An efficient application of particle swarm optimization in model predictive control of constrained two-tank system

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 12, No. 4, August 2022, pp. 3540~3550
ISSN: 2088-8708, DOI: 10.11591/ijece.v12i4.pp3540-3550  3540
Journal homepage: http://ijece.iaescore.com
An efficient application of particle swarm optimization in model
predictive control of constrained two-tank system
Ahmad Kia Kojouri1
, Javad Mashayekhi Fard2
1
Department of Electrical Engineering, Islamshahr Branch, Islamic Azad University, Islamshahr, Iran
2
Department of Electrical Engineering, Sabzevar Branch, Islamic Azad University, Sabzevar, Iran
Article Info ABSTRACT
Article history:
Received Aug 28, 2021
Revised Feb 15, 2022
Accepted Mar 10, 2022
Despite all the model predictive control (MPC) based solution advantages
such as a guarantee of stability, the main disadvantage such as an
exponential growth of the number of the polyhedral region by increasing the
prediction horizon exists. This causes the increment in computation
complexity of control law. In this paper, we present the efficiency of particle
swarm optimization (PSO) in optimal control of a two-tank system modeled
as piecewise affine. The solution of the constrained final time-optimal
control problem (CFTOC) is derived, and then the PSO algorithm is used to
reduce the computational complexity of the control law and set the physical
parameters of the system to improve performance simultaneously. On the
other hand, a new combined algorithm based on PSO is going to be used to
reduce the complexity of explicit MPC-based solution CFTOC of the
two-tank system; consequently, that the number of polyhedral is minimized,
and system performance is more desirable simultaneously. The proposed
algorithm is applied in simulation and our desired subjects are reached. The
number of control law polyhedral reduces from 42 to 10 and the liquid
height in both tanks reaches the desired certain value in 189 seconds. Search
time and apply control law in 25 seconds.
Keywords:
Complexity reduction
Constrained final time-optimal
control
Constrained piecewise affine
system
Particle swarm optimization
This is an open access article under the CC BY-SA license.
Corresponding Author:
Javad Mashayekhi Fard
Department of Electrical Engineering, Sabzevar Branch, Islamic Azad University
Sabzevar, Iran
Email: mashayekhi@iaus.ac.ir
1. INTRODUCTION
According to the system modeling, it is possible to introduce piecewise affine (PWA) systems, a
particular hybrid system class. It is described by partitioning the vast input-state space into Polyhedron
regions and assigning a PWA equation to each of these regions. Discrete-time PWA systems are a very
effective tool for modeling most hybrid systems [1]. These systems have established themselves as a
powerful class for identification and approximation of generic nonlinear systems by multi-linearization at
equilibrium [2]. A practical method in designing controllers of nonlinear systems is optimal control concepts
in constrained and non-constrained processes by linear discrete-time models in the form of state space.
Study [3] presents a constrained discrete model predictive control strategy for an inside greenhouse
temperature. Given what has been said about the modeling advantages of most systems based on the PWA
class, recently, there has been strong interest in computing the optimal form-package controller for PWA
systems. These problems became known as the constrained final time-optimal control (CFTOC) [4], [5]. The
most important methods of analysis of this problem are multi-parameter programming, receding horizon
control (RHC), or model predictive control (MPC). MPC is an effective way to deal with constrained control
problems and has found many applications and advances in industry and academic research. This method

Int J Elec & Comp Eng ISSN: 2088-8708 
An efficient application of particle swarm optimization in model predictive … (Ahmad Kia Kojouri)
3541
requires that the next position of the variables be predicted based on the current position, the controller input,
and the process model. In other words, the sequence of control inputs that optimize the objective function is
computed and applied to the process. This control concept is called the MPC [6].
Boiler turbine process control was performed with three manipulated variables and three controlled
variables using MPC technique [7]. Optimal control sequence allows recalculation and feedback performance
in MPC whenever a new measurement arrives; this is known as MPC. The stability of the control system and
the fulfillment of conditions, constraints, and requirements during operation are ensured by the proper
formulation of the objective function. RHC has the complexity and high volume of online computing related
to optimizing and reducing system robustness due to the difference between the actual and MPC processes.
Optimal methods based on the use of multi-parametric linear programming (MP-LP) and multi-parametric
quadratic programming (MP-QP) were provided [8], [9].
The offline calculation of the optimal control rule for constrained discrete-time linear systems was
performed using these methods. The resulting rules were made available as a PWA function on the
polyhedrons. At present, explicit MPC techniques enable a standard method in controller design for nonlinear
processes that are modeled in the PWA form, creating a substitute for intelligent controller design methods
such as fuzzy logic and neural networks in high-performance applications [10]. Unfortunately, one of the
main problems is the increasing complexity of the control rule obtained by increasing the prediction horizon
and its effect on system performance. On the other hand, it is necessary to increase the prediction horizon for
the system's optimal performance. For linear systems with parametric uncertainty by the Lyapunov function,
the PWA controller is designed with low complexity [11]. Effective representation and approximation are
provided by in-depth learning to MPC of linear time invariant (LTI) systems. Theoretically, at least neurons
and hidden layers are considered [12]. A nonlinear robust MPC with input-dependent perturbations and states
and uncertainty is presented [13]. The MPC algorithm with PWA control rules is presented for discrete-time
linear systems in the presence of finite perturbation [14]. The online burden of computational of the linear
MPC can be transferred to offline by using multi-parameter programming, which is named explicit
MPC [15]. A flexibility algorithm is proposed to reduce the calculation volume in [15] that the designer can
balance time and storage complexities.
This is done by hash tables and the associated hash functions. Two modified controllers instead of
the standard MP-QP are used [16] to reduce the complexity of the multi-parameter programming of MPC.
The problem of reducing the complexity of explicit MPC for linear systems is considered by PWA
employing separating functions [17]. A semi-continuous PWA model based on the optimal control method
for the nonlinear system is proposed [18]. First, the nonlinear system is approximated by multi-linear
subsystems, and then the subsystems are incorporated into a PWA system and formulated as an optimal
control problem. A computational method for optimizing and controlling a two-tank system with three
control valves is presented [19]. Easy implementation and optimizability of complex objective functions,
with many local minimums are the main advantages of particle swarm optimization (PSO). Furthermore,
PSO can be used to search significantly wider range of candidate solutions. The dynamics of the tank system
are nonlinear. The linear model is considered, and the parameters are adjusted so that the difference between
the actual system and the model is minimized by solving the optimal control problem. PSO has been used to
solve the problem of constrained optimization [20]. Self-adaptive of particle swarm optimization (SAPSO) is
recommended to increase PSO performance. Theoretically, the convergence of the method has been
investigated. Considerable interest has recently been generated to use PSO in optimization and engineering
problems [21]. A new algorithm that combines model predictive control with the PSO is proposed for
optimal control of constrained direct current (DC-DC) power system modeled as piecewise affine [22]. A
modern strategy based on model reference command shaping (MCRS) for an overhead crane, with dual
pendulum component impacts. The existing MRCS calculation has been making strides with the PSO
calculation to diminish plan complexity and guarantee that synchronous alteration can be done with the
feedback controller [23]. The proposed hybrid algorithm incorporates social interaction and elitism
mechanisms from PSO into manta ray foraging optimization (MRFO) strategy [24]. A control strategy based
on the self-tuning method and synchronous reference frame with proportional integral (PI) regulator is
proposed to achieve optimal power quality in an independent microgrid. Particle swarm optimization is used
to adjust the parameters of the PI controller, which ensures flexible performance and superior power quality
[25]. The tuning method used for proportional integral derivative (PID) with derivative filter controller for
liquid slosh system by implementing PSO algorithm [26]. PSO algorithm is applied to optimize the surface
grinding process parameter in both rough and final grinding conditions [27]. This paper is organized as
follows; first, the CFTOC of PWA system is briefly described. Having introduced the two-tank optimal
control in section 3, in section 4, the application of PSO for the solution of the expressed problem is
discussed, and finally, the simulation results and conclusion are presented.

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 12, No. 4, August 2022: 3540-3550
3542
2. CONSTRAINED FINAL TIME-OPTIMAL CONTROL (CFTOC) PROBLEM AND SOLUTION
We will do with the constrained PWA systems as (1) [4], [10]:
x(t + 1) = fPWA (x(t), u(t)) ≔ Aix(t) + Biu(t) + fi (
x(t)
u(t)
) ∈ Di (1)
where t≥0, the domain D ≔ ⋃ Di
ND
i=1 of fpwa(.,.) is a non-empty compact set in Rnx+nu , ND<∞ is the number of
system dynamics and D ≔ ⋃ Di
ND
i=1 denotes a polyhedral partition of the domain D. i.e., the closure of Di is
Di
̅ ≔ {(
x
u
) ∈ Rnx+nu|Di
x
x + Di
u
u ≤ Di
0
} and int(Di) ∩ int(Dj) = ∅ ∀i ≠ j. The CFTOC problem for
piecewise affine system (1) in the form (2a), (2b) and (2c) [4], [22]:
JT
∗
(x(0)) = min
UT
JT(x(0), UT) (2a)
subject to {
x(t + 1) = fPWA(x(0), u(t)
x(t) ∈ χf (2b)
JT(x(0), UT) = lT(x(T)) + ∑ l(x(t), u(t))
T−1
t=0 (2c)
where JT(. , . ) is the cost function, l(.,.) the stage cost, lT(.) the final penalty, UT optimization variable
described as the input sequence UT = {u(t)}t=0
T−1
, T<∞ receding horizon and χf is a compact terminal target
set in Rnx. If the solution of the CFTOC problem is not unique, uT
∗
(x(0)) = {u∗(t)}t=0
T−1
decides one
realization from the set of the conceivable optimizers.
The CFTOC problem determine a set of initial state and feasible inputs as χT ⊂ Rnx(x(0) ∈ χT),
UT−t ⊂ Rnu(u(t) ∈ UT−t , t = 0, … , T − 1) respectively. The explicit closed form solution can be expressed
as u∗(t): χT → UT−t, t=0, …, T-1. The considered system is PWA (1), and the cost is based on 1, ∞ norm
[22].
𝑙(𝑥(𝑡), 𝑢(𝑡)) ≔ ‖𝑄𝑥(𝑡)‖𝑝 + ‖𝑅𝑢(𝑡)‖𝑝 (3a)
𝑙(𝑥(𝑇)) ≔ ‖𝑃𝑥(𝑇)‖𝑝 (3b)
where ‖. ‖p with p={1,∞} speak to the standard vector standard 1,∞. The arrangement of (2) with
aforementioned limitations is a time varying PWA work of the starting state x(0)∈ 𝒫i.
𝑢∗(𝑡) = 𝜇𝑃𝑊𝐴(𝑥(0), 𝑡) = 𝐾𝑇−𝑡,𝑖𝑥(0) + 𝐿𝑇−𝑡,𝑖 (4)
where t = 0, … , T, {𝒫i}i=1
Np
is the polyhedral partition of a set of feasible state x(0), χT = ⋃ 𝒫i
Np
i=1
, with the
closure of 𝒫i stated as 𝒫
̅i = {x ∈ Rnx|𝒫i
x
x ≤ 𝒫i
o
} [2].
If a receding horizon control strategy is used for closed-loop, the control rule is expressed as
time-variant PWA state-feedback of the form [4]:
𝜇𝑅𝐻(𝑥(𝑡)) ≔ 𝐾𝑇,𝑖𝑥(𝑡) + 𝐿𝑇,𝑖 𝑖𝑓 𝑥(𝑡) ∈ 𝒫𝑖 (5)
where i=1,…, Np and for t≥0, u∗
(t) = μRH(x(t)). The CFTOC problem can be presented and solved for any
selection of P, Q, R, albeit here it is assumed that the parameters T, Q, R, P and χf
are selected by the
following assumptions [3]. To avoid additional control actions in steering states to the origin (equilibrium
point), matrices R, Q are required to have a full column rank.
3. OPTIMAL CONTROL OF TWO-TANK SYSTEM
The two-tank [28] shown in Figure 1 is a basic benchmark model to investigate and analyze the
control issues for PWA system. The tanks are filled by pump acting on tank 1, continuously manipulated
from 0 up to a maximum flow Q1. A switching valve V12 controls the flow between the tanks. This valve is
assumed to be either completely opened or closed (V12=0 or 1 respectively). The VN2 manual valve controls
the nominal outflow of the second tank. It is assumed in the simulations that the manual valves, VN1 is always
closed and that VN2 is open. The liquid levels to be controlled are denoted by h1, h2 for each tank respectively.

3543
Figure 1. Two-tank system schematic
It consists of two tanks that connected to each other. We assumed that: i) the valves behavior is
linear, ii) the initial volume of liquid in tanks is zero, and iii) the inflow of liquid to the first tank is constant
and has the maximum value. The liquid volume of tank 1 is defined by a time varying equation as (6a) and
(6b):
V1(t) = V0,1 + (Q1 − Q1,2) × t (6a)
𝑉1(𝑡) = 𝐴1 × ℎ1(𝑡) (6b)
where V0,1 is initial volume of liquid in tank 1and Q1, Q1,2 are inflow and outflow liquid of tank1 (Q1,2 can be
defined as inflow of liquid to tank 2), A1 and h1 are base area and the time varying height of liquid in tank1
respectively. The (6) can be repeated for the tank 2 with similar definition as (7a) and (7b):
𝑉2(𝑡) = 𝑉0,2 + (𝑄1,2 − 𝑄2) × 𝑡 (7a)
𝑉2(𝑡) = 𝐴2 × ℎ2(𝑡) (7b)
By combining (6), (7),
ℎ1(𝑡) =
1
𝐴1
(𝑉0,1 + (𝑄1 − 𝑄1,2) × 𝑡) (8a)
ℎ2(𝑡) =
1
𝐴2
(𝑉0,2 + (𝑄1,2 − 𝑄2) × 𝑡) (8b)
The system is expressed as a discrete time model with a sample time (Ts=10s) by (9).
{
ℎ1(𝑘 + 1) = ℎ1(𝑘) +
𝑇𝑠
𝐴1
(𝑄1(𝑘) − 𝑘12𝑉12
∗
(ℎ1(𝑘) − ℎ2(𝑘))
ℎ2(𝑘 + 1) = ℎ2(𝑘) +
𝑇𝑠
𝐴2
(𝑘12𝑉12
∗
(ℎ1(𝑘) − ℎ2(𝑘)) − 𝑘𝑁2𝑉𝑁2
∗
ℎ2(𝑘)
(9)
This model can be defined as a piecewise affine system of the shape (1), with four modes, depicted as takes
after [28]:
− First Mode: 𝑉12
∗
open, h1≥hv
− Second Mode: 𝑉12
∗
open, h1≤hv
− Third Mode: 𝑉12
∗
closed, h1≥hv
− Fourth Mode: 𝑉12
∗
closed, h1≤hv

 ISSN: 2088-8708
3544
For example, the first mode of the system matrices is:
𝐴1 = [
0.9542 −0.0393
0.0941 0.9670
] 𝐵1 = [
0.0699 0
0 0
] 𝐶1 = [1 0] 𝐷1 = [0] 𝑎𝑛𝑑 𝑓1 = [
0.0164
−0.0164
]
The CFTOC problem of the presented PWA system is solved by MPT [29] based on MPC for the
prediction horizon=3, norm =1, 𝑄 = [
1 0
0 1
], R= 10−5
× [
1 0
0 1
] and the explicit PWA control law has 78
polyhedral regions as shown in Figure 2 and the performance of the closed-loop system from a given initial
condition is presented in Figure 3.
Using PSO algorithm, the considered purposes are going to be fulfilled simultaneously:
− The number of polyhedral of explicit MPC-based control law is minimized to reduce the complexity,
− The liquid reaches a certain height in tanks in a short time and desirable manner.
Figure 2. Controller partitions
Figure 3. Closed loop system performance

3545
4. PSO ALGORITHM APPLICATION FOR SOLVING DEFINED PROBLEM
Before long a short time later, a brief audit of PSO calculation is displayed, and after that PSO
application is explored to fathom the defined problem. The PSO could be a worldwide exploratory
optimization strategy, to begin with, proposed by Kennedy and Eberhart in 1995 [30]. It is made of swarm
insights and is based on inquiring about the behavior of feathered creatures and the angle of development.
The motion of each molecule is impacted by its best-known nearby position and is additionally coordinated
to the best-known positions within the look space, which are upgraded by other particles to discover superior
positions. Agreeing to Figure 4, the premise of the strategies is as follows: “Each molecule can appear with
its current speed and position, the foremost ideal position of each individual, and the foremost ideal position
of the environment” [22], [30]. By selecting the beginning populace Xi, Vi, the speed and position of each
molecule around the look space alter concurring to correspondence (10) [30]:
𝑋𝑖 = [𝑥1,𝑖 𝑥𝑖,2 ⋯ 𝑥𝑛,𝑖]
𝑉𝑖 = [𝑣1,𝑖 𝑣𝑖,2 ⋯ 𝑣𝑛,𝑖]
𝑉𝑖𝑑
𝑘+1
= 𝑉𝑖𝑑
𝑘
+ 𝑐1 × 𝑟1
𝑘
× (𝑉𝑖𝑑
𝐿𝑏𝑒𝑠𝑡
) + 𝑐2 × 𝑟2
𝑘
× (𝑉𝑖𝑑
𝐺𝑏𝑒𝑠𝑡
) (10a)
𝑋𝑖𝑑
𝑘+1
= 𝑋𝑖𝑑
𝑘
+ 𝑉𝑖𝑑
𝑘
(10b)
𝑉𝑖𝑑
𝐿𝑏𝑒𝑠𝑡
= 𝑝𝑏𝑒𝑠𝑡𝑖𝑑
𝑘
− 𝑋𝑖𝑑
𝑘
(10c)
𝑉𝑖𝑑
𝐺𝑏𝑒𝑠𝑡
= 𝑔𝑏𝑒𝑠𝑡𝑖𝑑
𝑘
− 𝑋𝑖𝑑
𝑘
(10d)
where in this uniformity, 𝑉𝑖𝑑
𝑘
and 𝑋𝑖𝑑
𝑘
independently stand for the speed of the molecule i at its k times and the
d-dimension amount of its position; 𝑝𝑏𝑒𝑠𝑡𝑖𝑑
𝑘
speaks to the d-dimension amount of the person i at its most
ideal position at its k times; 𝑝𝑏𝑒𝑠𝑡𝑖𝑑
𝑘
is the amount d-dimension of the swarm at its most ideal position. To
avoid a molecule from being distant from the look space, the speed of the made molecule is restricted in each
course between -vdmax and vdmax [22]. In the event that the number of vdmax is as well tall, the arrangement is
distant from the most excellent arrangement, something else; the ideal arrangement will be nearby. c1 and c2
appear the speed digit, which alters the length when flying to the biggest molecule within the entire gather
and the ideal single molecule. In the event that the number of vdmax is as well little, the molecule is likely
distant from the target field; in case the shape is as well huge, the molecule may fly all of a sudden to the
target field or past the target field.
Figure 4. The basis of evolutionary PSO algorithm
Reasonable figures for c1 and c2 can control the molecule flight speed and will not be a halfway ideal
arrangement. c1 is more often than not rise to c2, and they are rising to 2. r1 and r2 speak to an irregular story
and 0-1 could be an arbitrary number. As said some time recently, our modern point is utilizing PSO for
complexity lessening of unequivocal MPC-based control law by diminishment the number of its polyhedral
and setting the physical parameter of framework to move forward the framework execution at the same time.
In this manner, the taking after objective work has been defined: Fitness-Function =Number of polyhedral +
Output specifications.
X
i
Xi+1
Vi
Vi+1
VLbest
VGbest

 ISSN: 2088-8708
3546
Where yield determinations are decided as a summation of operational determinations such as
settling time, overshoot, undershoot, steady-state deviation, time consistent, and so forward. Presently, the
unequivocal controller gotten in past segment as a portion of PSO ought to be considered and the taking after
unused execution list is being defined:
JnewT
∗
= 𝑀𝑖𝑛(𝐹𝑖𝑡𝑛𝑒𝑠𝑠 − 𝐹𝑢𝑛𝑐𝑡𝑖𝑜𝑛) = 𝑀𝑖𝑛[𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑜𝑙𝑦ℎ𝑒𝑑𝑟𝑎𝑙𝑠 𝑜𝑓 𝜇𝑅𝐻(𝑥(0)
+ 𝑂𝑢𝑡𝑝𝑢𝑡 𝑠𝑝𝑒𝑐𝑖𝑖𝑐𝑎𝑡𝑖𝑜𝑛] (11a)
S.T{
u∗
= μRH(x(t)) = KT,ix(t) + LT,i if x(t) ∈ 𝒫i
Output = ∑ desired output characteristic
(11b)
It is used according to the flowchart shown in Figure 5.
Figure 5. Flowchart of PSO algorithm application
Start
Evaluate Objective functions for each particle
Sort the population & determine the (Xbest,i) (XGbest)
Iter=iter+1
Define Input data & Generation of initial
population randomly & define initial
velocity zeros
iter=1 first Iteration
Update the particle
Evaluate Objective functions for each particle
Sort the population & determine the (Xbest,i) (XGbest)
Iter < iter-max
Yes
NO
finish
Print the best solution
Satisfied the constraints
Added the penalty factor
Satisfied the constraints
Added the penalty factor Yes
Yes
NO
NO

3547
The issue is illuminated as:
− The required information such as time step, fluid stature, and calculation parameters such as the number
of population and a number of the cycles is considered.
− The initial population is made as follows [22]:
Xi = [height of valve1, cross section of valves, maximum height, base area of each tank]
initial population = [
X1
X2
⋮
XN pop
]
− The yield determinations of the framework are measured based on introductory parameters, at that point,
the control law (gotten by MPT) is connected, and the number of polyhedral districts is calculated. In the
long-run objective work is characterized as [22]:
𝐹𝑖𝑡𝑛𝑒𝑠𝑠 − 𝐹𝑢𝑛𝑐𝑡𝑖𝑜𝑛 = 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑜𝑙𝑦ℎ𝑒𝑑𝑟𝑎𝑙 + 𝑜𝑢𝑡𝑝𝑢𝑡 𝑠𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑎𝑡𝑖𝑜𝑛𝑠.
where we consider yield determinations can be expected as yield details ≜ settling time + certain stature
of liquid. The best arrangement among the entire population is decided, and the population is overhauled
based on (10a) and (10b).
− For the predefined number of iterations, step 3 is done repetitively.
The convergence conditions are explored, and at last the leading arrangement is appeared within the
output. Then number of iterations is 15, the number of populaces is 5, c1= c2=2 is considered. The ideal
parameters are compiled within the Table 1.
Table 1. The optimal parameters
Sampling time Height of valve 1 Cross-section of valves Maximum height Area of each tank Num P Time(S)
10 s 0.1 m 1.00E-05 0.5 0.001 m^2 10 189
T is the required time to reach a certain height of the liquid. As presented, the number of control law
polyhedral reduces from 42 to 10. In Figure 6, the output flow variation of the first tank is shown. These
changes are linear. After 100 seconds, output flow increases to 10 cubic meters. In Figure 7, the liquid height
in the first tank is shown. These changes are nonlinear and incremental. After 90 seconds, the liquid height of
the first tank reaches 50 meters. Figure 8 shows the output flow variation of the second tank. These changes
are linear. At the last moment, there was a change in the function of the output valve.
Figure 6. The output flow variation of tank 1 Figure 7. The liquid height in tank 1
Figure 9 shows the liquid height changes of the second tank. As you can see, this change is
nonlinear and incremental. The height of the liquid in both tanks reached a certain value in 189 seconds.
0 50 100 150 200
0
1
2
3
4
5
6
7
8
9
10
Across Section Of Valves
Time (S)
Water
Flow
Rate
(m
3
)
0 50 100 150 200
0
10
20
30
40
50
60
Steady State Level In First Tank
Time (S)
Height
Of
Water
(m)

 ISSN: 2088-8708
3548
Figure 10 shows the unique convergence of the objective function using the PSO algorithm. It converges
after 4 iterations. Figure 11 shows the controller partition with 10 regions. Concurring to the comes about, it
can be concluded that by tuning the physical parameters utilizing PSO, the number of polyhedral of
MPC-based control law is minimized; in this way, the related complexity of arrangement to CFTOC will be
diminished together with the change of yield details at the same time so that the fluid tallness in both tanks
reach craved esteem in 189 seconds.
Figure 8. The output flow variation of tank 2 Figure 9. The liquid height in tank 2
Figure 10. Convergence diagram of objective
function
Figure 11. Controller with minimum partitions
5. CONCLUSION
Several analytical methods have been used for the CFTOC solution. Their main disadvantage is the
computational complexity of the solution; therefore, the problem can be considered as NP-hard. The meta-
heuristic algorithm is used to solve NP-hard optimization problems that have strategies to escape the local
optimal solution and are applied to a wide range of problems. In general, the developments of meta-heuristic
methods are done with the study and inspiration of optimization in nature, such as particle swarm
optimization. By using PSO and an appropriate definition of the objective function, the complexity of the
MPC-based solution of CFTOC was reduced and the system performance was improved simultaneously. The
most massive advantage of the recommended method is that if the mentioned purposes were not in one
direction, we can define a multi-objective function to fulfill aims. According to the simulation results, it is
demonstrated that the number of polyhedral and the dependent complexity of CFTOC solution are reduced,
the system performance such as reaching the liquid height at a certain time is desirable and the obtained
0 50 100 150 200
0
1
2
3
4
5
6
7
8
9
10
Output Valve Of Second Tank
Time (S)
Water
Flow
Rate
(m
3
)
0 50 100 150 200
0
10
20
30
40
50
60
Steady State Level In First Tank
Time (S)
Height
Of
Water
(m)
2 4 6 8 10 12 14
10
15
20
25
30
35
40
45
Iteration
Number
Of
Polyhedral

3549
steady-state error reaches zero. The number of control law polyhedral reduces from 42 to 10. The liquid
height in both tanks reaches desired value in 189 seconds.
REFERENCES
[1] P. Grieder, F. Borrelli, F. Torrisi, and M. Morari, “Computation of the constrained infinite time linear quadratic regulator,”
Proceedings of the 2003 American Control Conference, 2003., vol. 6, no. ue. 4, pp. 4711–4716, 2003, doi:
10.1109/ACC.2003.1242467.
[2] G. Ferrari-Trecate, M. Muselli, D. Liberati, and M. Morari, “A clustering technique for the identification of piecewise affine
systems,” Automatica, vol. 39, no. 2, pp. 205–217, Feb. 2003, doi: 10.1016/S0005-1098(02)00224-8.
[3] H. Hamidane, S. El Faiz, M. Guerbaoui, A. Ed-Dahhak, A. Lachhab, and B. Bouchikhi, “Constrained discrete model predictive
control of a greenhouse system temperature,” vol. 11, no. 2, p. 1223, Apr. 2021, doi: 10.11591/ijece.v11i2.pp1223-1234.
[4] F. J. Christophersen, Optimal control of constrained piecewise affine systems. Verlag Berlin Heidelberg: Springer, 2007.
[5] M. Baotic, “Polytopic computations in constrained optimal control,” Automatica, vol. 50, no. 3–4, pp. 119–134, 2009.
[6] D. Q. Mayne, J. B. Rawlings, C. V Rao, and P. O. M. Scokaert, “Constrained model predictive control: Stability and optimality,”
Automatica, vol. 36, no. 6, pp. 789–814, Jun. 2000, doi: 10.1016/S0005-1098(99)00214-9.
[7] S. Kumar Sunori, P. Kumar Juneja, and A. B. Jain, “Model predictive control system design for boiler turbine process,”
International Journal of Electrical and Computer Engineering (IJECE), vol. 5, no. 5, pp. 1054–1061, Oct. 2015, doi:
10.11591/ijece.v5i5.pp1054-1061.
[8] A. Bemporad, M. Morari, V. Dua, and E. N. Pistikopoulos, “The explicit linear quadratic regulator for constrained systems,”
Automatica, vol. 38, no. 1, pp. 3–20, Jan. 2002, doi: 10.1016/S0005-1098(01)00174-1.
[9] P. Tøndel, T. A. Johansen, and A. Bemporad, “An algorithm for multi-parametric quadratic programming and explicit MPC
solutions,” Automatica, vol. 39, no. 3, pp. 489–497, Mar. 2003, doi: 10.1016/S0005-1098(02)00250-9.
[10] F. Borrelli, M. Baotić, A. Bemporad, and M. Morari, “Dynamic programming for constrained optimal control of discrete-time
linear hybrid systems,” Automatica, vol. 41, no. 10, pp. 1709–1721, Oct. 2005, doi: 10.1016/j.automatica.2005.04.017.
[11] L. Lu and M. Kvasnica, “Low-complexity stabilizing PWA controllers for linear systems with parametric uncertainties,” IFAC-
PapersOnLine, vol. 53, no. 2, pp. 7286–7291, 2020, doi: 10.1016/j.ifacol.2020.12.569.
[12] B. Karg and S. Lucia, “Efficient representation and approximation of model predictive control laws via deep learning,” IEEE
Transactions on Cybernetics, vol. 50, no. 9, pp. 3866–3878, Sep. 2020, doi: 10.1109/TCYB.2020.2999556.
[13] J. Kohler, R. Soloperto, M. A. Muller, and F. Allgower, “A computationally efficient robust model predictive control framework
for uncertain nonlinear systems,” IEEE Transactions on Automatic Control, vol. 66, no. 2, pp. 794–801, Feb. 2021, doi:
10.1109/TAC.2020.2982585.
[14] M. Zhao and X. Tang, “Robust tube-based MPC with piecewise affine control laws,” Abstract and Applied Analysis, vol. 2014,
pp. 1–14, 2014, doi: 10.1155/2014/358148.
[15] F. Bayat, T. A. Johansen, and A. A. Jalali, “Using hash tables to manage the time-storage complexity in a point location problem:
Application to explicit model predictive control,” Automatica, vol. 47, no. 3, pp. 571–577, Mar. 2011, doi:
10.1016/j.automatica.2011.01.009.
[16] Y. Chen, S. Li, and N. Li, “Complexity reduced explicit model predictive control by solving approximated mp-QP program,” in
2015 10th Asian Control Conference (ASCC), May 2015, pp. 1–6, doi: 10.1109/ASCC.2015.7244434.
[17] M. Kvasnica, J. Hledík, I. Rauová, and M. Fikar, “Complexity reduction of explicit model predictive control via separation,”
Automatica, vol. 49, no. 6, pp. 1776–1781, Jun. 2013, doi: 10.1016/j.automatica.2013.02.018.
[18] X. Zhu, C. Song, J. Zhao, and Z. Xu, “A semi-continuous PWA model based optimal control method for nonlinear systems,”
Processes, vol. 8, no. 2, p. 170, Feb. 2020, doi: 10.3390/pr8020170.
[19] S. Y. Sim, S. L. Kek, and K. G. Tay, “Optimal control of a coupled tanks system with model-reality differences,” in AIP
Conference Proceedings, 2017, vol. 1872, doi: 10.1063/1.4996669.
[20] B. Tang, Z. Zhu, and J. Luo, “A framework for constrained optimization problems based on a modified particle swarm
optimization,” Mathematical Problems in Engineering, vol. 2016, pp. 1–19, 2016, doi: 10.1155/2016/8627083.
[21] Hoda Pourhossein, A. Zare, and M. Monfared, “Hybrid modeling and PID-PSO control of buck-boost chopper,” Przeglad
Elektrotechniczny-Electrical Review, vol. 88, no. 8, pp. 187–191, 2012.
[22] B. Mohammadi, S. Busaleiky, R. Modarres, E. Yarionsorudi, M. Fojlaley, and S. Andik, “Designing optimal control of
constrained DC-DC power converter using particle swarm optimization,” International Journal of Technical Research and
Applications, vol. 2, no. ue. 1, pp. 79–83, 2014.
[23] H. I. Jaafar, Z. Mohamed, M. A. Ahmad, N. A. Wahab, L. Ramli, and M. H. Shaheed, “Control of an underactuated double-
pendulum overhead crane using improved model reference command shaping: Design, simulation and experiment,” Mechanical
Systems and Signal Processing, vol. 151, Apr. 2021, doi: 10.1016/j.ymssp.2020.107358.
[24] M. F. M. Jusof, S. Mohammad, A. A. A. Razak, N. A. M. Rizal, A. N. K. Nasir, and M. A. Ahmad, “Hybrid manta ray foraging—
particle swarm algorithm for PD control optimization of an inverted pendulum,” in Recent Trends in Mechatronics Towards
Industry 4.0, Lecture Notes in Electrical Engineering, vol. 730, Singapore: Springer, 2022, pp. 1–13.
[25] U. Sultana, S. H. Qazi, N. Rasheed, and M. W. Mustafa, “Performance analysis of real-time PSO tuned PI controller for
regulating voltage and frequency in an AC microgrid,” International Journal of Electrical and Computer Engineering (IJECE),
vol. 11, no. 2, pp. 1068–1076, Apr. 2021, doi: 10.11591/ijece.v11i2.pp1068-1076.
[26] M. Z. Mohd Tumari, A. F. Zainal Abidin, A. S. Rahimi A Subki, A. W. Ab Aziz, M. Salihin Saealal, and M. Ashraf Ahmad,
“Liquid slosh control by implementing model-free PID controller with derivative filter based on PSO,” Indonesian Journal of
Electrical Engineering and Computer Science (IJEECS), vol. 18, no. 2, pp. 750–758, May 2020, doi:
10.11591/ijeecs.v18.i2.pp750-758.
[27] T. M. Shin, A. Adam, and A. F. Z. Abidin, “A comparative study of PSO, GSA and SCA in parameters optimization of surface
grinding process,” Bulletin of Electrical Engineering and Informatics (BEEI), vol. 8, no. 3, pp. 1117–1127, Sep. 2019, doi:
10.11591/eei.v8i3.1586.
[28] J. Thomas, “Complexity reduction of robust model predictive controller for uncertain piecewise affine systems,” in Proceedings
of the 2011 American Control Conference, Jun. 2011, pp. 857–862, doi: 10.1109/ACC.2011.5991337.
[29] M. Kvasnica, Real-time model predictive control via multi-parametric programming: Theory and tools. VDM Verlag, 2009.
[30] J. Kennedy and R. Eberhart, “Particle swarm optimization,” in Proceedings of ICNN’95 - International Conference on Neural
Networks, vol. 4, pp. 1942–1948, doi: 10.1109/ICNN.1995.488968.

 ISSN: 2088-8708
3550
BIOGRAPHIES OF AUTHORS
Ahmad Kia Kojouri received the B.S. degree in Control engineering from the
Iran University of Science and Technology, Tehran, Iran, in 2003, The MS degree in control
engineering from Islamic Azad University, South Tehran branch, Iran in 2006, and his Ph.D. in
Control Engineering from Islamic Azad University, Science and Research Branch, Tehran,
Iran in 2013. He is currently an Assistant Professor in the Electrical Engineering Department at
Islamic Azad University, Islamshahr, Iran. His research interests include optimal and robust
control, process control, and multivariable control systems. Email: kia@iiau.ac.ir.
Javad Mashayekhi Fard received the B.S. degree in power engineering from the
Islamic Azad University, Bojnourd, Iran, in 2003, The MS degree in control engineering from
Islamic Azad University, South Tehran branch, Iran in 2006, and his Ph.D. in Control
Engineering from Islamic Azad University, Science and Research Branch, Tehran, Iran in
2013. He is currently an Assistant Professor in the Electrical Engineering Department at
Islamic Azad University, Sabzevar, Iran. His research interests include optimal and robust
control, process control, and multivariable control systems. Email: mashayekhi@iaus.ac.ir.

An efficient application of particle swarm optimization in model predictive control of constrained two-tank system

More Related Content

Similar to An efficient application of particle swarm optimization in model predictive control of constrained two-tank system (20)

More from IJECEIAES (20)

Recently uploaded (20)

An efficient application of particle swarm optimization in model predictive control of constrained two-tank system