Summary of the invention:
The objective of the invention is to provide a kind of strategy and method based on nonlinear dynamic mathematical model, the nonlinear model predictor control strategy that adapts to controlled process and equipment characteristic and formation multivariate variable structural nonlinear model predictor controller (being called for short VSNMPC), the not only variation of (non-linear) adaptation controlled process on model, also structurally adapt to the variation of controlled process, make control system have more vitality, provide optimum control structure and control effect at any time.
The variable structural nonlinear model predictor controller that the present invention provides mainly comprises following five part compositions: online in real time configuration device 1; Nonlinear model predictor control counter 2; Performance variable output processor 3; Controlled variable constrained optimization device 4; Performance variable constrained optimization device 5.When enforcement is of the present invention, also need the following support equipment that is connected with variable structural nonlinear model predictor controller: slip-stick artist interface 9, operation interface 8, real-time data base 9, dcs 10, controlled process or equipment 6.These support equipments are to possess in controlled process or the equipment a bit, and other can adopt existing products.
More than its annexation of listed ingredient see Fig. 1.The input end of online in real time configuration device 1 connects slip-stick artist interface 7, operation interface 8, real-time data base 9, dcs 10, output terminal connects nonlinear model predictor control counter 2, controlled variable constrained optimization device 4 and performance variable constrained optimization device 5 input ends, also attended operation variable output processor 3 input ends.Controlled variable constrained optimization device 4 output terminals connect nonlinear model predictor control counter 2 input ends, nonlinear model predictor control counter 2 output terminals connect control action output processor 3 input ends, performance variable constrained optimization device 5 output terminal attended operation variable output processors 3 input ends, performance variable output processor 3 output terminals connect dcs 10, dcs 10 connects controlled device 6, and nonlinear model predictor control counter 2 connects real-time data base 9 and slip-stick artist interface 7.By online in real time configuration device 1 variation of detection controlled process structure in real time, determine the only control system structure of current time according to the multivariate co-ordination principle; Calculate by current optimum structure and corresponding nonlinear model by nonlinear model predictor control counter 2 and to provide current control action increment, provide actual adjustment amount by control action output processor 3 again, realize Advanced Control controlled process; Simultaneously, online in real time configuration device 1 gives the current controlled variable that is optimized and behaviour does variable, implements tuning by controlled variable constrained optimization device 4, performance variable constrained optimization device 5 respectively.
Major technique characteristics of the present invention and theing contents are as follows:
1. based on nonlinear state spatial model with multiple time lag:
$\stackrel{.}{X}\left(t\right)=\frac{\mathrm{dX}\left(t\right)}{\mathrm{dt}}=F[X(t-{\mathrm{\τ}}_{A}),U(t-{\mathrm{\τ}}_{B}),V(t-{\mathrm{\τ}}_{F})]---\left(1\right)$
Y(t)＝G[X(t)，V(t)] (2)
Wherein: F, G are given functional vector
X ∈ R
^{n}(state variable SV) Y ∈ R
^{r}(controlled variable CV)
U ∈ R
^{m}(performance variable MV) V ∈ R
^{q}(can survey disturbance variable FV)
τ
_{A}, τ
_{B}, τ
_{F}Be respectively SV, MV, (be SV the retardation time of FV, MV, the function of FV) matrix is considered state variable, performance variable and can survey disturbance variable the influence of state variable is all had time lag, and be state variable, performance variable and the function that can survey disturbance variable retardation time, is the actual conditions of many controlled processes and equipment, also is one of principal feature of the present invention.
2. replace state variable constraint and multiple controlled variable setting with Region control
Except that the controlled variable of general indication, many state variables of controlled process need remain in the bound of permission.The present invention is converted to regional controlled variable with these variablees, when estimating it and will transfinite, it is implemented control, estimates when not transfiniting, not as controlled variable.Make the model pre-estimating control algolithm be converted into " not having constraint " shortcut calculation.
The present invention is divided into float area and two kinds of controlled variables of fixed area with regional controlled variable, and the bound of float area can be floated with other variable.The present invention also can be provided with " condition controlled variable ", promptly under certain condition just as controlled variable, to adapt to the demand that the controlled process multivariate is coordinated, can realize these control requirements by change structure control strategy provided by the invention.
3. monodrome is estimated nonlinear weight control
The present invention adopts following optimization aim to calculate current control action:
$\underset{\mathrm{\ΔU}\left(k\right)\∈{R}^{m}}{\mathrm{Min}}\left[J\right]:\underset{U\left(k\right)\∈{R}^{m}}{J}=\underset{i=1}{\overset{r}{\mathrm{\Σ}}}{E}^{T}\left(P\right)\mathrm{WE}\left(P\right)---\left(3\right)$
Wherein: P=[p
_{1}P
_{r}]
^{T}It is the time domain of estimating of each controlled variable
E
^{T}(P)=[e
_{1}(p
_{1}) ... e
_{r}(p
_{r})] be that controlled variable is estimated deviation
${e}_{i}\left({p}_{i}\right)={Y}_{i}^{s}-{Y}_{0i}^{c}(k+{p}_{i}/k)$
${Y}_{0i}^{c}(k+{p}_{i}/k)$
w
_{i}The weighting coefficient of=the i controlled variable
The characteristics of this algorithm are:
1. to each controlled variable, only calculate the optimum control effect with a discreet value in its following a certain moment (estimating time domain P);
2. only calculate variation delta U (k)=U (the k)-U (k-1) of the control action of current time, after Constraints Processing, carry out;
More than two characteristics this algorithm is simplified greatly than common MPC algorithm, the appropriate selection estimated time domain P, can obtain identical control effect.
3. m≤r, promptly performance variable can be less than or equal the number of controlled variable, by weight matrices W each variable is coordinated;
4. nonlinear weight: weighting coefficient changes with the deviation of controlled variable or near the degree of constraint limit, so that a plurality of controlled variables are coordinated, each variable is not transfinited.
5. morbid state is handled: suitably select weighting coefficient and performance variable, the variable (morbid state) that makes this algorithm permission simple crosscorrelation is simultaneously as controlled variable, and system still can normally move.
4. have feedback of status and the nonlinear model predictor control algolithm of estimating feedforward
The nonlinear state spatial model that utilizes (1) (2) formula to provide, the value of estimating the time domain moment controlled variable future is estimated and revised in real time, and then calculate the optimum control effect of estimating deviation and satisfying above-mentioned optimization aim (3), be the basic ideas of Prediction Control algorithm.At the characteristics of nonlinear model predictor control, the present invention controls the time of running at each, does following calculating:
1. calculate to determine each of τ retardation time
_{A}, τ
_{B}, τ
_{F}
2. calculate the currency of each state variable with the state observer method, comprise and to survey and can not survey state variable.
3. the state variable of calculating with current actual measurement state variable or observation that can not survey is an initial value, calculates the step response of each controlled variable to each performance variable, and estimates time domain β relatively according to what set
_{I, j}That determines the pairing of each controlled variable-performance variable estimates time domain p
_{I, j}
Provide and estimate the time domain valuation Y (k+P/k) of controlled variable constantly future.
Annotate: k+P/k represents that the state with current time k is an initial value, and following (k+P) value is constantly estimated.
4. calculate the relative step response matrix of current time:
S
_{i，j}(p
_{i})＝Y
_{i，j}(k+p
_{i}/k)-Y
_{0,i，j}(k+p
_{i}/k)
Wherein: Y
_{0, i, j}(k+p
_{i}It is initial value that //k) calculated (can not survey) state variable with current actual measurement state variable or observation, at current and following performance variable constantly with can survey under the condition that disturbance variable remains unchanged, each controlled variable that is calculated by (1) (2) formula is at k+p constantly in future
_{i}The time response.
5. the model pre-estimating value is done to estimate more the online in real time correction of time domain:
State when estimating time domain in the past with current time is an initial value, is estimated the value of calculating each controlled variable current time by (1) (2) formula:
Y
_{i}(k/k-p
_{i}), Y
_{i}(k-σ
_{i}/ k-p
_{i}-σ
_{i}), Y
_{i}(k+ σ
_{i}/ k-p
_{i}+ σ
_{i}) i=1,2 ..., r provides the mean value that the current time difference is estimated time domain CV discreet value:
$\stackrel{\‾}{{Y}_{i}}(k/k-{p}_{i})=\frac{{Y}_{i}(k/k-{p}_{i})+{Y}_{i}(k-{\mathrm{\σ}}_{i}/k-{p}_{i}-{\mathrm{\σ}}_{i})+{Y}_{i}(k+{\mathrm{\σ}}_{i}/k-{p}_{i}+{\mathrm{\σ}}_{i})}{3}$
Wherein: σ
_{i}Be the positive integer that to set.
The corresponding online trim amount of discreet value is:
$\mathrm{\δ}{M}_{i}={Y}_{i}\left(k\right)-\stackrel{\‾}{{Y}_{i}}(k/k-{p}_{i})$
Through i controlled variable of online feedback correction at (k+p constantly in future
_{i}) discreet value be:
Y
_{i} ^{c}(k+p
_{i}/k)＝Y
_{0i}(k+p
_{i}/k)+δM
_{i} (6)
6. calculate and estimate deviation E:
Set point is controlled:
${E}_{i}={Y}_{i}^{s}-{Y}_{i}^{c}(k+{p}_{i}/k)$
To Region control:
When:
${\mathrm{LLM}}_{i}<{Y}_{0i}^{c}(k+{p}_{i}/k)<{\mathrm{HLM}}_{i},$ E
_{i}＝0；
When: Y
_{i} ^{c}(k+p
_{i}/ k)〉HLM
_{i}, E
_{i}=Y
_{i} ^{c}(k+p
_{i}/ k)-HLM
_{i}
When: Y
_{i} ^{c}(k+p
_{i}/ k)＜LLM
_{i}, E
_{i}=Y
_{i} ^{c}(k+p
_{i}/ k)-LLM
_{i}
HLM
_{i}=upper limit LLM
_{i}=lower limit
7. calculate the weighting coefficient of each controlled variable of current time
W
_{i}＝W
_{iO}W
_{ie} (7)
Wherein: w
_{I0}=initial weighting coefficients, w
_{Ie}=with the weighting coefficient of estimating the deviation size variation
8. the objective function that provides according to (3) formula, calculate and provide current control action:
ΔU(k)＝U(k)-U(k-1)＝[S
^{T}(P)WS(P)]
^{-1}S
^{T}(P)WE (8)
More than (1)-(8) formula, be the imbody of algorithm characteristic of the present invention.These algorithms are finished by " 2. nonlinear model predictor control counter " among Fig. 1.
5. become structure-determine current controller architecture automatically by " 1. online in real time configuration device "
The variation of controlled variable and available action variable quantity is the importance that controlled process changes, for adapting to this variation of controlled process, by " online in real time configuration device ", realize becoming the structural model Prediction Control, be that its major function of another characteristics of the present invention is as follows:
1. constantly detect the controlled process operation conditions in each control, determine following content in real time:
CV: the controlled variable that needs (estimated deviation or exceeded little deviation district, possessed certain condition) and possible (non-fault and be allowed to);
MV: the performance variable of available (comprise and allow that use, trouble-free, and the not super upper limit of correlated variables RV or not super lower limit) own;
SV: the surveyed state variable that can be used as (trouble-free) feedback of status;
FV: the actual measurement that can do (trouble-free) feedforward is disturbed or the observation disturbance variable;
2. when controlled variable breaks down, withdraw from advanced person's control (becoming conventional PID control) or (when fault is arranged) automatically and carry out " degradation " (changing control system structure) or " upgrading " (after fault is eliminated, reverting to original control system structure) processing automatically.
3. determine the optimum controlled variable-performance variable pairing of current time: provide in above detection on the basis of information, control priority and the control priority of optimizing priority, performance variable and optimization priority according to controlled variable, constantly control system is carried out structural coordination in each control, determine the optimal control system structure and the multivariate coordinate scheme of current time, calculating for the online in real time of above-mentioned model pre-estimating control action provides the foundation of determining to estimate time domain, control system structure, controlled variable weighting coefficient.
Whether 4. definite current time needs and may carry out constrained optimization and coordination to controlled variable-performance variable.
5. provide the information that controlled variable does not have performance variable, can be by the order of setting, (but in no performance variable time spent) withdraws from advanced control automatically, or (after having the available action variable again) recovers advanced control automatically.
6. the constrained optimization of controlled variable-performance variable and coordination
1. the constrained optimization of controlled variable: when requiring certain controlled variable to reach given optimization range, provide the information that related variable does not transfinite by " 1. online in real time configuration device " (see figure 1), by " 4. controlled variable constrained optimization device " (see figure 1) with the set-point of the controlled variable given optimal value of tuning progressively.In case forecast has variable to transfinite or reaches given optimization range, tuning stops at once.When not carrying out tuning, if related variable forecast transfinites, and when not having other means, can adjust the set-point of optimised controlled variable, related variable is not transfinited.Characteristics of the present invention are to be provided with to optimize the zone, in the time of in controlled variable reaches the optimization zone, promptly no longer adjust, and prevent vibration.
2. the unsteady coordination of controlled variable:
The controlled variable set-point can float with other variable by following two kinds of methods:
● proportional unsteady with the unsteady correlated variables of other controlled variable:
$\mathrm{\Δ}{\mathrm{SP}}_{\mathrm{fcv}}\left(k\right)=\underset{j=1}{\overset{{N}_{r}}{\mathrm{\Σ}}}{\mathrm{\μ}}_{j}\mathrm{\Δ}{R}_{j}(k-{\mathrm{\τ}}_{j})---\left(9\right)$
Wherein: μ
_{j}=CV float related coefficient (j=1 ..., Nr)
Δ SP
_{Fcv}(k)=SP
_{Fcv}(k)-SP
_{Fcv}(k-1) SP
_{Fcv}(k) be the current set-point of CV that floats
if：[R
_{j}(k)>R
_{j.hlm}]or[R
_{j}(k)<R
_{j，llm}]or[R
_{j，llm}＝R
_{j，hlm}]
ΔR
_{j}(k)＝R
_{j}(k)-R
_{j}(k-1)
else?if：[R
_{j，hlm}>R
_{j}(k)>R
_{j，llm}]ΔR
_{j}(k)＝0
R
_{j}(k)=and the relevant variable currency that floats of a j CV, τ
_{j}=retardation time
N
_{r}=CV correlated variables the number of floating
R
_{J, hlm}, R
_{J, llm}Be respectively the bound of the unsteady correlated variables of CV
When needs float, remain unchanged behind the controlled variable set-point tracking measurement value certain hour.
3. the constrained optimization of performance variable: provide the performance variable that those need and may (do not need use as controlled variable control, non-fault, permission performance variable) optimization by " 1. online in real time configuration device ", according to the priority orders that the performance variable of setting is optimized, push performance variable to the optimal value (see figure 1) step by step by " 5. performance variable constrained optimization device " seriatim.Its method one is: change progressively that PID is given to make it move towards to optimize the zone, method two is to keep that PID is given to have certain deviation with measured value, makes PID output move towards to optimize regional.
7. control action is exported " the 3. performance variable output processor " among processing-Fig. 1, has following function:
1. the conversion of model pre-estimating control action output: except that bound and the constraint of speed limit, during practical application, often with the controller of PID controller or two PID series connection as carrying out link, the present invention can be provided with and keep and do not keep two kinds of selections of PID closed-loop control.When not keeping the PID closed loop, the present invention provides and does not keep closed-loop control, allows to adjust pid parameter and impregnable transfer algorithm:
$\mathrm{\ΔMV}\left(k\right)=\frac{1}{{K}_{\mathrm{pid}}}[\mathrm{\ΔU}\left(k\right)+\mathrm{PV}\left(k\right)-\mathrm{PV}(k-1)]---\left(10\right)$
Δ U (k)=model pre-estimating calculates the adjustment amount of gained control action current time
Actual MV (PID the is given) adjustment amount of Δ MV (k)=current time
K
_{Pid}The enlargement factor of=PID controller
PV (k), the controlled variable of PV (k-1)=current and last control PID control constantly
SP(k)＝SP(k-1)+ΔMV(k)
Wherein: SP (k)=current time PID controller set-point (11)
The two-way unperturbed of 2. advanced control and conventional PID control system switches: no matter switch to advanced control from conventional PID, or switch to conventional PID control (containing the automatic switchover of barrier for some reason or other reasons) by advanced person control, all keep the set-point of PID controller constant during switching.
3. output keeps: when the corresponding PID of performance variable (PID is given) institute not as the performance variable of current model pre-estimating control, and need remain on the certain numerical value, be called the output maintenance.
First kind of situation: because of performance variable and correlated variables reaches the upper limit or lower limit can not use as performance variable, but will keep performance variable and crucial directly related with performance variable variable thereof on bound, not continuing transfinites, and prevents the integration saturated phenomenon of PID; Or when performance variable and the crucial variable directly related with performance variable transfinite the adjustment limit value, still performance variable to be remained on the new bound.The present invention provides following algorithm (being limited to example on super):
If:[MV can not be to raising]
if{[OP(k)>(OP
_{hlm}+BL)]and?abs[SP(k-1)-PV(k)]≤2δ}
MV(k)＝PV(k)-δ[OP(k)-(OP
_{hlm}+BL)]； (12)
else?MV(k)＝SP(k-1)-δ；
SP(k)＝MV(k)；
Wherein: SP (k), PV (k) are respectively set-point and the measured values of (as performance variable) PID;
OP (k), OP
_{Hlm}(it is directly related to be called the key operation variable for the output of=PID controller
Variable) and limit value;
BL〉0, δ〉0 be respectively can online adjustment parameter.
Second kind of situation: variable does not transfinite, but is not selected as model pre-estimating control operation variable, and pid control circuit that neither the constrained optimization variable needs to keep PID output (variable valve) constant.The invention provides following two kinds of disposal routes:
● the given tracker measured value of PID controller;
● be changed to " condition " controlled variable;
4. to electing the performance variable of constrained optimization as, press the constrained optimization rule and adjust performance variable.
The controller that constitutes by the present invention is called VSNMPC, and its theory diagram is seen Fig. 1.
The present invention calculates through non-linear time lag, the nonlinear model online in real time is estimated calculating, estimate the online correction of time domain more, SINGLE PREDICTION PREDICTIVE CONTROL, feedback of status, nonlinear weight, the multicycle control algolithm, provide the adjustment amount of each control model pre-estimating control constantly, be converted to the set-point of PID controller commonly used again, form closed-loop control system, detect the running status of controlled process in real time, structure and action command according to current controlled process, the online in real time self-organization is fit to controller architecture and the corresponding controller parameter and the algorithm of present case, form multivariate and coordinate variable structure control system, not only on model but also structurally adapt to the variation of controlled process, the controlled variable real-time constraint is optimized and is floated and coordinate, the performance variable real time coordination is optimized, and makes controlled process all be in the optimization running status in all cases.
Embodiment
Enforcement of the present invention can progressively be carried out in proper order by flow process shown in Figure 2.Wherein:
" 1. online in real time configuration device " contains data and reads in module 1.1, and performance variable availability judge module 1.2 can be surveyed or observable interference availability judge module 1.3, determines current controlled variable module 1.4, controlled variable-performance variable matching module 1.5.
" 2. nonlinear model predictor control counter " contains: state observation calculates and state variable availability judge module 2.1, determines weighting coefficient module 2.2, control law computing module 2.3.
" 3. performance variable output processor " contains: performance variable keeps module 3.1, SP tracking module 3.2, output processing module 3.3.
Read in module 1.1 by data and read the controlled variable that the slip-stick artist sets, performance variable, the variable directly related with performance variable, state variable can be surveyed or the numerical value of observable interference; Read the order that comes into operation of controller that slip-stick artist and operator set.
Do to judge by performance variable availability judge module 1.2:
1. whether the PID controller as performance variable is cut to " CAS " or " long-range given " (reading information by operation interface).
2. performance variable and the variable directly related non-fault (providing) whether by RTDB with performance variable.
3. whether performance variable is allowed to come into operation, and whether performance variable and the variable directly related with performance variable not super bound (slip-stick artist interface).
Whether other PID system that 4. guarantees the normal operation of performance variable drops into closed-loop control (RTDB).
Provide information such as " can freely adjust ", " can only to raise ", " can only to downward modulation ", " unavailable " according to above judgement.
Can survey or observable interference availability judge module 1.3: allowing use and trouble-free actual measurement interference or observation to disturb can be as feed forward variable.
State observation calculates and 2.1 pairs of all state variables of state variable availability judge module are all observed calculating with model.Calculate if actual measurement state variable and observation (can not survey) state variable are carried out model pre-estimating, constitute feedback of status.If when the actual measurement state variable has fault, or the slip-stick artist sets when not allowing to use, and uses the observation computing mode and carries out model pre-estimating.
Determine current controlled variable module 1.4: in all controlled variables that controlling schemes is set, according to the current controlled variable that drops into control of following conditional decision:
1. whether allow to come into operation (reading the order that comes into operation respectively by slip-stick artist interface and operator interface)
2. whether the crucial controlled variable (slip-stick artist's setting) that this controlled variable is relevant comes into operation
3. non-fault, and available performance variable is arranged.
Determine controlled variable weighting coefficient module 2.2: weighting coefficient is the function that controlled variable (when current and future operation variable is constant) is estimated deviation E, need carry out model pre-estimating and calculate.
Controlled variable-performance variable matching module 1.5: according to current permission and may drop into the controlled variable of control, performance variable, can survey or controlled variable-performance variable pairing and priority that observable interference and slip-stick artist set, determine current only controlled variable-performance variable pairing, determine four kinds of application of performance variable: control, constrained optimization, constraint keep and SP follows the tracks of; Determine to push to the controlled variable of optimal value simultaneously.
Control law computing module 2.3:, calculate the optimum control effect by (1)-(8) formula according to controlled variable-performance variable pairing that online configuration is determined.
Performance variable keeps module 3.1: when reaching the constraint limit performance variable can not be used because of the correlated variables of performance variable, the variable directly related with performance variable remained on the constraint limit, prevent the influence that the PID integration is saturated.Or keep PID output constant under certain condition.
SP tracking module 3.2: when the PID controller as performance variable does not switch to " CAS " or " long-range given ", make performance variable output tracking PID given on the spot, switch to this control by PID control to guarantee unperturbed ground.
Performance variable constrained optimization device 5: press performance variable constrained optimization control law and adjust performance variable.
Performance variable output processing module 3.3: to above " control operation variable ", " performance variable maintenance ", " SP tracking " exports to DCS and operator interface after the performance variable under " performance variable optimization " various situations is handled.
Controlled variable constrained optimization device 4: press controlled variable constrained optimization control law and adjust the controlled variable set-point.
For implementing the present invention, also need following support equipment:
10.DCS or other has the system of PID control, realizes the control that this controller provides by this system.
9. real-time data base RTDB: gather the controlled process variable by data-interface by DCS, comprise controlled variable, performance variable, the variable directly related with performance variable, state variable can be surveyed or observable interference and other related data, is the foundation of calculating and controlling.
7. slip-stick artist interface: controller is set and monitored, comprise the used variable controlled variable of this controller, performance variable, the variable directly related with performance variable, state variable can be surveyed or the setting of observable interference, the setting of all variable bounds, speed limit and the order that comes into operation, the setting of model and model parameter and controller parameter, the setting of controlled variable-performance variable pairing and priority, or the like.All setting datas all can leave among the RTDB, can adjust when controller moves.Simultaneously, controller is given RTDB with its running state data, so that the operation of controller is monitored.Relevant data among demonstration of trend display and the record RTDB monitors controller and controlled process ruuning situation.
8. operation interface:, comprise the come into operation switch and the state that comes into operation of each controlled variable, the availability of performance variable and the state that comes into operation, the adjustment of controlled variable set-point, the failure message of controlled variable etc. for the operator provides the interface that this controller is operated.Usually operator interface can be realized at the operating terminal of DCS or other classical control system.
Above support equipment all can utilize existing commercial equipment and software to realize.
Application examples:
Fig. 3 is the general structure that VSNMPC of the present invention uses in petrochemical production device, wherein:
1. the variable structural nonlinear model predictor controller VSNMPC that provides for the present invention
2. be that soft instrument 4. based on nonlinear model is data input/output interface (Data I/O)
5. be that PID controller (in DCS) 6. is controlled production run table
7. be that slip-stick artist interface 8. is operation interface (in the DCS operating side)
9. be that real-time data base RTDB 10. is DCS
The controlled variable of different process units and performance variable are as shown in Table 1.
Table one: VSNMPC is application examples in petrochemical production device