Sei sulla pagina 1di 4

AbstractAdvanced control is the most efficient technology

available to achieve the objective of any industrial process,


namely the reduction of variable costs while maintaining
product quality. Some advanced control methods are: model
based predictive control, internal model control, adaptive
control, fuzzy control, robust control, neural network based
control and optimal control. One advanced control method that
is successfully used in industrial control applications is Model
Based Predictive Control (MBPC). This paper presents a
control application using this advanced control technique for
controlling a binary distillation column.
I. INTRODUCTION
DVANCED control techniques such model predictive
control has proven to be a very successful controller
design strategy, both in theory and practice, because it
provides high performance controllers that can easily be
applied to difficult high-order and multivariable processes,
with constraints, linear or even nonlinear.
Predictive control is a control strategy based on the
explicit use of a process model in order to generate the
predicted values of the output over a future time horizon.
These predicted output values are then used to compute a
sequence of control moves that optimizes the future
behavior of the plant.
Theoretical and practical issues associated with MBPC
technology are summarized in several articles. [2,5,9]
Linear predictive control became popular since 70s
especially through the dynamic matrix control method,
which had been used on large scale in industry due to its
simplicity and its advantages.
The firsts MBPC algorithms, Model Predictive Heuristic
Control [10], Dynamic Matrix Control [3], Extended
Prediction Self-Adaptive Control [4], Generalized Predictive
Control [8] have used only linear process models.
But most of processes are nonlinear, with strong
nonlinearities and a large number of constraints, beginning
with 90s appeared the first concerns regarding nonlinear
predictive control.[1]
II. MODEL BASED PREDICTIVE CONTROL ALGORITHM
The main idea of predictive algorithms is the use of the
process model to predict the behavior of process outputs,
over a specified time horizon with respect to changes in
process inputs. [5] These predictions are used for finding a
control variable values sequence that minimizes an objective
function, without violating some input or output constraints.
Only the first value of the control strategy is
implemented, then the calculations are repeated starting
from the current state, yielding a new control value. The
prediction horizon keeps being shifted forward and for this
reason MBPC is also called receding horizon control (figure
1).


Fig. 1. Receding horizon control:
a) output prediction using the current step, k
b) output prediction for the next step
k the current step, y output, p prediction horizon, m control
variable horizon.

Model Based Predictive Control (MBPC) is a control
algorithm that uses:
- a dynamic model of the process
- a history of past control moves and
- an optimization cost function over the receding
prediction horizon, to compute the optimum control moves.
The optimization cost function is given by:

=
+ A I +

=
+ + +
A
= u
m
l
l k c
c
l
p
l
l k r k l k y
k C
1
,
2
)] 1 ( [
1
2
)] ( ) | ( [
) (
min
(1)

Using an Advanced Control Technique for Controlling a Distillation
Column
Paraschiv N.
*
, Baiesu A.
**
, Stamatescu G.
***

*
Petroleum-Gas University of Ploiesti-Romania IEEE Member,
**
Petroleum-Gas University of
Ploiesti Romania,
***
Politehnica University of Bucharest Romania IEEE Member
A
2009 IEEE International Conference on Control and Automation
Christchurch, New Zealand, December 9-11, 2009
WePT3.5
978-1-4244-4707-7/09/$25.00 2009 IEEE 581



where: y is the output value, r - the setpoint value, k the
current step, p the prediction horizon value, m the
control horizon value, c the control value and
c
l
I is the
control weight.
The control variable variation which makes the output
process value to follow the setpoint trajectory must not be
too aggressive. This disadvantage can be eliminated by
using a control variable weight
c
l
I . As these weight values
are bigger, the control variable variations will be smallest,
and the output will not follow the setpoint trajectory close
enough. Though, finding the optimal value for this weight
will have as result the setpoint trajectory following with low
energy consumption.
The predictive control tuning parameters are:
- the sampling time;
- the control variable horizon;
- the prediction horizon;
- the weight matrices that are used in the optimization
procedure.
III. THE DISTILLATION COLUMN
The distillation column (figure 2) has the L-B, Shinskey
approach [11] controlled structure (the reflux flow is used
for controlling the propylene composition and the bottom
product flow is used for controlling the propane
composition).

Fig. 2. Propylene(C3)/propane(C3) distillation column:
PC pressure controller, FC flow controller, LC level controller, AT
composition transducer, L reflux flow, Li reflux flow setpoint, P -
pressure, Pi pressure setpoint, B bottom product flow, Bi bottom
product flow setpoint, HVR reflux tank level, HVri reflux tank level
setpoint, HB bottom column level, HBi bottom column level setpoint,
F feed flow, xF feed composition, xB bottom composition, xBi -
bottom composition setpoint, xD top composition, xDi top composition
setpoint.

The process has two outputs (the propylene (top) and
propane (bottom) composition) and four main inputs, two
controlled variable (the reflux and bottom product flows)
and two disturbances (the feed flow and feed composition).
(figure 3)

Fig. 3. Distillation process input and output variables.

Using available data of an industrial column the process
was simulated using the HYSYS

simulation environment,
observing that it has a nonlinear behavior, characterized by
different gains and transient times for each operating point
and each process channel.
Using the Markov parameters identification procedure
(figure 4), models of the process were determined for every
process channel. The simulation data are the process
response (top and bottom composition variation) to input
(control variables and disturbances) step changes.

Fig. 4. Markov identification procedure.

The process models are described by second order
transfer functions with dead time (2), having different gains
and time constants for each operating point and process
channel.
1
1
2
2
+ +
=

s T s T
e k
G
s
m
t
(2)
k
m
is the process gain, is the dead time and
2
T

and T
1

are time constants.
IV. DISTILLATION COLUMN MODEL BASED PREDICTIVE
CONTROL
Because the distillation process is a nonlinear one it is
wise to use a model based control structure, which can take
account of the process nonlinearities by changing the
process model according to the operating point. The process
is a nonlinear one, but represented as a reunion of linear
models one for each channel and operating point.
The distillation column is simulated using ASPEN
HYSYS simulation environment, and the control structure is
implemented in MATLAB.
At each simulation initialization step, using a
communication system between HYSYS and MATLAB, the
bottom and top composition, the feed flow and composition
are sent from the process to the control structure, in
MATLAB. Here, using these data the model parameters
(time constants and gain) are determined and loaded in
582



controller. Using these model parameters, the two controlled
variables (reflux flow and bottom product flow) are
computed and sent at each sampling instant to the process, in
HYSYS. (figure 5)


Fig.5. Proposed model based control system structure.

The dynamic system behavior analysis consisted of
modifying the compositions setpoint, the disturbances and
the controllers tuning parameters. (figures 6 to 9)
For top composition MBP controller has the following
default simulation parameters:
- prediction horizon is variable, and it is calculated
using the process model time constant T
1
, from (2), being
T
T
1
4
; [7]
- control variable time horizon 30 sampling times;
- output weight 1 (minimum value: 0, maximum value:
1);
- control variable weight 0.2 (minimum value: 0,
maximum value: 1).

Fig. 6. Top composition trend when the controller setpoint increases
from 0.89 mol. fr. to 0.92 mol. fr..

Fig.7. Top composition trend when the controller setpoint increases from
0.89 mol. fr. to 0.92, control variable time horizon is 25.

Fig.8. Top composition trend when the controller setpoint increases from
0.89 mol. fr. to 0.92, control variable weight is 0.7.

Fig.9. Top composition trend when the feed flow increases from 241.5
kmol/h to 246.5 kmol/h.
583



V. CONCLUSION
In this paper was presented a solution for controlling a
distillation column from a catalytic cracking unit, using a
model based predictive control technique.
The MBPC tuning parameters are: the sampling time, the
prediction horizon, control variable horizon, output and
control variable weight.
As can be seen from the presented trends, the behavior of
the process with the control system was studied for different
values of the tuning parameters, observing that a decreasing
of the control variable horizon or a decreasing of the control
variable weight, from the default values, can lead to an
increasing of the transient time.
Also we can observe that the process output value reaches
the setpoint value with the best dynamic performances for
the default values of the tuning parameters.
The control system has a robust behavior when a
disturbance appears in the process.
REFERENCES
[1] Allgwer F., Badgwell T. A., Qin J. S., Rawlings J. B., Wright S. J.,
Nonlinear predictive control and moving horizon estimation An
introductory overview. Advances in Control, Highlights of ECC99,
1999, pp. 391449.
[2] Camacho E.F., Bordons C., Model Predictive Control. Springer
Verlag, 1999.
[3] Cutler C. R., Ramaker B. L., Dynamic Matrix Control a computer
control algoritm. AIChE National Mtg., Houston, Texas, 1980.
[4] De Keyser R. M. C., Van Cauwenberghe, Extended prediction self-
adaptive control. IFAC Symp. on Identification and System Parameter
Estimation, York, 1985, pp.1255-1260.
[5] De Keyser R. M. C., Van de Velde, Dumortier F.,A comparative study
of self-adaptive longrange predictive control methods. Automatica,
24, 1988, pp. 149-163.
[6] De Keyser R. M. C., A gentle introduction to Model Based Predictive
Control. The Conference on Control Engineering and Signal
Processing, Piura, 1998.
[7] Morari M., Lee J., Garcia C., Model predictive control, 2002.
[8] Peterka V., Predictor-based self-tuning control. Automatica, 20, 1984,
pp. 39-50.
[9] Rawlings J. B., Tutorial overview of model predictive control. IEEE
Contr. Syst. Magazine, 20(3), 2000, pp. 852.
[10] Richalet J. A., Rault A., Testud J. L., Papon J., Model Predictive
Heuristic Control: application to an industrial process. Automatica,
14, 1978, pp. 413-428.
[11] Shinskey F.G., Distillation Control, McGraw-Hill, 1984.

584

Potrebbero piacerti anche