ISA Transactions 48 (2009) 273–282

Contents lists available at ScienceDirect

ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans

Design of nonlinear PID controller and nonlinear model predictive controller for
a continuous stirred tank reactor
J. Prakash ∗ , K. Srinivasan
Department of Instrumentation Engineering, Madras Institute of Technology, Anna University, Chennai-44, India

article info abstract

Article history: In this paper, the authors have represented the nonlinear system as a family of local linear state space
Received 7 September 2008 models, local PID controllers have been designed on the basis of linear models, and the weighted sum of
Received in revised form the output from the local PID controllers (Nonlinear PID controller) has been used to control the nonlinear
4 February 2009
process. Further, Nonlinear Model Predictive Controller using the family of local linear state space models
Accepted 9 February 2009
Available online 19 March 2009
(F-NMPC) has been developed. The effectiveness of the proposed control schemes has been demonstrated
on a CSTR process, which exhibits dynamic nonlinearity.
Keywords:
Crown Copyright © 2009 Published by Elsevier Ltd on behalf of ISA. All rights reserved.
CSTR
PID controller and model predictive
controller

1. Introduction A simple way to describe a nonlinear dynamic system using

PID controller and linear model predictive controller are
the two most popular control schemes that have been widely
implemented throughout the chemical process industries for the
past two decades. However, control of nonlinear system using
above linear control schemes don’t give satisfactory performance
at all operating points, the reason being that the process
parameters of the nonlinear process will vary with the operating
conditions. Moreover, the PID controller tuned at one operating
condition may not provide satisfactory servo and regulatory
performances at shifted operating points. It should be noted that,
to achieve improved closed loop performance a different set of
controller settings for each operating condition have to be used.
In the case of model based control schemes, the accuracy of the
model will have a significant effect on the closed loop performance
of the control system. The multiple-linear models concept has been
used in the recent years for modeling of nonlinear systems [1].
In addition, multiple-linear model based approaches for controller
design [2–5] have attracted the process control community. A
plethora of multiple-model adaptive control schemes have been
proposed in the control literature [6–9]. Gao et al. [10] has
proposed a nonlinear PID controller for CSTR using local model
networks. Omar Galan et al. [11] have reported the real-time
implementation of multi-linear model based control strategies on
the laboratory scale process.
∗ Corresponding author.
E-mail address: prakaiit@rediffmail.com (J. Prakash).
0019-0578/$ – see front matter Crown Copyright © 2009 Published by Elsevier Ltd on behalf of ISA. All rights reserved.
doi:10.1016/j.isatra.2009.02.001
multiple linear models has been proposed by Takagi–Sugeno [12]
and it is being used in this paper to develop Nonlinear PID con-
troller and Nonlinear Model Predictive Controller. The proposed
control scheme consists of a family of controllers (Local Con-
trollers) and a scheduler. As suggested by Kuipers and Astrom [13],
either local PID controller outputs or the local PID controller pa-
rameters can be interpolated. In the case of interpolation of con-
troller parameters, the controllers’ structure have to be assumed
as homogeneous, whereas interpolation of controllers output does
not impose any such constraints. At each sampling instant, the
scheduler will assign weights for each controller and the weighted
sum of the outputs will be applied as an input to the plant in the
case of interpolation of local controller outputs.
As suggested, one can also apply operating regime approaches
to develop an operating regime based model that can be applied
in a model-based controller [14,15]. Since global information can
be applied to determine the control input at each sampling instant,
the nonlinear model based controller is expected to achieve better
control performance. Recently, stability analysis of a multi-model
predictive control algorithm with an application to the control
of chemical reactors has been reported by Leyla, Özkan and
Kothare, [16].
The key unit operation in chemical plants namely the continu-
ous stirred tank reactor (CSTR) exhibits highly nonlinear dynamic
behavior. Hence, there arises a need to develop computationally
non-intensive control schemes in order to achieve tighter con-
trol of strong nonlinear processes. A plethora of advanced con-
trol schemes such as neural adaptive controller [17], nonlinear
internal model control scheme [18] and fuzzy model predictive
274 J. Prakash, K. Srinivasan / ISA Transactions 48 (2009) 273–282

2. Takagi–Sugeno (T–S) fuzzy model

Nomenclature
Consider a nonlinear system represented by the following
CA Concentration (mol/l) nonlinear differential equations:
T Temperature (K)
−
qc Coolant flow rate (l/min) ẋ = f (x, u, d) (1)
q Feed flow rate (l/min)
−
CA0 Feed concentration (mol/l) y = g (x, u, d). (2)
T0 Feed temperature (K)
Eq. (1) describes a deterministic system evolution and can be
Tc0 Inlet coolant temperature (K) obtained from the material and energy balances of the process
V CSTR volume (l) under consideration. Eq. (2) describes the relationships between
hA Heat transfer term (cal/(min K)) the measurements and the state variables. In order to describe a
k0 Reaction rate constant (min−1 ) discrete nonlinear system, Eqs. (1) and (2) can also be functionally
E /R Activation energy term (K) represented in discrete form as
−1H Heat of reaction (cal/mol)
x(k) = f [x(k − 1), u(k − 1), d(k − 1)] (3)
ρ, ρc Liquid density (g/l)
Cp , Cpc Specific heats (cal/(g K)) y(k) = g [x(k − 1), u(k − 1)] (4)
x(k) True state variable where, x(k) is the system state vector (x(k) ∈ Rn ), u(k) is the
y(k) Measured variables system input/known deterministic input (u(k) ∈ Rm ), d(k) the
u(k) Process inputs unmeasured disturbance/unknown input (d(k) ∈ Rq ), and y(k) is
A State transition matrix (continuous domain) the measured variable (y(k) ∈ Rr ). The parameters k represents
B Input matrix (continuous domain) the sampling instant and the symbol f and g represent an n-
C Measurement matrix dimensional function vectors. We assume that measurements are
Ki Steady State gain of the ith process model made at discrete sampling instants with sampling period T . Note
K c ,i Proportional gain of ith PID controller that the d(t ) term described in Eq. (1) is assumed to be piecewise
T r ,i Integral time of ith PID controller constant for kT ≤ t < (k + 1)T
Td,i Derivative time of ith PID controller A T–S fuzzy model has been proposed to represent a nonlinear
NP Prediction horizon system using locally linearized models [12]. Two different
Nc Control horizon methods for developing a T–S fuzzy model have been suggested
WE Error weighting matrix (N-MPC) in the literature, namely (i) the black box identification via
fuzzy clustering technique [20] and (ii) Linearization of an
WU Controller weighting matrix (N-MPC)
existing nonlinear system around the centers of the fuzzy region
partitioning the state space. The T–S fuzzy model is nothing
Greek letter words
but a piecewise interpolation of local linear models through
Φ State transition matrix (Discrete domain) membership function. The T–S fuzzy model is described by IF-
Γ Input coupling matrix (Discrete domain) THEN rules, which represent local linear relations of the nonlinear
ξ Damping factor system. The rule to describe the nonlinear system around an
ωn Un-damped natural frequency operating point is as follows:
λ Tuning parameter (IMC-PID controller) Rule i (i = 1 : N )
If z1 (k) is Mi,1 and . . . and zg (k) is Mi,g then
xi (k) = Φi (x(k − 1) − x̄i ) + Γi (u(k − 1) − ūi ) (5)
control scheme [19] have been already attempted on the CSTR pro-
yi (k) = Ci xi (k) (6)
cess which is considered for the simulation study in this paper.
Even with the introduction of powerful nonlinear control strate- where, zj (k) are the premise variables and Mij (k) are the fuzzy sets.
Φi , Γi , and Ci are known time invariant matrices of appropriate
gies such as nonlinear internal model control schemes and neural
dimensions. In this work it is assumed that such a model of the
adaptive control, the proposed control schemes remain an attrac-
process can be developed from the first principles by linearizing
tive control strategy, because it offer advantages such as simple de- them around different operating steady state values (x̄i and ūi ). The
sign and low computational complexity. global system behavior is described by a fuzzy fusion of all linear
The main contributions of the paper are as follows: firstly, the model outputs. For a given input vector, u(k), the global state and
nonlinear system is represented as a family of local linear state output of fuzzy model are inferred as follows:
space models. Secondly, local PID controllers have been designed N
X
on the basis of local linear models, the weighted sum of the x(k) = hi (z (k))[Φi (x(k − 1) − x̄i )
output from local PID controllers has been used to control the i=1
nonlinear process, and finally a nonlinear model predictive control + Γi (u(k − 1) − ūi ) + x̄i ] (7)
scheme using the family of local linear state space models has been
y(k) = Cx(k) (8)
proposed to control nonlinear process.
where the membership grades hi (z (k)) are defined as
The organization of the paper is as follows. Section 2 discusses
the T–S fuzzy model. Section 3 presents the design of nonlinear PID µi (z (k))
hi (z (k)) = (9)
controller. Section 4 deals with nonlinear model predictive control µ(k)
schemes formulation using local linear models. Section 5 deals g
Y
with analytical (first principle) model based predictive control µi (z (k)) = Mij (10)
formulation. The process considered for simulation study has j =1
been discussed in Section 6. Simulation results are presented in
N
Section 7 and the conclusions drawn from the simulation studies
X
µ(k) = µi (z (k)). (11)
in Section 8. i=1
 J. Prakash, K. Srinivasan / ISA Transactions 48 (2009) 273–282 275

It should be noted that the grade of membership should be To account for plant model mismatch and unmeasured
disturbances, a simple unmeasured disturbance estimator similar
N
X to the dynamic matrix control scheme is incorporated as follows:
hi (z (k)) ∈ [0, 1] and hi (z (k)) = 1. (12)
i=1 yc (k + j/k) = y(k + j/k) + d(k + j/k) (17)
where
3. Nonlinear PID controller (N-PID) design using local linear
models d(k + j/k) = d(k/k) = (ym (k) − y(k)) for j = 1, . . . , Np . (18)
In the above Eq. (18), ym (k) represents the measured output
In this section, the design of local PID controllers on the basis at the kth instant and y(k) represents the model output at the
of local linear models, which were described in the previous kth instant. Given a future setpoint trajectory yr (k + j/k), (j =
section, is discussed. Further, the method to combine the local 1, . . . , Np ), the nonlinear model predictive controller design
PID controller outputs yielding a global controller output has been problem can be formulated as:
outlined. The global controller output u(k) has been determined by
the following rules: u(k/k) .min
. . u(k + Nc − 1/k) J (19)

Rule i (i = 1 : N ) where,
If z1 (k) is Mi,1 and . . . and zg (k) is M1,g then Np
X
J = [E (k + j/k)]T WE [E (k + j/k)]
K c ,i (Kc ,i ∗ Td,i )
ui (k) = Kc ,i (e(k) − e(k − 1)) + T e(k) + j =1
T r ,i T Nc −1
X
× (e(k) − 2 ∗ e(k − 1) + e(k − 2)) + ui (k − 1) (13) + [1u(k + j/k)]T Wu [1u(k + j/k)]
j =0
where, T is the sampling time. It should be noted that PID
E (k + j/k) = yr (k + j/k) − yc (k + j/k) (20)
controller could be designed to satisfy the stability, performance,
and robustness criteria for each local linear model. Kc ,i , Tr ,i and 1u(k + j/k) = u(k + j/k) − u(k + j − 1/k). (21)
Td,i are the proportional gain, integral time and derivative time Subject to the following constraints
values of the ith PID controllers determined using standard optimal
PID tuning methods. The global controller output is described by a uL ≤ u(k + j/k) ≤ uH for j = 0, . . . , Nc − 1 (22)
fusion of all linear PID controller outputs.
yL ≤ yc (k + j/k) ≤ yHfor j = 1, . . . , Np (23)
1u(k + Nc /k) = 1u(k + Nc + 1/k)
4. Nonlinear model predictive controller using local linear
models (F-NMPC) = · · · 1u(k + Np − 1/k) = 0̄. (24)
The resulting constrained optimization problem can be solved
In the proposed N-MPC formulation, at every sampling instant
using any standard optimization technique.
the fuzzy dynamic model (Refer Eqs. (7) and (8)) is used for
predicting the future behavior of the plant over a finite number
of future time steps, say Np which is called prediction horizon. 5. Nonlinear model predictive controller using first principle
A set of Nc future manipulated input moves {u(k/k), u(k + (analytical) model–A-NMPC
1/k) . . . u(k + Nc − 1/k)}(where Nc is called the control horizon)
are determined by constrained optimization with the objective of The objective of the A-NMPC is to calculate a set of future
minimizing the predicted deviation of the process output from control moves (Control horizon) by minimization of a cost function
the target over the prediction horizon as well as minimizing the on a moving finite horizon (Prediction horizon). The optimization
expenditure of control effort in driving the process output to target, problem is solved on-line, based on the predictions obtained
from a nonlinear model. It is possible to use different empirical
subject to pre-specified operating constraints. The proposed F-
nonlinear models for predictions in the controller, but the most
NMPC is implemented in a moving horizon framework, that is,
attractive approach is to use the first principle models [21]). In the
only u(k/k) is implemented at each sampling instant and the
analytical model based N-MPC formulation, given a sequence of
optimization is repeated at each sampling instant based on the
future control moves {u(k/k) · · · u(k + 1/k) · · · u(k + Nc − 1/k)},
updated information from the plant.
a Np step ahead output prediction using the first principle model
The fuzzy dynamic model developed in the Section 2 can
could be written as follows:
be used recursively to obtain multi-step prediction. Given a
sequence of future control moves {u(k/k) · · · u(k + 1/k) · · · u(k x(k + j + 1|k) = x(k + j|k)
+ Nc − 1/k)}, a Np step ahead output prediction can be written Z (k+j+1)T
F x(τ ), u(k + j|k), d̄ dτ ;
 
as follows: +
(k+j)T
N
j = 0, 1, . . . .Np − 1
X
x(k + j + 1/k) = hi z (k + j)[Φi [(x(k + j/k)) − x̄i ] (25)
i =1 y(k + j + 1/k) = Cx(k + j + 1/k); for j = 0, 1, . . . , NP − 1. (26)
+ Γu,i (u(k + j/k) − ūi ) + x̄i ] for j = 0, . . . , Nc − 1 (14)
To account for plant model mismatch and unmeasured distur-
N
X bances, a simple unmeasured disturbance estimator similar to the
x(k + j + 1/k) = hi z (k + j)[Φi [[x(k + j/k)] − x̄i ] dynamic matrix control scheme is incorporated as follows:
i =1

+ Γu,i (u(k + Nc − 1/k) − ūi ) + x̄i ] yc (k + j/k) = y(k + j/k) + d(k + j/k) (27)

for j = Nc , . . . , NP − 1 (15) where

y(k + j/k) = C x(k + j/k); for j = 1, . . . , NP . (16) d(k + j/k) = d(k/k) = (ym (k) − y(k)) for j = 1, . . . , Np . (28)
276 J. Prakash, K. Srinivasan / ISA Transactions 48 (2009) 273–282

Table 1 7.1. Fuzzy dynamic model for the CSTR process

Operating data for CSTR process.
Process variable Normal operating condition For fuzzy dynamic model design for the CSTR process, the
Process flow rate (q) 100.0 l/min coolant flow rate (qc ) has been chosen as the premise variable
Feed concentration (CA0 ) 1 mol/l and triangular membership functions have been used to partition
Feed temperature (T0 ) 350 K the input space qc . The universe of discourse is divided into five
Inlet coolant temperature (Tc0 ) 350 K
intervals which are defined by the linguistic variables, very low
CSTR volume (V ) 100 l
Heat transfer term (hA) 7 × 105 cal/(min K) [97 100], low [97 100 103], medium [100 103 106], high [103 106
Reaction rate constant (k0 ) 7.2 × 1010 min−1 109] and very high [106 109] respectively. Further, local model
Activation energy term (E /R) 1 × 104 K parameters (consequent part of T–S fuzzy model) are determined
Heat of reaction (−1H) −2 × 105 cal/mol by linearizing the nonlinear differential equations ((35) and (36))
Liquid density (ρ , ρc ) 1 × 103 g/l
around the centers of the fuzzy region partitioning the operating
Specific heats (Cp , Cpc ) 1 cal/(g K)
space of the system. The detailed design procedure of the fuzzy
dynamic model for the CSTR process has been reported in Senthil
Given a future setpoint trajectory yr (k + j/k), (j = 1, . . . , Np ), et al. [22]. The linear time invariant discrete state space models
the nonlinear model predictive controller design problem can be (Refer Eqs. (5) and (6)) for five different operating points of CSTR
formulated as: are:
u(k/k) .min
. . u(k + Nc − 1/k)J (29) Operating point: 1 (qc = 97C̄A = 0.0795 and T̄ = 443.4566)
where, 1.2040e−001 −3.1008e−003
 
Φ1 =
Np
X 1.5350e+002 1.4438
J = [E (k + j/k)]T WE [E (k + j/k)]
1.2927e−004
 
j =1 Γ1 = .
−9.6293e−002
Nc −1
X
+ [1u(k + j/k)]T Wu [1u(k + j/k)] Operating point: 2 (qc = 100, C̄A = 0.0885 and T̄ = 441.1475)
1.7133e−001 −3.2672e−003
j =0
 
Φ2 =
E (k + j/k) = yr (k + j/k) − yc (k + j/k) (30) 1.4362e+002 1.4733
1u(k + j/k) = u(k + j/k) − u(k + j − 1/k). 1.3035e−004
 
(31)
Γ2 = .
Subject to the following constraints −9.4559e−002

uL ≤ u(k + j/k) ≤ uH for j = 0, . . . , Nc − 1 (32) Operating point: 3 (qc = 103, C̄A = 0.0989 and T̄ = 438.7763)
2.2479e−001 −3.4252e−003
 
y ≤ yc (k + j/k) ≤ y
L
for j = 1, . . . , Np
H
(33) Φ3 =
1.3333e+002 1.5012
1u(k + Nc /k) = 1u(k + Nc + 1/k)
1.3074e−004
 
= · · · 1u(k + Np − 1/k) = 0̄. Γ3 = .
(34) −9.2643e−002
The resulting constrained optimization problem can be solved
Operating point: 4 (qc = 106, C̄A = 0.1110 and T̄ = 436.3091)
using any standard optimization technique.
2.8071e−001 −3.5731e−003
 
Φ4 =
6. Continuous stirred tank reactor (CSTR) 1.2254e+002 1.5270
1.3038e−004
 
The first principle model of the continuous stirred tank system Γ4 = .
−9.0506e−002
and the operating point data (Refer Table 1) as specified in the
paper titled Fuzzy Model Predictive Control by Huang et al. [19] Operating point: 5 (qc = 109, C̄A = 0.1254 and T̄ = 433.6921)
have been used in this simulation study. In the process considered
3.3941e−001 −3.7084e−003
 
for simulation study, an irreversible, exothermic reaction A → B Φ5 =
1.1123e+002 1.5504
occurs in constant volume reactor that is cooled by a single coolant
1.2913e−004
 
stream. The process is modeled by the following equations:
Γ5 = .
−8.8085e−002
dCA (t ) q(t )
 
−E
= (CA0 (t ) − CA (t )) − k0 CA (t ) exp (35)  
dt V RT (t ) C =
1 0
.
0 1
dT (t ) q(t ) (−1H )k0 CA (t )
 
−E
= (T0 (t ) − T (t )) − exp
dt V ρ Cp RT (t ) For the CSTR process considered for the simulation study Senthil
et al. [13] has shown that the linear dynamic model is not able
ρc Cpc
  
−hA
+ qc (t ) 1 − exp (Tc0 (t ) − T (t )) . (36) to capture the dynamic behavior of the CSTR process, whereas the
ρ Cp V qc (t )ρ Cp fuzzy dynamic model is able to capture the dynamic nonlinearity
The state x(t ) and input u(t ) vectors are given by x(t ) = [CA ; T ] adequately.
and u(t ) = [qc ].
7.2. Nonlinear PID controller (N-PID) design for CSTR process
7. Simulation studies
In this work, we have intended to interpolate five PID controller
In all the simulation runs, the process is simulated using the outputs. That is, for each local linear model described in the
nonlinear first principle model (Eqs. (35) and (36)). The true previous subsection, a PID controller has been designed. In order
state variables are computed by solving the nonlinear differential to prevent a sharp spike in the controller output, at the time
equations using differential equation solver in Matlab 6.5. of step change in the setpoint, the derivative of the measured
 J. Prakash, K. Srinivasan / ISA Transactions 48 (2009) 273–282 277

Fig. 1. Servo response of CSTR with F-NMPC, N-PID, and A-NMPC (a) Process output (b) Controller output.

Fig. 2. Histogram of computation time per sampling instant.

output has been used in the control law instead of a derivative Table 2 provides the values of the second-order transfer
of the error. That is, PV derivative type PID controller form has function model parameters such as process gain, damping factor
been implemented in this work (P and I on setpoint error and and un-damped natural frequency at different operating points.
D on Process Variable). Further, the tuning parameters of each The IMC based PID tuning procedure will yield the following
PID controller have been determined using the IMC [23] tuning controller parameters:
rules proposed by Morari and Zafiriou, [24]. The process transfer 2ξi 2ξi 1
function relating the reactor concentration to the coolant flow rate K c ,i = ; Tr ,i = ; Td,i = .
ωn,i Ki λ ωn,i 2ξi ωn,i
at all operating points has been found to be of the form:
It should be noted that using the model parameters reported in
Ki Table 2, we have obtained the controller parameters of each local
Gi (s) = ∀ i = 1 : 5.
s2 + 2ξi ωn,i s + ωn2,i PID controller. The PID controllers’ parameters at five different
278 J. Prakash, K. Srinivasan / ISA Transactions 48 (2009) 273–282

Table 2
Damping factor and un-damped natural frequency at different operating points.

Operating point Damping factor Freq. (rad/s) Process gain

At qc = 97, C̄A = 0.0795, T̄ = 443.4566 0.661 3.93 0.0028
At qc = 100, C̄A = 0.0885, T̄ = 441.1475 0.540 3.64 0.0032
At qc = 103, C̄A = 0.0989, T̄ = 438.7763 0.416 3.34 0.0037
At qc = 106, C̄A = 0.1110, T̄ = 436.3091 0.285 3.03 0.0043
At qc = 109, C̄A = 0.1254, T̄ = 433.6921 0.141 2.71 0.0052

Table 3
PID controllers’ parameters at different operating points.

Operating point Kc ,i Tr , i Td , i

119.4321
At qc = 97, C̄A = 0.0795, T̄ = 443.4566 λ
0.3367 0.1926
92.6928
At qc = 100, C̄A = 0.0885, T̄ = 441.1475 λ
0.2973 0.2546
67.4294
At qc = 103, C̄A = 0.0989, T̄ = 438.7763 λ
0.2491 0.3601
43.2812
At qc = 106, C̄A = 0.1110, T̄ = 436.3091 λ
0.1876 0.5792
19.1813
At qc = 109, C̄A = 0.1254, T̄ = 433.6921 λ
0.1037 1.3124

Table 4 NMPC schemes for CSTR have been developed with the sampling
ISE values of F-NMPC, N-PID, and A-NMPC for setpoint tracking.
time of 0.083 min, prediction horizon of NP = 5, and control
Sampling instants interval A-NMPC F-NMPC N-PID horizon of Nc = 1.The error weighting matrix and the controller
10 ≤ k ≤ 49 2.52e–05 3.68e–05 5.07e–05 weighting matrix used in the N-MPC formulation are WE =
50 ≤ k ≤ 79 8.84e–05 9.42e–05 1.47e–04 1e4 and WU = 0. The following constraints on the manipulated
80 ≤ k ≤ 120 1.55e–05 2.67e–05 3.10e–05 input (coolant flow rate) are imposed 95 < qc < 108.

Table 5 7.4. Servo performance

ISE values of A-NMPC, F-NMPC and N-PID in the presence of setpoint change and
load change.
The setpoint variations as shown in Fig. 1(a) have been
Sampling instants interval A-NMPC F-NMPC N-PID introduced for assessing the tracking capability of the proposed (i)
10 ≤ k ≤ 69 1.16e–05 1.18e–05 1.66e–05 F-NMPC formulation using the local linear models, (ii) proposed
70 ≤ k ≤ 119 1.17e–04 1.43e–04 2.03e–04 N-PID control scheme using multiple-linear PID controllers and
120 ≤ k ≤ 175 5.51-06 2.77e–06 1.11e–05 (iii) analytical/first principle model based NMPC(A-NMPC). From
the response, it can be inferred that, the F-NMPC formulation
based on local linear models, N-PID control scheme using multiple-
linear PID controllers and A-NMPC scheme are able to maintain
the reactor concentration at the setpoint. The ISE values of F-
NMPC, N-PID and A-NMPC are reported in Table 4. From Table 4,
it can be inferred that the ISE values of A-NMPC and F-NMPC have
been found to be considerably less than N-PID. The variation in
the controller outputs is presented in Fig. 1(b). The observations
(qualitative) of the above simulation study are as follows:
Both the proposed controllers and A-NMPC are able to maintain
the setpoint at the desired value. However, the performances of
F-NMPC and A-NMPC at all the operating points are found to be
better than N-PID, as there is less overshoot and settles to the
setpoint faster. The F-NMPC provides performance comparable to
that of A-NMPC.
Fig. 2 presents histograms of computation time at each
sampling instant obtained using A-NMPC and F-MPC. It can be
concluded that the computation time per iteration (Matlab 7.0,
Intel Core 2 Duo Processor-2.13 GHz) of the proposed F-NMPC
algorithm is in the range 0.01–0.7 s, whereas, for the A-NMPC
Fig. 3. Variation in feed temperature of CSTR. algorithm, the value is in the range of 0.25–1.6 s. For the N-PID the
computation time per iteration has been found to be in the range
operating points have been reported in Table 3. It should be noted 0.001–0.0016 s.
that the controller gain has been found to be the function of the The proposed F-NMPC helps to reduce the number of compu-
filter time constant lamda (λ). tations needed, compared to the rigorous model based NMPC (A-
NMPC). Also, in the A-NMPC, the nonlinear differential equations
7.3. Nonlinear model predictive controller for CSTR process have to be numerically integrated to obtain the predicted estimates
of the output variables. On the other hand, in the F-NMPC, although
A simple model predictive control for CSTR has been developed more matrices are needed, all of them have constant values, which
using the local linear models (Refer Section 7.1). F-NMPC and A- limit the calculation to (i) the determination of weights, which will
 J. Prakash, K. Srinivasan / ISA Transactions 48 (2009) 273–282 279

Fig. 4. Servo and regulatory responses of CSTR with F-NMPC, A-NMPC and N-PID (a) Process output (b) Controller output.

Fig. 5. Servo response of CSTR with N-PID for various values of filter time constant (a) Process output (b) Controller output.

be provided by the operating region membership functions, and
(ii) state propagation calculations of each model using the appro-
priate matrices and a weighted average of the local linear model
outputs. Since, it is not necessary to carry out numerical integra-
tion of nonlinear differential equations; the proposed F-NMPC ap-
proach has better implementation capabilities than the A-NMPC
approach. Note that the computation time of the F-NMPC for even
higher-order problems will be always less demanding, in compar-
ison to that of the rigorous model based NMPC(A-NMPC).
7.5. Servo-regulatory performance
Simulation studies have been carried out to demonstrate
the disturbance rejection capability of the proposed F-NMPC
formulation based on local linear models, analytical model based
NMPC (A-NMPC) and N-PID Controller at nominal and at shifted
operated points.
A step change in the feed temperature of magnitude 2 ◦ K (from
350 ◦ K to 352 ◦ K) has been introduced at the 10th sampling
280 J. Prakash, K. Srinivasan / ISA Transactions 48 (2009) 273–282

Fig. 6. Servo response of CSTR with F-NMPC for various values of prediction horizon (a) Process output (b) Controller output.

Fig. 7. Performance of N-PID in the presence of measurement noise (a) Process output (b) Controller output.

instants and the value has been maintained upto the 110th
sampling instants and is then brought back to 350 ◦ K. (Refer Fig. 3).
The ISE values are computed for A-NMPC, F-NMPC and N-PID and
are reported in Table 5. The ISE value of F-NMPC and A-NMPC have
been found to be considerably less than that of N-PID for the servo-
regulatory performance case also. The following observation can be
drawn from the simulation studies
• From 10th sampling instants to 70th sampling instants of
Fig. 4(a), it can be inferred that the controllers (A-NMPC, N-
PID and F-NMPC) are able to reject the disturbance quickly and
bring the reactor concentration back to the nominal value of
the setpoint. This part of the simulation demonstrates that the
controllers are able to reject the disturbance at the nominal
operating point.
• With the disturbance being persistent, a step change in the
setpoint has been introduced at 70th sampling instant and it
can be noted that both the controllers are able to maintain the
 J. Prakash, K. Srinivasan / ISA Transactions 48 (2009) 273–282 281

Fig. 8. Performance of F-NMPC in the presence of measurement noise (a) Process output (b) Controller output.

Table 6
Mean and standard deviation of the true value of the controlled variable for various values of alpha.
Alpha Sampling instants interval F-NMPC µ (σ ) N-PID µ (σ )

0.25 51 ≤ k ≤ 100 0.0989(3.28e–04) 0.0989(2.32e–04)

251 ≤ k ≤ 300 0.1107(7.15e–04) 0.1108(4.24e–04)
451 ≤ k ≤ 500 0.0989(3.57e–04) 0.0989(2.64e–04)
0.5 51 ≤ k ≤ 100 0.0988(4.31e–04) 0.0989(2.93e–04)
251 ≤ k ≤ 300 0.1107(8.09e–04) 0.1108(4.31e–04)
451 ≤ k ≤ 500 0.0988(4.48e–04) 0.0989(2.61e–04)
0.75 51 ≤ k ≤ 100 0.0988(4.58e–04) 0.0989(3.09e–04)
251 ≤ k ≤ 300 0.1107(7.15e–04) 0.1108(4.23e–04)
451 ≤ k ≤ 500 0.0988(3.59e–04) 0.0989(2.72e–04)

concentration at the setpoint, as evident from 70th sampling
instants to 110th sampling instants of Fig. 4(a).
• At 110th sampling instants a simultaneous step change in the
setpoint (Refer Fig. 4(a)) as well as a step change in the feed
temperature (Fig. 3) has been introduced and it can be inferred
that the performance of the controllers has been found to be
satisfactory. This part of the simulation demonstrates that the
controllers are able to reject the disturbance as well as maintain
the process variable at the setpoint. It should be noted that the
performance of F-NMPC is found to be better than A-NMPC and
N-PID.
7.6. Performance of N-PID for various values of filter time constant
In order to show the tradeoff between performance and
robustness of the proposed nonlinear PID control scheme, we
performed simulation studies for various values of filter time
constant (lamda). The closed loop responses for step changes in
the setpoint for various values of lamda are shown in Fig. 5(a).
The manipulated variable profiles for various values of filter time
constants are shown in Fig. 5(b). It should be noted that we have
used single tuning parameter (lamda) for all the local controllers.
However, for each local controller appropriate values of filter
constant can be chosen. As we increase the value of lamda, it was
observed that the responses have been found to be of over-damped
type (lamda 0.5 and 1.5). For lamda value equal to 0.25, it can
be observed that controller output was found to be aggressive, as
compared to other values of lamda (Refer Fig. 5(b)).
7.7. Performance of F-NMPC for various values of prediction horizon
In order to assess the effect of the prediction horizon, we
have performed simulation studies for various values of prediction
horizon. The closed loop responses to step changes in the setpoint
and for various values of prediction horizon are shown in Fig. 6.
In all the simulation runs, a control horizon of 1 is used. For the
process considered for simulation study the prediction horizon
seems not to have appreciable effect as shown in Fig. 6. The
setpoint tracking performance has been found to be almost the
same for all the values of prediction horizon.
7.8. Performance of N-PID and F-NMPC in the presence of measure-
ment noise
The performances of the proposed N-PID and F-NMPC control
schemes in the presence of measurement noise are shown in
Figs. 7 and 8 respectively. In both the control schemes, Gaussian
white noise of mean zero and standard deviation of 0.0012 mol/l
has been added to the true value of the process variable (reactor
concentration). A digital first order filter has been used to filter
the noisy process measurement and the control calculations are
282 J. Prakash, K. Srinivasan / ISA Transactions 48 (2009) 273–282
performed based on the filtered value of the process variable, in
case of N-PID and F-NMPC. From the Figs. 7 and 8, the performance
of both the control schemes has been found to be satisfactory.
The mean and the standard deviation of the true value of the
measured variable (concentration) for various values of alpha have
been reported in Table 6. As we reduce the value of alpha, the
controller action of both the control schemes has been found to be
smooth and the standard deviation of the controlled variable has
been found to be less.
8. Conclusions
In this paper, the authors have proposed a simple and
straightforward procedure for designing a Nonlinear PID control
(N-PID) scheme and Nonlinear Model Predictive Control scheme
(F-NMPC) using local linear models for the CSTR process, which
exhibits significant variation in the damping factor and un-
damped natural frequency. From the extensive simulation studies,
it can be concluded that the proposed controllers have good
setpoint tracking, disturbance rejection capabilities at nominal
and shifted operated points and robustness properties. Further,
the performance of the proposed nonlinear model predictive
control scheme using local linear models, has been compared
with nonlinear model predictive control using an analytical model.
From the extensive simulation study, it can be concluded that the
proposed F-NMPC helps to reduce the number of computations
needed, compared to the analytical model based NMPC. The
proposed model based control scheme (F-NMPC) can be considered
as an alternative to analytical model based control scheme (A-
NMPC).
References
[1] Johansen T, Murray-Smith R. Multiple model approaches to modeling and
control. London: Taylor and Francis Limited; 1997.
[2] Xue ZK, Li SY. Multi-model modeling and predictive control based on Local
model networks. Control and Intelligent Systems 2006;34(2):105–12.
[3] Arslan E, Camurdan MC, Palazoglu A, Arkun Y. Multi-model scheduling control
of nonlinear systems using gap metric. Industrial & Engineering Chemistry
Research 2004;43:8275–83.
[4] Boling Jari M, Seborg Dale E, Hespanha JP. Multi-model adaptive control of a
simulated pH neutralization process. Control Engineering Practice 2007;15:
663–72.
[5] Prakash J, Senthil R. Design of observer based nonlinear model predictive
controller for a continuous stirred tank reactor. Journal of Process Control
2008;18:504–14.
[6] Dougherty Danielle, Cooper Doug. A practical multiple model adaptive
strategy for multivariable model predictive control. Control Engineering
Practice 2003;11:649–64.
[7] Narendra KS, Balakrishnan J. Adaptive control using multiple models. IEEE
Transactions on Automatic Control 1997;4(2):171–87.
[8] Wojsznis WK, Blevins TL. Evaluating PID adaptive techniques for industrial
implementation. In: Proceedings of the 2002 American control conference.
2002. p. 1151–5.
[9] Tan W, Marquez HJ, Chen T, Liu J. Multi-model analysis and controller
design for nonlinear processes. Computers & Chemical Engineering 2004;28:
2667–75.
[10] Gao Ruiyao, O’dywer Aidan, Coyle Eugene. A Nonlinear PID control for
CSTR using local model networks. In: Proceedings of 4th world congress on
intelligent control and automation. 2002. p. 3278–82.
[11] Galan Omar, Romagnoli, Palazoglu Ahmet. Real-time implementation of multi-
linear model-based control strategies-An application to a bench-scale pH
neutralization reactor. Journal of Process Control 2004;14:571–9.
[12] Takagi T, Sugeno M. Fuzzy identification of systems and its applications to
modeling and control. IEEE Transactions on Systems Man Cybernetics 1985;
15:116–32.
[13] Kuipers B, Astrom K. The composition and validation of heterogeneous control
laws. Automatica 1994;30(2):233–49.
[14] Aufderheide B, Bequette BW. Extension of dynamic matrix control to multiple
models. Computers and Chemical Engineering 2003;27:1079–96.
[15] Tian Z, Hoo KA. Multiple model-based control of the Tennessee–Eastman
process. Industrial & Engineering Chemistry Research 2005;44:3187–202.
[16] Özkan Leyla, Kothare MV. Stability analysis of a multi-model predictive control
algorithm with application to control of chemical reactors. Journal of Process
Control 2006;16:81–90.
[17] Krishnapura VG, Jutan Arthur. A neural adaptive controller. Chemical
Engineering Science 2000;55:3803–12.
[18] Nahas EP, Henson MA, Seborg DE. Nonlinear internal model control strategy
for neural network models. Computers and Chemical Engineering 1992;
16(12):1039–57.
[19] Huang YL, Lou HH, Gong JP, Edgar Thomas. Fuzzy model predictive control.
IEEE Transaction on Fuzzy Systems 2000;8:5–77.
[20] Babuska R, Verbruggen H. Fuzzy set methods for local modeling and
identification in multi-model approaches to modeling and control. London:
Taylor and Francis Limited; 1997. p. 75–100.
[21] Minh Tran, Dimitrios K. Varvarezos Mohamad Nasir The importance of first-
principles model-based steady-state gain calculations in model predictive
control—a refinery case study. Control Engineering Practice 2005;13(8):
1369–82.
[22] Senthil R, Janarthanan K, Prakash J. Nonlinear state estimation using
fuzzy kalman filter. Industrial Engineering Chemistry Research 2006;45:
8678–8688.
[23] Panda Rames C, Yu Cheng-Ching, Huang Hsiao-Ping. PID tuning rules for
SOPDT system: Review and some new results. ISA Transactions 2004;43(2):
283–295.
[24] Morari, Zafiriou. Robust process control. Englewood Cliffs (NJ): Prentice Hall;
1989.