Professional Documents
Culture Documents
Model Predictive Control of Distillation Column Based Recursive Parameter Estimation Method Using HYSYS Simulation
Model Predictive Control of Distillation Column Based Recursive Parameter Estimation Method Using HYSYS Simulation
Abstract Some chemical plants such as distillation column model of system available online while the system in
have highly nonlinear behavior. these processes demand a operation. The main goal of this paper consist on
powerful identification method such as linear and obtaining a mathematical model that adequately
nonlinear models. In this paper, a distillation column is describes the stream composition variation process
simulated in a rather realistic environment by HYSYS
inside a distillation column by using system
and the obtained data is in connection with MATLAB for
identification and control purpose. In this case, the identification techniques. The distillate and bottoms
identified model is characterized by two structures, Linear streams compositions can be inferred by measuring the
model structure based on ARX ( Autoregressive with temperature of trays near each of the column. The tray
external input) and nonlinear model structure based on temperature measurement dynamics are typically more
neural network. For control goals, two linear and rapid than the product stream composition
nonlinear model predictive controllers are applied. measurements, so tighter disturbance rejection may be
General predict control (GPC) and nonlinear predict possible with this strategy. The paper is organized as
control (NPC) are compared based adaptive identification follows. The system identification procedure is
of model. Since, practical systems change with time and
described in Section 2. Section 3 discusses the Adaptive
the parameters of system are time varying, using real-time
identification based recursive parameter estimation is identification methodology and recursive estimation
necessary, although, desired control strategy is reached method. predictive control based real-time
with a good parameter estimation. The algorithm has been identification is introduced in Section 4. System
tested on an distillation column. The resulting identification and control of a distillation column is
performances show the successful and promising developed in Section 5 and the conclusion is given in
capabilities of the proposed algorithm . the last section.
Keywords Adaptive identification, Kalman filter, Distillation
column, Neural network, predictive control, Parameter
estimation, HYSYS.
2 System Identification Procedure
309
) In this method, the search direction is introduced:
value q (k ) and covariance matrix as follow (Ljung,
1999):
f ( ) = - B ( ) G (U ( ) ( t ))
i i i
t -1 -1
(15)
P (t ) = [ j ( k )j T (k )] = p -1 (k - 1) + j (k )j T (k ) (10)
k =1
-1 Where, G is gradient vector and B specifies the
= p (k - 1) - p ( k - 1)j (k ) I + j T ( k ) p ( k - 1)j (k ) j T ( k ) p ( k - 1)
approximation of the inverse Hessian [1].
So the kalman filter gives the conditional expectation of 6 System identification and control of a distillation
q (k ) as : column
) ) )
q ( k ) = q ( k - 1) + K ( k ) y ( k ) - j T ( k )q ( k - 1) (11) In this Section, two types of predictive control are
investigated and their corresponding results are
Where the K is introduced kalman gain as follow: compared. To link between MATLAB and HYSYS, a
MATLAB file from O.T.Berglihn with some
p(k - 1)j (k )
K (k ) = (12) modifications is used [9]. In this paper, an online
T
1 + j (k ) p(k - 1)j (k ) connection between HYSYS and MATLAB is
implemented to control of Distillate product
The problem of time varying parameters of system is composition. In industrial application, control of
tackled by assigning less weight to older measurements product composition was accomplished by top tray
that are no longer representative for the system [4]. temperature control . whereas, accessing to product
composition data needs an expensive sensors and
4 model predictive control based real time implement large lag to the plant , using top tray
identification temperature is more common to use .Hence top tray
temperature will be used to control of product
4.1 Generalized Predictive Control (GPC) composition . In distillation column, reflux mass flow is
selected as a manipulated variable. Reflex mass flow,
GPC is predictive control approach with linear models has direct effect on the top tray temperature and
such as ARX, ARMAX and ARIX. In GPC consequently on the product composition. The steady
implementation the first step is system identification. state value of reflux mass flow is 100 kg/h and the top
ARIX model has been chosen for this algorithm. The tray temperature is 58.93C. Earlier, discussed that using
structure of ARIX models has been demonstrated model for GPC algorithm is known as Auto Regressive
below: and Moving Average (ARIX) model. Thus, the process
model adopted by GPC is:
y ( k ) = G ( z -1 )u ( k ) + H ( z -1 )e ( k )
A( z -1 ) 1 z - k B( z -1 ) C ( z -1 )
G ( z -1 ) = , H ( z -1 ) = y (t ) = -1
u (t ) + e(t ) (16)
B ( z -1 ) 1 - z -1 (13) A( z ) A( z -1 ) D
310
track the reference when the exerted changes are small. 2
Nonlinear
Control
Predictive
2
Fig .4. NPC reference trajectory and manipulated variable
0
-
2
-
7 Conclusion:
40 5 10 15 20 25 30 35 40 45 50
0 0 0 0 0 0 0 0 0 0
inpu
60
0
t
In this paper, the task of applying model predictive
40
0 control based a complete system identification
20
0
procedure for a distillation column process is
0
0 5 10 15 20 25 30 35 40 45 50
considered. Whereas, distillation column has strong
0 0 0 0 0 0 0 0 0 0
nonlinearity and wide range of operating points, Real-
time identification based recursive parameter estimation
Fig .2. Ordinary GPC reference trajectory and manipulated variable
is used. The results show that adaptive GPC based
output recursive parameter estimation has successful and
promising capabilities in real-time identification and
15
10
5
control.
0
Reference:
-5
input
600 and implementation using MATLAB. 2009 Springer-
Verlag London Limited.
400
[2] J.R Rodrigues Vasquez, R.Rivas Perez, J.Sotomayor
200 Moriano, J.R.Peran Gonzalez. System identification of
steam pressure in a fire-tube boiler. 2008 elsevier. Journal
0
0 50 100 150 200 250 300 350 400 450 500 of computers and chemical engineering
[3] M.Norgaard, O.Ravn, N.K.Poulsen and L.K.Hansen.
neural network for modeling and control of dynamic
Fig .3. Adaptive GPC reference trajectory and manipulated variable systems. Springer, verlag London limited 2000.
[4] Lennart Ljung. System identification, theory for the user.
NPC, which is abbreviated of Nonlinear Predictive ISBN 0-13-881640-9
Control, is appropriate algorithm for highly nonlinear [5] William.Luyben. distillation design and control using
process. In order to remove the superfluous weights aspen simulation. Copyright # 2006 by John Wiley &
Sons, Inc. 2005023397
from the network, pruning determines the optimal [6] William L.Luyben. process modeling, simulation, and
network architecture according to the Optimal Brain control for chemical engineers. McGraw-Hill
Surgeon strategy [3]. Levenberg Marquardt algorithm is [7] Totok R. Biyanto, Totok Suhartanto and Bambang L.
used for network training. Set point tracking behavior Widjiantoro. Predicting Liquid-Vapor (LV) composition
of the regulator system with NPC, corresponding to top at distillation column.
[8] Oliver Nelles. Nonlinear system identification. ISBn 3-
tray temperature is shown in Fig .4. It can be seen that 540-67369-5 Springer-Verlag Berlin Heidelberg New
NPC couldnt track reference under large disturbances York.
exertion. So, adaptive GPC based recursive parameters [9] O.T.Berglihn. a toolbox for using MATLAB as an
estimation has powerful ability to control of time- active/COM controller for HYSYS, MATLAB centeral,
http://www.pvv.org/olaf/hysyslib/,1999.
varying systems.
311