correspondingly, three kinds of control algolithms are arranged.1. CV1-MV1, action direction is "-"; 2. CV1-MV2, action direction is "+" 3. CV1-MV3, action direction is "-"

In each control constantly, all need the state according to controlled device, select the control corresponding algorithm, reach control and the requirement of coordinating, its step is as follows: 1. estimate the discreet value Y that the time domain online in real time is calculated CV1 and CV2 by maximum ₁ ⁰And Y ₂ ⁰2. judge Y ₁ ⁰And Y ₂ ⁰Zone of living in: judge 3. whether MV reaches bound (whether being in the optimization state); 4. select the CV-MV pairing by running status, adopt following IF-THEN method.IF:[Y ₁ ⁰＜Y ^S(set-point)

(MV is optimized with the control that reduces the CV deviation same function direction is arranged)

AND MV1 and MV3 all have been in optimal value (lower limit)

AND MV2 is (allow use, the not super upper limit, not standard-sized sheet of bypass is cut down in threeway) effectively] OR[Y ₁ ⁰〉=Y ^S(MV is optimized with the control that reduces the CV deviation different directions is arranged),

AND CV1 is in (Y in the little deviation district ₁ ⁰-Y ^S＜setting value)]

AND MV1 is (allow use, not super lower limit, not complete shut-down of bypass is cut down in threeway) effectively] THEN: select CV1-MV2 pairing and corresponding algorithm to control.ELSE IF:[Y ₁ ⁰＜Y ^SAND MV3 is not in optimal value (lower limit);

OR[Y ₁ ⁰〉=Y ^SAND CV1 does not locate (Y in the little deviation district ₁ ⁰-Y ^S＞setting value)

AND MV2 has reached upper limit AND MV3 effectively (not reaching the upper limit)]

THEN: select CV1-MV3 pairing and corresponding algorithm to control ELSE and select CV1-MV1 pairing and corresponding algorithm to control END IF Y ₂ ⁰〉=the upper limit, THEN Δ MV2 must not be greater than zero;

Δ MV1 and Δ MV3 must not be less than zero; IF Y ₂ ⁰≤ lower limit, THEN Δ MV2 must not be less than zero;

Δ MV1 and AMV3 must not be greater than zero; IF ABS (Δ MV) 〉=speed limit, THFN ABS (Δ MV)=speed limit.Embodiment three: the control of debutanizer

The effect of debutanizer is that the butane in the raw material (C4) and lighter hydro carbons are separated, and distillates from the top of tower, and component heavier in the raw material is discharged at the bottom of tower.C5 content is lower than given standard in the requirement overhead, the also low and given standard of the butane content at the bottom of the tower in the draw-off.Usually at the bottom of cat head and tower, temperature pi controller TC1 is set respectively and TC2 controls debutanizer.

For improving separation accuracy, guarantee cat head and bottom product quality, can adopt model pre-estimating coordinating controller provided by the invention, on the basis of conventional tower top temperature pi controller TC1, column bottom temperature pi controller TC2, tower precompression pi controller PC, realize Advanced Control, shown in figure four.Wherein: 1: overhead condensation refrigeratory 2: return tank of top of the tower 3: tower bottom reboiler T ₁T ₂T ₃T ₄T ₅T ₆: actual measurement debutanizer each point temperature T C1:: tower top temperature pi controller TC2: column bottom temperature pi controller PC: tower top pressure pi controller F ₀T ₀: material flow and temperature F ₂Gas flow P, L: liquid level F at the bottom of tower top pressure and the tower ₁T ₆T ₅: the heat carrier flow and the temperature A that give the reboiler heat supply ₁A ₂: at the bottom of the tower with overhead product quality (C ₅With butane C ₄Content) analyser or observer debutanizer control system argument table

Sequence number	????CV	????MV	????RV	?FBV	????FFV
						?1	Bottom product C 4Content A 1	TC2 is given	Control valve opening	?T 2	Feed rate F 0
?2	Column bottom temperature T 4??(Z)	TC1 is given	Control valve opening	?T 3	Feeding temperature T 0
						?3	Overhead product C 5Content A 2	PC is given	Control valve opening	?T 5	Heat carrier flow F 1
?4	Tower top temperature T 1??(Z)		Liquid level L at the bottom of the tower		The thermal barrier temperature T 6
						?5	Tower top pressure (Z)		Gas flow F 2

IF[Y ₁ ⁰〉=Y ^S(MV is optimized with the control that reduces the CV deviation different directions is arranged),

AND CV1 is in (Y in the little deviation district ₁ ⁰-Y ^S＜setting value)]

AND MV2 is (allow use, not super lower limit, not complete shut-down of bypass is cut down in threeway) effectively]

IF MV3 does not reach optimal value, then pushes MV3 to optimal value by given pace;

IF MV3 has reached optimal value, and MV1 does not reach optimal value, then pushes MV1 to optimal value by given pace.

IF MV1, MV2, MV3 all lost efficacy, and then asked upper level (overall situation) control to be coordinated.

Example: MV2 reaches lower limit (top is followed threeway and cut down the bypass contract fully), MV1 and MV3 reach the upper limit, can reduce MV12 (feed rate), increase tower bottom heat-obtaining amount (reducing tower temperature of lower set-point) or increase tower at the bottom of heat-obtaining amount (close threeway is cut down bypass or increased tower at the bottom of the tower at the bottom of circular flow)

Claims (3)

1. universal multi-variable quantity model pre-estimating coordinating control method is used in particular for the multivariate pre-estimating coordinating control method based on mathematical model of continuous flow procedure, it is characterized in that:

Available any controlled process discrete time model: comprise the universal method of the variation in the moment in future being estimated with Multi-Nominal Matrix model, convolution model (pulse or step response), pulsed transfer function, state-space model or time series models (CARMA or ARMAX); On being difficult during model, the method for estimating by structure, response time and the steady-state characteristic of controlled process;

Have the correction time domain, discreet value is proofreaied and correct with the weighted mean of the difference of each sampling instant actual measurement state variable (containing controlled variable) and model pre-estimating value in this time domain;

The SINGLE PREDICTION PREDICTIVE CONTROL algorithm: according to each controlled variable in following certain discreet value of (time domain estimated in title) constantly and the weighting quadratic performance index minimum of the deviate between its set-point, the calculating operation variable is at the adjustment amount of each control cycle;

(controlled variable reaches the multiple of its steady-state value to the response of performance variable step with estimating horizontal β to set priority by the requirement of controlled variable, 0＜β＜1), according to Model Calculation response index and correlation index, the pairing of selected controlled variable and performance variable, according to dynamic responding speed be set the different sampling periods, and then determine to estimate time domain and estimate step number accordingly, if estimate step number greater than 25, should increase the sampling period, form and estimate time domain and multicycle Prediction Control strategy more;

Three kinds of variable dynamic feedback: utilize all available real measured data currencys and history value to feed back and feedover, comprise three kinds of variablees: the feedback of controlled variable, (surveying beyond the controlled variable) state variable, performance variable, can survey the feedforward of estimating of interference, form multiloop dynamic feedback control system structure;

Based on this architectural feature, by the steady-state characteristic and the response time of controlled device, to determine to estimate time domain and estimate step number accordingly, the feedback matrix of selected each backfeed loop does not need the mathematical model of controlled device accurately, constitutes the robust predictor controller;

Be applicable to that two kinds of controls require: controlled variable is maintained the set point control on the set-point and controlled variable maintained Region control in the given area;

The quantity that is applicable to controlled variable and performance variable changes with object running environment, be less than, equal at performance variable or the various situations more than controlled variable between when transforming, make controlled variable and performance variable all reach the control and the coordination of optimal value, for this reason, size by the deviation (estimating deviation) of controlled variable discreet value and its set-point or given area, a plurality of zones are set, when the controlled variable discreet value is in zones of different, adopt different control strategies or weighting coefficient (nonlinear Control or nonlinear weight);

When performance variable is less than controlled variable, three kinds of Real-time and Dynamic coordination approachs are arranged: controlled variable is estimated the deviation weighting, selects satisfied difference to control the intermediate value of the performance variable that requires or implemented step control;

,, control during at performance variable by the priority orders selection operation variable of optimization if the Optimizing operation variable is consistent with the action direction of control controlled variable more than controlled variable; When control and requiring of optimizing are contradictory,, select for use the fast performance variable of response to reduce deviation rapidly if the controlled variable discreet value is in the large deviation district; If the controlled variable discreet value is in little deviation district, select for use the slower performance variable of response to keep little deviation, simultaneously, push its optimal value to the performance variable that given pace will need to optimize;

Situation according to controlled device, the variation of residing zone of controlled variable discreet value and controlled variable quantity, the variation of validity of performance variable (whether limited or do not allow to use) and respective numbers, automatically select different controlled variables and performance variable for use, coordinate in the following order: each variable is not transfinited, make controlled variable reach given, make performance variable reach optimization;

Online in real time is revised control algolithm, adapts to the performance variable variation of pure retardation time.

2. method according to claim 1 is characterized in that: the adjustment amount to each performance variable multiply by attenuation coefficient α, the main adjustment parameter during as on-line operation.

3. method according to claim 1 is characterized in that: size and absolute value thereof to the each adjustment amount of each performance variable are provided with the bound constraint, and the adjustment amount of performance variable can not make relative variable transfinite simultaneously.

Images