Process Optimisation using Predictive Control (*)
Dr. V.M. Becerra
City University, Control Engineering Research Centre
Northampton Square, London EC1V 0HB
Tel: 0171 - 477 8132, Fax: 0171 - 477 8568
e-mail: v.m.becerra@city.ac.uk
ABSTRACT
Traditional approaches to steady state optimisation using linear
model-based
predictive controllers use a separation between setpoint regulation and
setpoint economic optimisation. The objective function of traditional
predictive controllers depends mainly on the error between the predicted
variables and the corresponding setpoints, while the setpoints may be
computed by a separate steady state optimisation algorithm.
The predictive control approach presented here is innovative because the
objective function of the predictive controller contains the steady
state
economic objective, such that there is no need to use a separate steady
state optimiser and the setpoints are calculated by the predictive
controller
itself. The approach uses an adaptive linear state space model of the
process
to perform predictions. The predictions are initialized from state
values
measured from a rigorous model operating in parallel with the plant
which
is tuned with reconciled process data. The fact that the linear model
is
adaptive enables the optimiser to drive the process to its steady state
optimum.
Several types of constraints may be handled, including magnitude and
rate
constraints on the manipulated variables and output magnitude
constraints.
The approach has advantages over traditional optimisation techniques
based on
steady state models of the process which use on-line steady state
information.
Firstly, the optimisation is based on dynamic information and, hence,
there is
no need to wait for the system to settle down to collect measurements,
perform
computations and update the setpoints. Secondly, the predictive nature
of the
scheme gives it the ability to anticipate the activation of constraints
into the future, and to take appropriate action in advance. Thirdly, the
periodic
operation of the algorithm enables it to dynamically track changes in
the
optimal operating conditions of the process due to disturbances, price
or
constraint changes. Finally, state constraints are dynamically enforced
while driving the process towards the optimum, and state constraint
infeasibilities are
handled using a penalty approach.
The technique has been tested with realistic simulations of an
industrial
multicomponent distillation column, using the rigorous process simulator
OTISS.
(*) This work is supported by the Engineering and Physical Sciences
Research Council, Grant No. GR/J7345