Professional Documents
Culture Documents
1)
Minh et al. / Model Predictive Control of ... / pp. 4-12
Submitted: 10/12/2009
Accepted: 12/01/2010
Appeared:10/02/2010
HyperSciences.Publisher
Abstract— This paper develops a mathematical model and simulation for a condensate distillation
column based on the dynamic continuity and nonlinearity of the mass and the energy. The linear-reduced
order model is used as a regulator of model predictive control (MPC) to verify the ability of a
conventional MPC controller that satisfies the output specifications subject to the input and output
constraints. A modified MPC controller with zone regions instead of setpoints as a new control objective
function for the controlled outputs has been investigated for improving the control performance of this
ill-conditioned process.
Keywords: Distillate purity/impurity; Composition control; Model predictive control; Ill-conditioned
process; Zone regions; Robust control.
♠
1. INTRODUCTION The goals of this paper are twofold: firstly, to present a
theoretical calculation procedure of a condensate column for
Distillation is the most popular and important separation simulation and analysis as an initial step for a project
method in the petroleum industries for purification of final feasibility study, and secondly, for the controller design: a
products. Distillation columns are made up of several linear reduced order such that it best reflects the dynamics of
components, each of which is used either to transfer heat the distillation process is derived and used as a regulator for a
energy or enhance mass transfer. A typical distillation MPC controller. A modified MPC controller with zone
column contains a vertical column where trays or plates are regions is studied to increase the ability of the controller to
used to enhance the component separations, a reboiler to deal with the model uncertainty for an ill conditioned
provide heat for the necessary vaporization from the bottom process.
of the column and a condenser to cool and condense the
vapor from the top of the column, a reflux drum to hold the An L-V configuration or the energy balance method is
condensed vapor so that liquid reflux can be recycled back selected as the control structure for the distillation column. In
from the top of the column. this control structure the liquid flow rate L and the vapor flow
rate V are the control inputs. The objective is to maintain the
Most of distillation control systems, either conventional or specification of the product concentration outputs xB and xD
advanced, assume that the column operates at a constant despite disturbance in the feed flow F and the feed
pressure. Pressure fluctuations make the control more concentration xF as shown in Fig. 1.
difficult and reduce the performance. The L-V structure,
which is called the energy balance structure, can be The MPC has been selected for controlling the distillation
considered as the standard control structure for a dual column since MPC has been originally developed to meet the
composition control distillation. specialized control needs of petroleum refineries (Qin, J. and
Badgwell, T. (1997)). MPC combines the best of both open-
loop and closed loop control methods. It generates an online
♠ feedback control by using the open-loop optimization. The
The authors would like to acknowledge Universiti Teknologi PETRONAS
and Universiti Kuala Lumpur for their necessary and valuable supports. basic ideas involved in MPC design are: predict future plant
response using a process model and minimize a finite horizon
objective function comprising a sum of the future predicted Process.However, this will fail to achieve a steady state offset
errors and the control moves. of the controlled outputs. This problem is investigated in this
paper.
Condenser
2. PROCESS MODELLING AND SIMULATION
Reflux Drum
Input u1 Distillate Flow D The distillation column in this paper is a part of a petroleum
project to build a condensate processing plant to raise the
Reflux Rate L Output xD
Feed Flow F utility value of condensate. The nominal capacity of the plant
is 130,000 tons of raw condensate per year based on 24
operating hours per day and 350 working days per year. The
Boilup Rate V
quality of the output products is the purity of the distillate, xD,
Input u2 higher/equal than 98% and the impurity of the bottoms, xB,
Bottom Flow B less/equal than 2%. The feed stock data and its compositions
Output xB are based on the reference of PetroVietnam Gas Company,
Reboiler
(1999).
Fig. 1. Distillation Flowsheet. The feed stock can be considered as a pseudo binary mixture
of Ligas (iso-butane, n-butane and propane) and Naphthas
A real plant usually has to work within certain input and (iso-pentane, n-pentane, and higher components). The
output limitations. And the MPC controller is very well column is designed with 14 trays. The model is simplified by
suited to handle online of these constraints. Fig. 2 illustrates lumping some components together (pseudocomponents) and
the geometry of the MPC constrained optimum. modelling the column dynamics on these pseudocomponents
only as referred to in Kehlen, H. and Ratzsch, M. (1987).
Unconstrained Optimum
The vapor boilup V generated by the heat input to the reboiler
∆u (k + 1) Ji Q − BcB (t B − t F )
J i +1 is calculated as in Franks, R. (1972): V = B ,
λ
Constrained where QB: heat input; B: flow rate of bottom product; cB:
Optimum specific heat capacity; tF: inlet temperature; tB: outlet
temperature; λ : the latent heat or the heat of vaporization.
The latent heat at any temperature is described in term of the
latent heat at the normal boiling point as shown in Nelson,
∆u (k ) T
Feasible Region
W. (1982): λ = γλB , where λ : latent heat at absolute
TB
Receding horizon
prediction temperature T; λB : latent heat at absolute normal boiling
point TB, γ : correction factor obtained from the empirical
Fig. 2. MPC for the unconstrained and the constrained chart.
Once the process is ill-conditioned, i.e. it has a large Some major design parameters to determine the liquid holdup
process condition number (defined as the ratio between the on trays, column base and reflux drum are calculated mainly
maximum and the minimum singular value of the gain based on the references of Joshi, M. (1979), McCabe, W. and
matrix). It is difficult to control because the process gain is Smith J. (1976) and Wuithier, P. (1972): Velocity of vapor
strongly dependent on the input direction. For this reason, ρ − ρG
conventional MPC controllers cannot provide an adequate phase arising in the column: ωn = C L (m / s ) , where
ρG
performance with set point tracking and disturbance rejection
as mentioned in Qin J. and Badgwell T. (1997). The ρ L : density of liquid phase; ρG : density of vapor phase; C :
traditional method for controlling an ill-conditioned process correction factor depending flow rates of two-phase flows.
with the same number of manipulated inputs as controlled
outputs is to delete one or more controlled variables from the The actual velocity ω is normally selected that
control objective as referred in Grosdidier P. et al. (1988). ω = (0.80 − 0.85)ωn for paraffin vapor.
Multi-variable nonlinear MPC of an ill-conditioned
distillation column can also be solved by Jonas B. W. and The diameter of the column is calculated with the formula:
Jari M. B. (2005) using the quasi-ARMAX model with fuzzy
4Vm
logic. Minh V.T and Nitin A. (2005) also proposed some Dk = (m) , where Vm : the mean flow of vapor in
soften constraints to improve the robustness of Model 3600πω
Predictive Control for Ill-Conditioned Distillation the column.
5
International Journal of Systems Control (Vol.1-2010/Iss.1)
Minh et al. / Model Predictive Control of ... / pp. 4-12
6
International Journal of Systems Control (Vol.1-2010/Iss.1)
Minh et al. / Model Predictive Control of ... / pp. 4-12
( K n −1V ) ( K V + L + LF )
an , n −1 = , an.n = − n ,
M M
1
( L + LF )
xD an , n +1 =
0.9 M
0.8
Feeding Section: for tray 7÷8 or for n=8,
0.7
0.6
( K 7V ) ( K V + L + LF ) (L )
a8,7 = , a8.8 = − 8 , a8,9 =
0.5 M M M
0.4
for n=9,
0.3
0.2
( K 8V ) (K V + L) (L)
a9,8 = , a9.9 = − 9 , a9,10 =
0.1 M M M
xB
0
0 50 100 150 200 250 300 350
Time
Rectifying Section: for tray 7÷14 or for n=10÷15,
( K n −1 (V + VF )) ( K (V + VF ) + L )
Fig. 4. The Steady State Values of Concentration xn an , n −1 = , an.n = − n ,
M M
(L)
an , n +1 =
M
7
International Journal of Systems Control (Vol.1-2010/Iss.1)
Minh et al. / Model Predictive Control of ... / pp. 4-12
8
International Journal of Systems Control (Vol.1-2010/Iss.1)
Minh et al. / Model Predictive Control of ... / pp. 4-12
Processes usually operate with the desired target values or the −1% ≤ ∆u (k ) ≤ +1% of u(k) at steady state or
output setpoints. In some operations, the desired setpoint
values might change over time. Therefore, the target tracking
-0.0739 ≤ ∆u1 (k ) ≤ +0.0739 (m3 / h) and
is an important part of any controller design. The objective
function for the target tracking is to determine the feasible
steady-state outputs to which the regulator converges, that 0.0647 ≤ u2 (k ) ≤ 0.0647 (m3 / h) .
minimize their deviation from the desired target values.
Target tracking can be formulated as an optimization MPC The output constraints:
problem that uses the quadratic objective function to
minimize the deviation of the steady-state outputs and inputs
98% ≤ y1 (k ) ≤ 100% and 0% ≤ y2 (k ) ≤ 2% .
from the desired target values:
The amplitude constraints: Fig. 5 shows the tracking performance of the MPC controller
with the setpoint trajectory changes.
−100% ≤ u (k ) ≤ +100% of the steady state or
The MPC controller allows a tracking performance to step
0 ≤ u1 (k ) ≤ 14.7832 (m / h) and 3 trajectories changed within the output targets with less than
0.08% overshoot, no final tracking offset errors in the
outputs.
0 ≤ u2 (k ) ≤ 12.9354 (m3 / h) .
9
International Journal of Systems Control (Vol.1-2010/Iss.1)
Minh et al. / Model Predictive Control of ... / pp. 4-12
program: min J = ∑ {∈ k
T
Q ∈k +∆ukT R∆uk } .
∈k , ∆uk
k =0
5. MODIFIED MPC WITH ZONE REGION Compared to traditional methods for controlling ill-
conditioned processes which are to delete some of the
controlled variables, this modified MPC only changes some
Ill-conditioned process is difficult to control. The traditional setpoints into zones so that all controlled outputs remain in
method for controlling the ill-condition process is to delete the control objective.
some controlled variables from the objective function. This is
because that if some output setpoints are deleted, the system Output Step Disturbance Rejection
becomes looser and the probability that the MPC controller
can find a solution will increase (Qin, J. and Badgwell, T
The difference between the model and the actual plant can
(1997)) and Grosdidier, P. et al. (1998)). However, if the
cause the steady-state offset errors or the closed loop
outputs jump out of the designed regions, the controller is not
instability. In Fig. 6, the system is initially at the steady state
able to push them back since their constraints have been
and a step disturbance of 0.5 enters the output variable y1.
deleted.
The conventional MPC system becomes unstable because of
the plant-model mismatch. While with the zone region, the
In this paper, a new method for handling the ill-conditioned disturbance does not cause the outputs deviate from its zone
process has been considered – soft constraints for the outputs limits. No control action is taken because none of the control
once they jump out of the desired regions: As long as the objectives has been violated and the system remains stable
outputs still lie inside the designed regions, no control action
is taken but when the outputs violate the designed regions,
the control objective in the MPC regulator will activate the
soft constraints and push them back to the desired regions.
10
International Journal of Systems Control (Vol.1-2010/Iss.1)
Minh et al. / Model Predictive Control of ... / pp. 4-12
11
International Journal of Systems Control (Vol.1-2010/Iss.1)
Minh et al. / Model Predictive Control of ... / pp. 4-12
AUTHORS PROFILE
12