You are on page 1of 42

APACT’04, Bath

April 26-28, 2004

Model Predictive Control:
A Success Story Continues
Frank Allgöwer *
Institute for Systems Theory in Engineering
University of Stuttgart

Some Facts about Model Predictive Control

• Widely accepted in process industries as the advanced control
methodology
• Tremendous impact on academic process control community, e.g.:
1. 919 papers published in 2002 in area of MPC (Inspec)
2. IFAC Adchem 2003, 30% of submitted control papers are in
the MPC area
3. (at least) four special journal issues being prepared at present
• Commercially very successful

Non-exhaustive MPC Vendor List






















ABB
ACT
Adaptics
Adaptive Resources
Adersa Home Page
Aspen Technology
Aurel Systems Inc.
Batch CAD
Bonner and Moore
Brainwave
C.F. Picou and Associates
Chemstations
Comdale Technologies
Control Arts Inc.
Control Consulting Inc.
Control Dynamics Homepage
Controlsoft Incorporated
Cybosoft
DOT Products
Trieber Controls
Yokogawa APC
US Process Control L.L.C.
Eldridge Engineering Inc.























Elsag Bailey
Envision Systems Inc.
Gensym
Enterprise Control Technologies
Fantoft Process Group
MATHWORKS
Honeywell
Hyprotech
Inferential Control Company
IntellOpt
Knowledgescape
MDC Technology
Neuralware
Nexus Engineering
Objectspace
Optimal Control Research
Pavilion Technologies
Predictive Control Ltd.
Process System Consultants
RSI
Simulation and Advanced Controls Inc.
Simtech
Texas Controls Inc.

Structure of Presentation

• A short introduction to theory and practice of MPC
• Linear vs. nonlinear MPC
• Efficient nonlinear MPC formulation

Conclusion of presentation:
Nonlinear model predictive control of realistically sized chemical
processes possible, if
(i) state of the art NMPC formulations,
(ii) specialized NMPC optimizers
are used.

u(τ ))dτ t • satisfies constraints u(τ ) ∈ U x(τ ) ∈ X .Typical Control Problem x˙ = f (x. x(0) = x0 Find stabilizing control strategy that • minimizes objective functional Z ∞ J = F (x(τ ). u).

Different Solution Strategies Open-loop vs. unstable systems. Closed-Loop Optimal Control Closed-loop optimal control: Open-loop optimal control: Feedback u=k(x) s. . closed-loop trajectories satisfy optimality conditions • Feedback present! • suited for uncertainty. • Finding closed solution hardly possible Input trajectory u = u(t. . .t. x0) solving optimization problem • Computation often feasible • • • • No feedback! Unstable systems? Uncertain systems? No reaction to disturbances . disturbances.

u(τ ))dτ t • Apply optimal open-loop input for τ ∈ [t. t + δ] • Usually use finite Prediction horizon Tp . x(t)) = u(·) Z TP F (x(τ ).Model Predictive Control MPC= repeated open-loop optimal control • Solve open-loop optimization problem all δ sampling instances min J(u(·).A Possible Solution .

MPC past future prediction horizon set-point predicted state x(t) state x(·) optimal input u at time t input u(·) t t + TP .

MPC past future prediction horizon set-point predicted state x(t + δ) optimal input u at time t + δ state x(·) input u(·) t t+δ t + TP t + δ + TP .

Characteristics of MPC • Moving horizon implementation • Performance oriented time domain formulation • Incorporation of constraints • Explicit system model used to predict future plant dynamics .

Linear MPC . u) • Nonlinear constraints h(x.Nonlinear MPC Linear MPC • Uses linear model: x˙ = Ax + Bu • Quadratic cost function F = xT Rx + uT Ru • Linear constraints Hx + Gu < 0 • ⇒ Quadratic program Nonlinear MPC (NMPC) • Nonlinear model: x˙ = f (x. u) < 0 • Nonlinear program . u) • Cost function can be nonquadratic F (x.

1967) .” (Lee & Marcus. The first portion of this function is then used during a short time interval. The process is then repeated.A Brief History of MPC Idea is rather old: “One technique for obtaining a feedback controller synthesis from knowledge of open-loop controllers is to measure the current control process state and then compute vary rapidly for this the open-loop control function. after which a new value of the function is computed for this measurement.

. . . Mayne & Michalska 1990. • Predictive control theory: Keerthi & Gilbert 1988. . Early industrial MPC applications: • Model Predictive Heuristic Control (IDCOM) Richalet et al. . . . Adersa • Dynamic Matrix Control (DMC) Cutler & Ramaker 1979 . Thomas 1975. .A Brief History of MPC cont. . . . 1976 . Shell Oil Academic research: • Few early theoretical investigations: Kleinmann 1970. Chen & Shaw 1982.

Linear MPC: Commercial Products By now linear MPC is a fully established advanced control method: • Already in 1995 over 2200 applications reported • First. mainly applications in refining/petrochemical sector By now much broader scope • Large applications (603 x 283. DMC Corp.) • Significant benefits reported Virtually all MPC applications up to mid 90s use linear step or impulse response models • Models found by suitable identification • MPC Software often includes identification modules • Online computations require solution of QP .

Joe Qin and T. A. Badgwell .Linear MPC Applications Area Refining Petrochemicals Chemicals Pulp and Paper Air and Gas Utility Mining/Metallurgy Food Processing Polymer Furnaces Aerospace/Defense Automotive Unclassified Total Aspen Honeywell Adersa PCL SGS Total Tech Hi-Spec 1200 480 280 25 1985 450 80 20 550 100 20 3 21 144 18 8 50 10 10 6 7 41 4 16 10 17 42 13 40 1833 40 696 UT/ TWMCC/AspenTech 1 1045 1438 3 7 26 125 450 450 68 10 14 37 51 17 45 13 7 1601 4542 © S.

.Possible Benefits of NMPC • Improved controller performance due to the ability to use better (=nonlinear) models • Enlargement of application field – regulation problems with ∗ highly nonlinear processes ∗ large disturbances – servo control problems with frequently changing operation points – control of batch processes – ...

Application of NMPC to a Chemical Reactor Cyclopentenol Production V. c˙ B = −V˙ cB + k1(ϑ) cA − k2(ϑ)cB . 3 ϑ . 1 ϑ˙ = V˙ (ϑ0 − ϑ) − k1(ϑ)cA∆HRAB + k2(ϑ)cB ∆HRBC ρCp  k w AR 2 +k3(ϑ)cA ∆HRAD + (ϑK − ϑ) ρCpVR   1 ˙ ˙ ϑK = QK + kw AR (ϑ − ϑK ) mK CP K   Ei ki(ϑ) = ki0·exp . A. i = 1. B. A. Tf Klatt/Engell/Kremling/Allg¨ower 93 Van der Vusse reaction scheme: A k k 2 1 −→ B −→ C k VR V. cA0 2 c˙ A = V˙ (cA0 − cA) − k1(ϑ)cA − k3(ϑ)cA . D m QK K 3 2A −→ D controlled variables: manipulated variables: disturbances: concentration cB flow rate V˙ feed temperature Tf . C. 2.

Operation at Point of Optimal Yield 1.1 2 Yield: .

cBs .

.

Φs = cA .

0 at optimum: .

∂Φ .

= ∂u opt → Kopt Φ s 0→ =0 .

∂y .

= ∂u opt 1 1.9 5 10 15 us • gain changes sign at optimal operating point ⇒ strong nonlinearity ⇒ not integral controllable 20 25 30 35 .0 Ks 0 0 -1 0.

D m K • • • • rejection of disturbances in Tf . cA0 setpoint changes in cBs input constraints in V˙ control tolerance 0. A.NMPC of the Chemical Reactor V. C.02 mol l QK NMPC setup: • performance index: J = R t+Tp t • output feedback case (EKF) (cB − cBs)2dt . A. Tf Control task: VR V. B.

Performance of NMPC-controller 1.1 setpoint 1 0.9 30 Input  −1  ˙ V h 20 10 0 0 x scenario 500 1000 1500 ↑ Tf = 100o C 2000 Time[s] 2500 3000 3500 4000 ↑ Tf = 115o C | {z }| {z } model/plant mismatch model/plant mismatch case 1 case 2 .2 Output  mol  cB l 1.

5 ↑ 5.5 = 104.1 l 1 0.5 ↑ 4.2 1.1 ↑ 4.3 Output  mol  cB R t+Tp t c2B (τ )dτ 1.7 ↑ 4.Control for Optimal Yield task: on-line optimization of yield J = 1.9 40 30 Input  −1  ˙ V h 20 10 0 0 x cA0 h mol i l Tf [o C] 500 1000 1500 2000 2500 Time[s] 3000 3500 4000 4500 = ↑ 5.9 1 100 1 115 5000 .

2 0. h i cA0 mol l 400 300 200 0 ↑ 5.22 yield Φ 0.1 500 1000 1500 2000 Time[s] 2500 ↑ 5.Comparison to Set-point Control setpoint control yield optimization 0.21 0.7 production rate doubled 3000 3500 4000 .19 production rate  mol  ˙ V ·cB h ∼ NT $ x.

Linear vs. Nonlinear MPC Reasons for present dominance of LMPC: • Linear models can be straightforwardly identified from process test data • Efficient implementation of LMPC possible LMPC → QP has to be solved NMPC → NLP has to be solved • Tuning for stability easy for LMPC .

... robustness.. .. . modeling. • Implementation issues: efficient and reliable real-time optimization. performance.Issues in Nonlinear MPC • System theoretic formulation and investigation: stability.

Efficient Numerical Solution of the Open-Loop Optimal Control Problem • Typically solved by finite dimensional input parameterization • Can.] ) . for example be solved by: direct multiple shooting method u 6 6 u xu *i u xi u u u u u u u u u sN sN−1 u u x0 uu u u t t1 t2 u u u- tN−1 tN x0 uu us1 u t t1 t 2 u u u - tN−1 tN • Optimization problem shows special sparse structure • Consecutive optimization problems are similar ⇒ use efficient hot starting/embedding strategy State of the art solution strategy. specially tailored to NMPC ⇒ allows real-time solution for realistically sized problems (based on MUSCOD II. [Diehl et al.

Approximate Solution: Real-time Iteration Scheme Question: Is it really necessary to solve optimal control problem “exactly”? • Often sufficient to perform only one SQP subiteration per sampling time if: – special update is used – close to optimal solution • Can be performed fast • Rigorous stability proof available [Diehl/Findeisen/Allg¨ ower/Bock/Schl¨ oder ’03] Real-time iteration scheme • Expands range of applicability (faster systems. ..) • Utilizes inherent robustness of NMPC ..

. • Implementation issues: efficient and reliable real-time optimization. modeling.Issues in Nonlinear MPC • System theoretic formulation and investigation: stability. .. performance. robustness. ....

An Important Issue in (N)MPC Theory Even in nominal case: • no model plant mismatch • no disturbances Predicted open-loop trajectories 6= Closed-loop trajectories x(t) x(t + δ) finite horizon t + TP t + δ + TP .

What is achieved R t+Tp by repeatedly minimizing t F (x(τ ). u(τ ))dτ ? • Stability? Why should the closed-loop be stable? Schemes that guarantee stability? . u(τ ))dτ .An Important Issue in (N)MPC Theory x(t) Even in nominal case: • no model plant mismatch • no disturbances x(t + 2δ) x(t + 3δ) Predicted open-loop trajectories 6= Closed-loop trajectories R∞ • Performance? Goal: min 0 F (x(τ ).

s. .t. . F .Dilemma of NMPC Guaranteed Stability versus Performance good performance ⇒ Large horizon ⇒ closed loop stability computationally not feasible Need for short horizon formulations that achieves good performance and stability 2 possible approaches: • choose Tp. closed loop is stable • design NMPC scheme. . closed loop is stable independent of choice of Tp. s. ⇒ “guaranteed stability” NMPC scheme . . .t. .

u(τ ))dτ x(t) t subject to: x˙ = f (x. Gilbert ’88.computationally not feasible x(t + TP ) = 0 . Mayne.Zero Terminal Constraint NMPC [Keerthi.feasibility. u) u Z t+Tp J(·)= F (x(τ ). system dynamics x(t) given “state feedback” u(τ ) ∈ U input constraints x(τ ) ∈ X state constraints x(t + Tp) = 0 zero terminal const. Michalska ‘90] Simplest NMPC scheme with guaranteed stability min J(x(t). u). + stability . performance for short horizons? .

NMPC Schemes with guaranteed stability Avoid ZTC. u(τ ))dτ+E(x(t + Tp)) t x(t) subject to: x˙ = f (x. u) u Z J(·)= t+Tp F (x(τ ). system dynamics x(t) given “state feedback” u(τ ) ∈ U input constraints x(τ ) ∈ X state constraints x(t + Tp)∈ Ω terminal constraint x(t + TP ) E(t + TP ) Ω Terminal penalty term E(x(t + Tp)) approximates infinite horizon cost in terminal region i. u). instead use • terminal region constraint • virtual local control law inside of terminal region that allows “infinite” prediction approximation via E(x(t + Tp)) min J(x(t). Quasi-infinite horizon NMPC [Chen. Allg¨ ower 1997] .e.

t.al. even for small TP Advantages: • guaranteed stability .Guaranteed Stability [Chen&Allg¨ ower ’98]. [Fontes ’01] min J(x(t). [Mayne et. u) u Z with: J(·)= t+Tp F (x(τ ). u) < 0 ∂x b) optimization feasible for t = 0 ⇓ R x(t) E(x(t + TP )) Ω Asymptotic Stability Guaranteed Region of Attraction: Set R of states satisfying b) • ∞ horizon performance approx.: ∀x ∈ Ω ∃u ∈ U with ∂E f (x. u(τ ))dτ +E(x(t + Tp)) t and: x(t + Tp)∈ Ω Nominal Stability: If a) E(·) and Ω s. ’00]. u) + F (x.

al.’97] ’99. u) = xT(τ )Qx(τ ) + uT(τ )Ru(τ ) ⇒ E(x) = xT P x ⇒ Ω := {x ∈ Rn|xT P x ≤ α} P is solution of Lyapunov equation α is solution semi-infinite optimization . – . ’00] Exemplatory quasi-infinite horizon NMPC: – Jacobian linearization stabilizable – quadratic cost function F (x. How to determine E.. al.al.. Primbs et. Ω? [Chen&Allg¨ ower ’97] [De Nicolao et.Comments • Many schemes fit into this setup: – Quasi-infinite horizon NMPC – Simulation-approximated infinite horizon NMPC – CLF approaches [Jadbabaie et.

δ → 0 (instantaneous implementation) predicted state x(t + δ) input u x(t) state x(·) input u(·) t t+δ Unrealistic assumptions. however (i) allows to develop system theoretic understanding (ii) lays basis for development of methodological tools key focus on stability issue t + TP t + TP + δ . i.e.NMPC Theory in the 90s: Instantaneous State Feedback NMPC Assumptions: • On-line computation of optimal input trajectory u? requires no computation time (instantaneous solution) • Continuous application of optimal input u = u?(x).

Sampled-Data Output Feedback NMPC: Wrap Up Under certain conditions it is possible to extend the state feedback stability results to the sampled-data output-feedback case Problems: • When are the assumptions satisfied? – Value function uniformly continuous in R? • Separation requires fast enough error convergence of observer −I measurement noise? Present results must be seen as a first step towards a practical output feedback NMPC scheme .

Application of NMPC to a Distillation Column Separation of liquid mixtures into its components L xD 40 D 28 T28 F. XF 21 14 V 1 Q xB B T14 .

164th. V F. XF 21 14 • Sampling time δ = 30s T14 V 1 Q • Models of different sizes available: 5th. xB at setpoints inferential control: T14. T28 • Quadratic weights on deviation from set-points xB B • Methanol/n-Propanol • High purity separation • Performance? • Computational feasibility? . 204th order • States estimated from T14. 42nd. xF • Manipulated variables: L. T28 controlled variables • Disturbances: F .Application of NMPC to a Distillation Column L xD 40 D 28 T28 • Objective: keep xD .

−15. +5 %) • Quasi-infinite horizon NMPC • Efficient dynamic optimization approach (multiple shooting. MUSCOD-II) Only Simulations. How about “real” application? .Simulations: Performance + Computation −3 x 10 x28−x28s 1 0 −1 −6 V−Vs [kmol/s] x 10 2 1 0 CPU time [s] −1 164th order model 42nd order model 5th order model 5 0 0 500 1000 1500 2000 time [s] 2500 3000 3500 • Feed concentration disturbances (+10.

Pilot Plant for Experiments Located at Institute for System Dynamics and Control. University of Stuttgart .

Experimental Results NMPC PI Controller T28 [oC] 71 • 204th order model 70.5 • 2 PI control loops: T14 → L. T28 → Q 70 Q [kW] 3 • feed flow increase by 20% at t = 1000s 2.5 2 0 1000 2000 3000 time [s] 0 1000 2000 3000 time [s] NMPC is computationally feasible and delivers good performance .

Conclusions • • • • Gave a short introduction to NMPC Linear MPC is a well established advanced control technology Nonlinear MPC is generally seen to hold much promise In the past decade a NMPC control theory was developed based on instantaneous considerations • Current research focuses on extensions to realistic scenarios – output feedback – sampled-data implementation – robustness – computational issues • Showed experimental results that prove feasibility of methodology for industrially sized problems NMPC will have a significant impact in nonlinear control applications over the next five years .