See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/235668177

Liquid level control of nonlinear Coupled Tanks System using linear model
predictive control

Conference Paper · October 2012

DOI: 10.1109/ICET.2012.6375434

CITATIONS READS

26 4,280

2 authors:

Muhammad Usman Khalid Muhammad Bilal Kadri

Pakistan Navy Engineering College PAF Karachi Institute of Economics & Technology
3 PUBLICATIONS   35 CITATIONS    62 PUBLICATIONS   397 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

HOSMC of ROVs View project

Design and development of Intelligent Mobile Robots (IMRs) for disaster mitigation and firefighting View project

All content following this page was uploaded by Muhammad Bilal Kadri on 16 May 2014.

The user has requested enhancement of the downloaded file.

 Liquid Level Control of Nonlinear Coupled Tanks
System using Linear Model Predictive Control

Muhammad Usman Khalid, Dr. Muhammad Bilal Kadri

Pakistan Navy Engineering College
National University of Sciences & Technology
Karachi, Pakistan
[email protected], [email protected]

Abstract—This paper investigates the usage of Discrete-time and For Multi-Input and Multi-Output (MIMO) configuration
Linear Model Predictive Control in controlling a nonlinear both of these tuning parameters can be selected independently
Coupled Tanks System. Two different schemes of Model for each input.
Predictive control are employed. To begin with, a basic Model
Predictive Control based on Generalized Predictive Control is Finally, simulation results are given to demonstrate the
used and then a Model Predictive Control approach based on performance achieved when both approaches are applied to
Laguerre functions. Simulation results have been included which Single-Input and Single-Output (SISO) nonlinear CTS. Also,
demonstrate the performance of both controllers when used to the DMPC using Laguerre functions approach is applied to
control Single-Input Single-Output Coupled Tanks System and MIMO nonlinear CTS.
the performance when Laguerre based Model Predictive Control
is applied to Multi-Input Multi-Output Coupled Tanks System.
II. THE COUPLED TANK SYSTEM
Keywords—Model Predictive Control, Laguerre functions, The coupled tank system consists of two vertical tanks
Nonlinear Coupled Tanks System, Generalized Predictive interconnected by a flow channel (Fig.1) which causes the
Control. levels of the two tanks to interact. Each tank has an
independent pump for inflow of liquid. The sectional area of
I. INTRODUCTION the outlets present and the base of each tank and the channel
connecting the two tanks can be varied with rotary valves. The
Model Predictive Control (MPC) has made a significant coupled tank system can be configured as a SISO or as a
impact on modern control engineering. It has found a wide MIMO system via manipulation of pumps input and sectional
range of applications in the process, chemical and food area of rotary valves.
processing industries. Model Predictive Control refers to a
specific procedure in controller design from which many kinds
of algorithms can be developed for different systems, linear or
nonlinear, discrete or continuous. The main difference in the
various methods of MPC is mainly the way the control problem
is formulated. One of the most popular methods of MPC is
Generalized Predictive Control (GPC). GPC was developed by
Clarke [1]. The idea of GPC is to calculate future control
signals in such a way that it minimizes a cost function defined
over a prediction horizon. GPC is capable of controlling
processes with variable dead-time, unstable and non-minimum
phase systems.
In this paper Discrete-time Model Predictive Control Figure 1. Layout of the coupled tanks system.
(DMPC) is used to control the liquid level of a nonlinear
Coupled Tanks System (CTS) in MATLAB, Simulink Considering mass balance, the dynamic equation of each
environment. At first, Model Predictive Control based on tank is developed.
Generalized Predictive Control [2] which is a restricted model
approach, is employed. Then a different approach using
(1)
Laguerre functions [4] is used. Laguerre functions when used
with DMPC have many benefits such as, the number of terms
used in the optimization problem can be reduced to a fraction (2)
of that required by the basic procedure, allows substantial
improvements in feasibility [5], two explicit tuning parameters where, H1 ,H2 are heights of liquid in tank 1 and tank 2
can be used for tuning the closed loop performance with ease respectively.A1 and A2 are cross-sectional areas of tank 1 and
tank 2.Q03 is the flow rate between tanks.Qi1,Qi2 are pump TABLE I. PARAMETERS OF THE COUPLED TANK SYSTEM
flow rate into tank 1 and tank 2 respetively.Q01 and Q02 are the Symbol Quantity Value
flow rate of liquid out of tank 1 and tank 2 respectively. A1,A2 tank section area 9350.10-6 m2
s1,s2,s3 channel sectional area 78.50.10-6 m2
By Bernoulli’s equation for a non-viscous, incompressible discharge coefficient of channel 1 and
a0 1
channel 2
fluid in steady flow,
a1 discharge coefficient of channel 3 0.5
(3) g gravitational constant 9.8 m/s2
(4) III. MPC ALGORITHM BASED ON GPC
(5) This basic MPC approach uses GPC to formulate the
where, α1, α2 and α3 are proportionality constants which control problem [2]. Considering a single-input and single-
depend on the coefficients of discharge, the cross-sectional output system:
area of each area and gravitational constant. By using values (14)
from (3) to (5) in (1) and (2) the nonlinear equations that
describe the dynamics of the multi-input and multi-output (15)
system are derived: where, u(k) is the manipulated variable, y(k) is the process
output and Xm is the state variable vector having a dimension
(6) of 2.
The system has u(k) as input. To design a predictive
(7) controller this needs to be altered, thus the model is augmented
with an integrator. Hence the state space form becomes:
For single-input and single-output configuration Qo1 and
Qi2 are zero, since the outlet valve of tank 1 is closed and (16)
liquid supply from pump 2 is also stopped.

(8)
(17)

(9) Or simply:

A. Linearized Perturbation Model (18)

Considering a small variation of q1 and q2 in both the (19)
control inputs respectively. Let h1 and h2 be the resulting
change in heights of the two tanks due to this variation. The where, 0m=[0 0…0] is a 1 x n vector, n being the order of the
linearized model can thus be derived: system. ΔXm(k+1)= Xm(k+1)-Xm(k) and augmented state
vector X(k)=[ ΔXm(k)T y(k) ]T .
1) Muli-Input and Multi-Output Linear Model
The future states of the system are calculated by using
(18), assuming current sampling instant ki, with ki >0 :

(10)
(20)
…

where, Nc is called the control horizon. It dictates the number

(11) of parameters used to capture the future control trajectory and
Np is the prediction horizon. It is also the length of the
optimization window.
2) Single-Input and Single-Output Linear Model
(12) Using (20) the output prediction can be found as:
(21)
(13) where,
,

and
 Thus at time k discrete Laguerre functions can be used to
capture the control trajectory:

(26)

where, m=0,1,2…Np ,
The value of x(ki+m) for single-input is found as:
(27)
φ= If the system is extended from single-input to multi-input
input. The control variable will be formulated as:
(28)
where, p is the number of inputs and
Now the prediction of future state at time m :
Let the data vector containing the set point information be (29)
RsT=[11 .. 1]r(ki). The cost function is now defined as:
where,
(22)
where, Rd=rw x INc x Nc with rw ≥0 is used as a tuning (30)
parameter for desired closed-loop performance.
IV. MPC USING LAGUERRE FUNCTIONS The cost function in terms of Laguerre parameter η is given
by,
Laguerre functions are orthonormal functions and can be
used to approximate the increments of control signals. First the
single-input case for MPC algorithm using Laguerre functions (31)
is described and is then extended to multi-input case. The z-
where,
transforms of discrete-time Laguerre functions are written as:

(23)
where, 0≤ a<1 is the scaling factor and N=1,2,.. is the number
of Laguerre terms.
Discrete-time Laguerre functions can be found by using After optimal value of η is found, the receding horizon
the following relation: control is realized.
Constrained solution can be found in terms of η by solving
(24) the quadratic programming problem:
Let the inverse z-transforms in vector form be, L(k). Using the (32)
realization (24), the discrete-time Lagueree functions can be
written in the form of difference equation with initial condition V. SIMULATION STUDIES
L(0): Simulation results show the performance comparison
between the basic MPC approach based on GPC and MPC
(25)
approach based on Laguerre functions when applied to SISO
where, nonlinear Coupled Tanks System. MPC approach based on
Laguerre functions is also applied to MIMO nonlinear CTS.
Prediction horizon for all simulations is set to 40.
A. SISO nonlinear CTS results
The reference level H2 used to evaluate the response of
MPC controllers applied to CTS is a pulse train whose
amplitude varies between 0 and 0.12 meter.

1) MPC approach results (Fig.2), (Fig.3)and (Fig.4) show

that in order to improve performance a long control horizon is
required and not much improvement is seen after the control
horizon exceeds the value 10. There are lots of undershoots
and overshoots in the plant response which are undesirable.
Irrespective of the valiue of the control horizon the overshoots
and undershoots are not eliminated.

Figure 4. The basic MPC algorithm with a control horizon of 30

Figure 2. The basic MPC approach with a control horizon of 2
2) Simulation result (Fig.5) indicates improvement in
performance when MPC with Laguerre functions is employed
for control of SISO CTS as compared to the performance of
basic MPC (Fig.4). The size of the undershoot and overshoot
are reduced considerably. This approach also reduces the
computational cost since there are only 4 parameters required
to capture the control trajectory. Parameter values for this
simulation are a=0.4 and N=4.

Figure 3. The basic MPC approach with a control horizon of 10

Figure 5. MPC based on Laguerre functions with scaling factor of 0.4

B. MIMO nonlinear CTS results CONCLUSION
The reference levels used for H1 and H2 are pulse train
with amplitude variations of 0 to 0.12m and 0 to 0.15m In this paper two MPC schemes are employed to control
respectively. Parameter values are a=0.4 and N=4. The nonlinear CTS having parameters given in Table 1. The
performance of the MPC with Laguerre functions with simulation results presented in this paper indicate that the MPC
MIMO plants outperforms the basic MPC approach. using Laguerre functions reduces computational cost and
Tighter control is observed throughout the simulation. performs better than the basic MPC approach using GPC.
Further aspects of the Discrete-time Model Predictive
Controller concerning the reduction of noise in steady state
without degradation of system performance and formulation of
control problem with constraints on the control variable
incremental variation and constraints on the amplitude of the
control variable, are under research.

REFERENCES

[1] D. W. Clarke, C. Mohtadi, and P. S. Tuffs, “Generalized predictive

control-part 1: The basic algorithm,” Automatica, 137-148 vol.23, 1987.
[2] Tan, K.K., Lee, T.H., Hunag, S.N. & Leu, F.M., “Pid control design
based on a gpc approach,” Industrial & Engineering Chemistry Research
41(8), 2002.
[3] D.Y. Valeryl, Predictive pid control of non-minimum phase systems,
Advances in pid control, InTech, Croatia, 2011.
[4] L. Wang, “Discrete time model predictive control design using laguerre
functions,”Proceedings of the American Control Conference,2430-2435
Vol.3, 2001.
[5] B. Khan and J. A. Rossiter, “Triple mode or laguerre mpc: a
comparison,”American Control Conference, 2011.
[6] A.Boubakir, F.Boudjema, C.Boubakir, and S.Labiod, “A fuzzy sliding
mode controller using nonlinear sliding surface applied to the coupled
Figure 6. Tank 1 level control using MPC based on Laguerre functions. tanks system,” International Journal of Fuzzy Systems, Vol. 10, 2008.
[7] J. Huang, L. Wang and Y. Huang , “Continuous time model predictive
control for a magnetic bearing system,”PIERS ONLINE, vol.3, No.2,
2007.

Figure 7. Tank 2 level control using MPC based on Laguerre functions.

View publication stats