The Design of The MISO Model Predictive Controller

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

TELKOMNIKA Indonesian Journal of Electrical Engineering

Vol.10, No.6, October 2012, pp. 1163~1170


 1163

The design of the MISO Model Predictive Controller for


Bioreactor
1 2 2 2
FanLiping* , Zhang Jun , Huang Xing , Huang Dong
1
College of Environment and Safety Engineering,Shenyang University of Chemical Technology, Shenyang,
110142, China,
2
College of Information Engineering, Shenyang University of Chemical Technology,
Shenyang, 110142, China
*corresponding author, e-mail: [email protected]*, [email protected]

Abstract
Bioprocesses have the inherent characteristics of high nonlinearity, parameter uncertainty and
information imperfection. These specific natures may cause some difficulties in control. Jacobian
linearization method was used in this paper to linearize the bioreactor model firstly, and then a Multiple-
input and single output (MISO) model predictive controller was designed for the reactor. PID controller and
fuzzy-PID controller were also introduced to compare with the MPC controller. Simulation results prove
that the model predictive control has several advantages over the other controls for bioreactors.

Keywords: bioreactor, linearization, model predictive control, rolling optimization

Copyright © 2012 Universitas Ahmad Dahlan. All rights reserved.

1. Introduction
As a high and new technology of life sciences, biotechnology is one of the fastest
growing technologies. The application and development of biotechnology has opened up
broad prospects for medical, pharmaceutical, food, environmental and some other
industries. Due to the needs of high production meeting product quality, process safety
and environmental regulation, control systems play a key role in chemical and
biochemical plants operation [1]. Monitoring and control of biological processes is
becoming more and more important [2-3].
Bioprocesses have complicated dynamics, they also are inherently concerned with
nonlinearity and non-stationarity, which make modeling and parameter estimation
particularly difficult, therefore their control is a challenging and delicate task [4-5].
Moreover, the scarcity of on-line measurements of the component concentrations makes
this task more sophisticated [6]. Hence, conventional control methods do not succeed in
such task [7].
It is well known that the design of high-performance model-based control
algorithms for biotechnological processes is hampered by two major problems which call
for adequate engineering solutions. First, the process kinetics is most often poorly
understood nonlinear functions, while the corresponding parameters are in general time-
varying. Second, up till now there has been a lack of reliable sensors suited to real-time
monitoring of the process variables which are needed in advanced control algorithms. The
control of biotechnological processes has been an important problem attracting wide
attention [8]. The main engineering motivation in applying control methods to such
processes is to improve operational stability and production efficiency [9-10].
Model predictive control (MPC) is an optimization strategy for the control of
constrained dynamic systems [11-12]. MPC uses multi-step prediction, rolling optimization
and feedback correction control strategies [13], so it can not only give a good control
effect and strong robustness, but also have an advantage of less demand on the accuracy
of the model. MPC can handle the uncertainty, non-linearity and parallel nature
effectively, and can make it easy to deal with the various kinds constraints existing in
process controlled variables and manipulated variables. It is an effective method to solve
complex industrial process control [14-15]. Good progress has recently been made in
simplifying the implementation of MPC [16]. MPC technology can now be found in a wide

Received June 14, 2012; Revised September 9, 2012; Accepted September 20, 2012
1164  e-ISSN: 2087-278X

variety of application areas including chemicals, petroleum, metallurgy, food processing,


aerospace, machinery, wastewater treatment and some other industrial sectors [17].
Model predictive control is used in this paper to a continuous bioreactor with
uncertainty. Jacobian matrix linearization method is applied to linearize the nonlinear
biochemical reactor model and get a reduced linearization system model which is suitable
for controller designing. Furthermore, model predictive controller is designed and is
simulated in the MATLAB/Simulink simulation environment.

2. Model of bioreactor
The biochemical reactor is an essential unit operation in a wide variety of
biotechnological processes. Biochemical reactors are used to produce a large number of
intermediate and final products, including medical products, food, beverages, and
industrial solvents [18]. A biotechnological process with growth of a single microbial
population on a single growth rate limiting substrate is described as [19]:

dγx
 dt = µ( γ s ) γ x − Dγ x
dγ (1)
 s = − 1 µ( γ s ) γ x + D( γ sin − γ s )
 dt Yx / s
where
µ max γ s
µ( γ s ) = (2)
K s + γ s + γ s 2 / Ki

And ࢽ࢞ is biomass concentration, ࢽ࢙ is substrate concentration, ࢽ࢙࢏࢔ is the inlet substrate


concentration, D is dilution factor, µ max is the maximum specific growth rate, Ki、Ks and Yx/s are
inhibition constant, saturation constant and yield coefficient for cells based on substrate
consumed. The nominal values of the model parameters used in the process simulation are
given in Table. 1.
When the dilution rate D and the influent substrate concentrationࢽ࢙࢏࢔ are chosen as the
control variables and the output are chosen as substrate concentration ࢽ࢙ and the biomass
concentrationࢽ࢞ , the output curves can be simulated by the upper described model and are
shown in Figure 1.

3. Linearizing Bioreactor Model


Biological processes involve living organisms. The bioprocess advancement is
determined by the living cells capabilities and characteristics, the bioreactor performance
as well as by the cultivation media composition and the main parameters evolution. The
high metabolic network complexity inside the cells often determine very sophisticated,
non-linear growth and product formation kinetics, with further consequences on the
bioprocess behavior, but at the same time on the product quality and yield.
The high nonlinearity and parameter uncertainty exist in system models make it
rather difficult to design a control scheme using the non-linear model directly[20-21]. In
order to make the controller designing easy, the nonlinear model should be linearized to
get a linearization model.
Considering the non-linear system described as the following form

 x& = f ( x, u, t )
 (3)
 y = g ( x, u, t )
wherex(t) ∈ ℜ n is the state vector, and u(t) ∈ ℜ m is the control vector. Thus, in the case of the
initial value of x, u and the input time are determined, the following equation is obtained:

TELKOMNIKA IJEE Vol. 10, No. 6, October 2012 : 1163–1170


TELKOMNIKA IJEE e-ISSN: 2087-278X  1165

δ x& = Aδ x + Bδ u
 (4)
δ y = Cδ x + Dδ u

Table1. Model Parameters


State/Parameters Range Nominal values
γ s , gl − 1 0.1-2 -
−1
D,h 0-2.75 -
γ x , gl −1
0.1-19 -

µ m ax , h −1
0.24-0.36 0.31
−1
Yx/s , g g 0.4-0.6 0.5
−1
K s , gl 0.08-0.12 0.1
−1
K i , gl 40-60 40

γ sin , gl −1
10-30 20

7.5 0.9

0.8
7

0.7
6.5

0.6
rs/(g/l)
rx(g/l)

6
0.5

5.5
0.4

5
0.3

4.5 0.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t/h t/h

Figure 1. Curves of the model

0.8
linearized
non-linearized
0.7

0.5
rs/(g/l)

0.4

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t/h

Figure 2. Comparison of linearized model and the non-linearized model

The design of the MISO Model Predictive Controller for Bioreactor (Fan Liping)
1166  e-ISSN: 2087-278X

In which, the coefficient matrixA, B, C, D can be obtained by using Jacobian linearization


method:

∂ ∂
A= f '( x , u , t ), B = f '( x , u , t )
∂x ∂u (5)
∂ ∂
C = g '( x , u , t ), D = g '( x , u , t )
∂x ∂u

For the actual bioreactor system,both dilution rate and influent substrate concentration
have great impacts on the output (substrate concentration). From the practical point of view, the
influent substrate concentration should and could become a control variable. A controller with
dual inputs often can get better outputs than that with single input. So, a dual input MPC
controller is used to control the reactor in this paper.
For the biochemical reactor model described in equation (1), if the start point of the
linear system is chosen as follows: the dilution rate is D = 0.2h -1 , the biomass concentration is
γ x = 5gl−1 , the substrate concentration is γ s = 0.202gl−1 , and the dilution factor D and the inlet
substrate concentration γ sin are chosen as the control variables, and the substrate concentration
is chosen as the output, the linearized point is chosen at steady state, then the linearized state
space equation can be described as

x& = Ax + Bu (6)

where

 −1.327 ×10−5 1.617   −5 0 


A= , B =   (7)
 −0.4 −3.434 14.8 0.2 

The linearization model and the non-linearization model are compared in Figure 2. As
the curve shows, the linearization model can give a valid description of the real system. Thus
the linear model can be used to replace the original nonlinear model approximately.

4. Controller Design
The main idea of MPC algorithms is to solve an optimization problem in order to find the
control vector trajectory that optimizes the cost function over a future prediction horizon [22].
Different from the traditional optimal control strategy, the model predictive control uses the non-
parametric model based on impulse response as its internal model. By finding the optimal
solution of the performance indicator according to the historical information and the future input,
the future control action can be decided and the future output can be predicted [23-25]. The
performance indicator can be described as

N
J (k ) = ∑ y (k + j ) − r (k + j ) + λ 2 u (k + j − 1)
2 2
(8)
j =1

MPC state space matrix can be expressed as a general form as follow:

xm (k + 1) = Am xm (k ) + Bmu (k )
(9)
y (k ) = Cm xm (k )

where

TELKOMNIKA IJEE Vol. 10, No. 6, October 2012 : 1163–1170


TELKOMNIKA IJEE e-ISSN: 2087-278X  1167

 ∆x (k + 1)   A OT  B 
xm (k + 1) =  m  , A =   , Bm =   , C = [O 1] (10)
 y (k + 1) 
m
CA 1  CB 

For the bioreactor model described above, if the predictive length ischosen as N P = 10
,the control length is N c = 8 , and the Robustness and Rapidity is chosen as rw = 0.8 , then
MPC matrix of the state space and correlation matrix can be obtained as follows

 0.9971 0.1367 0   −0.3925 0.0014 


Am =  −0.0338 0.7068 0  , Bm =  1.2602 0.0169 
(11)
 −0.0338 0.7068 1   1.2602 0.0169 
Cm = [0 0 1]

The output matrix of this system is

Y = FX + φ∆U (12)

in which

F = [Cm Am Cm Am 2 Cm Am 3 ... Cm Am 20 ]T , ∆U = (φ T φ ) −1 (φ T RS − φ T FX ) (13)

where (φ T φ ) −1φ T RS corresponds to the set-point change,while −(φ T φ )−1φ T F corresponds to the
state feedback control within the framework of predictive control, and

Cm Bm 0 0 0 
C A B Cm Bm 0 0 
 m m m 
φ = Cm Am 2 Bm Cm Am Bm Cm Bm 0  (14)
 
 M 
 19 
Cm Am Bm Cm Am18 Bm Cm Am17 Bm 16
Cm Am Bm 

So far, all parameters appeared in the block diagram of MPC system are given, and the
MPC controller is finished designing.

5. Simulation and Results Analysis


To examine and certify the control effect of the proposed approach, simulation of the
bio-reactor using model predictive control is carried out. In the simulation process, the set point
output is 1, the input weight is 0, the rate weight is 0.1, the output weight is 1.2, the input
constraints u1 ∈ [0, 0.275] and u2 ∈[10,30] . The Simulink simulation model of the MPC control
and fuzzy-PID control based on the linear model of the bioreactor is shown in Figure 3.
The running curves of substrate concentration and dilution rate are shown in Figure 4.
When the PID controller is used, the three parameters are: KP=10, KI=3.5, KD=0.01. The output
under MPC is very different from which under Fuzzy-PID control and PID control. From the
comparison of the output curves, the model predictive control is proved to have better
robustness such as stability and performance of closed-loop system in the whole domain and
could deal with the constraints over time.

The design of the MISO Model Predictive Controller for Bioreactor (Fan Liping)
1168  e-ISSN: 2087-278X

Figure 3. MPC and fuzzy-PID controller of the bioreactor

0.35
pid
mpc
0.3 fuzzypid

0.25

0.2
rs/(g/l)

0.15

0.1

0.05

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t/h

1.2
pid
1 mpc
fuzzypid
0.8

0.6

0.4
D/h-1

0.2

-0.2

-0.4

-0.6
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t/h

Figure 4. Substrate concentration and dilution rate

TELKOMNIKA IJEE Vol. 10, No. 6, October 2012 : 1163–1170


TELKOMNIKA IJEE e-ISSN: 2087-278X  1169

It can be seen from Figure 4 that the designed model predictive control scheme can
handles multivariable control problems effectively and meet the demands for both precise and
fast tracking.

6. Conclusions
Appropriate model linearization method can facilitate the designing of the model
predictive controller. By using MPC controller to the bio-reactor, the system can not only have
fast response characteristic, but also have good steady-state behavior and strong robustness.

Acknowledgements
This work was financially supported by the National Natural Science Foundation of China
(61143007).

References
[1] Mailleret L, Bernard O, Steyer JP.Nonlinear Adaptive Control for Bioreactors with Unknown
Kinetics.Automatica, 2004; 40(8): 1379-1385.
[2] Xie J, Mao G, Zhang S, Zhang W. An Integrated Computing Environment for Bio-Molecular
Networks.Journal of Convergence Information Technology, 2010; 5(4): 164-171.
[3] Rivai M, Purwanto D, Juwono H, Sujono HA. Electronic Nose using Gas Chromatography Columnand
Quartz Crystal Microbalance. TELKOMNIKA Indonesian Journal of Electrical Engineering. 2011; 9(2):
319-326.
[4] Meszaros A, Voros J. Design of Intelligent Controllers for Biotechnological Process. Acta Chimica
Slovaca. 2008; 1(1): 208-220.
[5] Roman M, Selişteanu D. Nonlinear on-line Estimation and Adaptive Control of a Wastewater
Treatment Bioprocess. Electronics and Electrical Engineering. 2012; 117(1): 23-28.
[6] Renard F, Wouwer AV. Robust Adaptive Control of Yeast Fed-Batch Cultures.Journal of Computers
and Chemical Engineering. 2008; 32(6): 1238–1248.
[7] Horiuchi J. Fuzzy Modeling and Control of Biological Processes.Journal ofBioscience and
Bioengineering. 2002; 94(6): 574–578.
[8] Smets Y, Claes JE, November EJ, Bastin GP, Vanimpe JF. Optimal Adaptive Control of (Bio)
Chemical Reactors: Past, Present and Future. Journal of Process Control. 2004; 14(7): 795–805.
[9] Caramihai M, Dumitrache I, Chirvase AA. Progress in Characterization, Modeling and Control of
Aerobic Bioprocess. Revue Roumaine de Chimie. 2009; 54(8): 619–635.
[10] Dimitrova N, Krastanov M. Nonlinear Stabilizing Control of an Uncertain Bioprocess Model.
International Journal of Applied Mathematics and Computer Science. 2009; 19(3): 441–454.
[11] AshooriA, Moshiri B, Sedigh AK, Bakhtiari MR. Optimal Control of A Nonlinear Fed-batch
Fermentation Process Using Model Predictive Approach. Journal of Process Control. 2009; 19(7):
1162-1173.
[12] Liu D, Li M. Adaptive Predictive Control System with Disturbance Compensation based on Self-
recurrent Wavelet Neural Network. International Journal of Advancements in Computing Technology.
2011; 3(10): 330-338.
[13] Domınguez F, Pistikopoulos EN. Recent Advances in Explicit Multiparametric Nonlinear Model
Predictive Control. Industrial & Engineering Chemistry Research. 2011; 50(2): 609-619.
[14] Holkar S, Waghmare LM. An Overview of Model Predictive Control. International Journal of Control
and Automation. 2010; 3(4): 47-63.
[15] Lee J. Model Predictive Control: Review of the Three Decades of Development. International Journal
of Control, Automation and Systems. 2011; 9(3): 415-424.
[16] Alford JS. Bioprocess Control: Advances and Challenges. Computers and Chemical Engineering.
2006; 30(10): 1464–1475.
[17] Qin SJ, Badgwell TA. A survey of industrial model predictive control technology. Control Engineering
Practice. 2003; 11(7): 733–764.
[18] Rajinikanth V, Latha K. Identification and Control of Unstable Biochemical Reactor. International
Journal of Chemical Engineering and Applications. 2010; 1(1): 106-111.
[19] Tham HJ, RamachandranKB, Hussian MA. Sliding Mode Control for a Continuous Bioreactor.
Chemical and Biochemical Engineering Quarterly. 2003; 17(4): 267-275.
[20] Caramihai M, Dumitrache I, Chirvase AA. Progress in Aerobic Bioprocess Modeling and Control:
Case Study. Innovative Romanian Food Biotechnology. 2009; 3(9): 1-9.
[21] Harikrishna D, SrikanthNV. Dynamic Stability Enhancement of Power SystemsUsing Neural-Network
Controlled Static-Compensator. TELKOMNIKA Indonesian Journal of Electrical Engineering. 2012;
10(1): 9-16.

The design of the MISO Model Predictive Controller for Bioreactor (Fan Liping)
1170  e-ISSN: 2087-278X

[22] Jadlovska A, Kabakov N, Sarnovsky J.Predictive Control Design Based on Neural Model of a Non-
linear System. Acta Polytechnica Hungarica. 2008; 5(4): 93-109.
[23] Gondhalekar R, Jones CN. MPC of Constrained Discrete-Time Linear Periodic Systems-A
Framework for Asynchronous Control: Strong Feasibility, Stability and Optimality via Periodic
Invariance. Automatica. 2011; 47(2): 326-333.
[24] Wang L. Continuous Time Model Predictive Control Design Using Orthonormal Functions.
International Journal of Control. 2001; 74(16): 1588-1600.
[25] Ghasemi A. Application of Linear Model Predictive Control and Input-Output Linearization to
Constrained Control of 3D Cable Robots. Modern Mechanical Engineering. 2011; 1(2): 69-76.

TELKOMNIKA IJEE Vol. 10, No. 6, October 2012 : 1163–1170

You might also like