A Two-Level Strategy of Integrated Dynamic
Optimization and Control of Industrial Processes A Case Study
Lehrstuhl für Prozesstechnik, RWTH Aachen, Germany
J. V. Kadam, M. Schlegel, W. Marquardt*
Systems and Control Group, TU Delft, The Netherlands
R. L. Tousain, D. van Hessem, J. van den Berg, O. H.Bosgra
Two-level strategy
Vertical decomposition
Optimal process operation
Implications
•
Objectives:
Constraints:
•
Maximize profit
•
Changing market conditions
•
On-spec production
•
Process disturbances
•
Feasible operation profiles •
•
Decomposed optimization-control problem
Complex dynamic optimization
and control problem
D-RTO
Involves repetitive decision
making
MPC
ref
min

(
x
,
u
, t0 , t fi )
ref
Operational & safety constraints
u
min
u
,t fi
dE
t
Vertical decomposition approach
d , x, y , u
d,x
y, u
Estimation
D-RTO
trigger
s.t. 0  f ( x , x , u , d , t ), x (t0i )  x0i
y ref  g ( x , u ref , d , t )
yref , uref
~
d , x, y, u t
0  h ( x , u ref , d )
t  [t0i , t fi ]
MPC
u
Plant (model)
(incl. base control)
d (t )
Interplay between D-RTO and MPC
•
•
•
Soft constraints can be moved from MPC to
D-RTO
D-RTO
Longer time horizon for D-RTO to ensure
feasibility
-
~
t0i 1  ~
t0i   ~
t, ~
t fi 1  ~
t fi   ~
t
Decomposition based on objectives  economic optimization (D-RTO) & tracking (MPC) subproblems
•
Different models, derived from a first principle model, at each level
•
Different set of constraints at each level
Implementation
An MPC using linear
time variant model
updated
updated
yref , uref
yref , uref
An MPC using
sequential approach
dynamic optimization
D-RTO trigger
A Matlab implementation of
an EKF for constrained
state estimation
LTVMPC
EKF
(Estimator)
ADOPTmpc
(MPC)
based on disturbance sensitivity analysis of
optimal solution
a re-optimization is triggered only if the
detected persistent disturbances have high
sensitivities
fi
•
t
D-RTO trigger for a possible re-optimization
0i
t0i 1  t0i  t , t fi 1  t fi  t
ŷ
~
t
~ ~ ~ ~ ~ ~
s.t. 0  f ( x , x , u , d , t ), x (t0i )  ~
x0i
~ ~
~
~
y  g ( x , u, d , t )
~ ~ ~
0  h ( x , u, d )
~
~
~
t [t , t ]
ref
D-RTO
ref T
ref
(
y

y
)
Q
(
y

y
)
 i
i
i
ref T
ref
 (ui  u ) R(ui  u )
DYNOPC
.......
~
t
Scheduler
ADOPTrho
(D-RTO)
INCA-OPC server
A simultaneous
approach based
dynamic optimizer
(in collaboration with CMU,
Pittsburgh, USA)
MPC
Connection to DCS
(process plant)
possible
 a strict operation envelope is computed which is used on the MPC level
An extension of a sequential
approach based dynamic
optimizer (ADOPT) for real-time
applications
gPROMS
(Process model)
•
Delta-mode MPC computes updates to the control profiles for tracking the process in the strict
operation envelope: rejects fast frequency process disturbances
•
D-RTO optimization problem is initialized with the solution on previous time horizon
•
A flexible software architecture for implementation of the two-level strategy
•
MPC optimization problem is initialized with the D-RTO solution
•
Is being applied to large-scale industrial processes
Case study: Semi-batch reactive distillation column
Problem description
R
xD
D
V
Discussion
•
•
Methyle acetate (MA) semi-batch reactive distillation column:
gPROMS model with 817 DAEs
Control variables: reflux ratio R, vapor stream V
•
Disturbance scenario: 50% drop in side stream feed rate and
other nominal process disturbances
•
•
Open-loop operation: the desired product quality (xD) is not met
•
A different strategy than the traditional MPC approach
•
Delta mode MPC and NMPC only:
•
Flexible plant operation in changing market and operating
conditions can be achieved by two-level strategy
•
Guaranteed overall (economical and operational) feasibility
that might not be achievable by an MPC only
Objective: Maximize production of MA for a fixed batch time of
4 hours (optimum by an off-line optimization)
•
Application of two-level strategy, nonlinear MPC (NMPC),
delta-mode MPC and open-loop operation
Conclusion
-
rigorous nonlinear model (the best option that can be considered) is used
produce off-spec product (economically infeasible)
 not economically viable
•
Two-level strategy: Real-time dynamic optimization and delta mode MPC
-
-
re-optimization triggered by the sensitivity-based approach
 can handle large-scale industrial problems
Future research work:
 new reference trajectories are determined
•
Rigorous strategy for D-RTO trigger, disturbance forecasting
desired product quality is met in the closed loop operation
•
Relation of process models on different levels
•
Fast numerical algorithms on different levels, etc…
 economically feasible operation
Funded by the European Commission under grant G1RD-CT-1999-00146
in the “INCOOP” project
http://www.lfpt.RWTH-Aachen.de/INCOOP * [email protected]