Professional Documents
Culture Documents
Comparison Between PID and FOPID Controllers Based On Particle Swarm Optimization
Comparison Between PID and FOPID Controllers Based On Particle Swarm Optimization
Comparison Between PID and FOPID Controllers Based On Particle Swarm Optimization
net/publication/260417316
CITATIONS READS
6 8,209
2 authors:
All content following this page was uploaded by Mohamed Jasim Mohamed on 28 February 2014.
Abstract—The intelligent optimization method for designing solution. One of these methods is PSO which is an
Fractional Order ( PIλDδ ) controller (FOPID) based on Particle evolutionary computation technique. This technique combines
Swarm Optimization (PSO) is used in this paper . The Fractional social psychology principles in socio-cognition human agents
Order ( PIλDδ ) controller (FOPID) same as conventional PID and evolutionary computations. It depends on behaviors of
controller but integral order (λ) and derivative order (δ) are organisms such as fish schooling and bird flocking. It has
fractional. The new and good performance extension for FOPID some characteristics such as a simple concept, easy to
can be provided by Fractional calculus because of the flexibility implement, computationally efficient algorithm, flexible and
order of fractional calculus. The Fractional Order ( PIλDδ )
well-balanced mechanism to enhance global and local
controller (FOPID) and conventional PID controller are applied
to the three problems (stable, unstable and non-minimum phase exploration abilities and it has more efficiency than GA [6].
systems). The parameters of FOPID comprise proportionality
constant, integral constant, derivative constant, integral order (λ)
and derivative order (δ). The design of ( PIλDδ ) controller needs II. FRACTIONAL CALCULUS
to optimize five parameters while the design of conventional PID Fractional order system is a dynamical system that can be
controller needs only three parameters to optimize. Therefore, modeled by a fractional differential equation that is containing
the task of designing FOPID controller is more challenger than derivatives of non-integer order. Fractional order systems are
conventional PID controller. The design of controllers focuses on
useful in studying the anomalous behavior of dynamical
minimizing the integral square error (ISE) and integral time
square error (ITSE) criterion.
systems in electrochemistry, biology, viscoelasticity, and
IndexTerms—Fractional Calculus, ISE & ITSE Criteria, chaotic systems.
Particle Swarm Optimization, PID and FOPID or PIλDδ The fractional calculus is one of mathematical analysis that is
Controllers. used in different fields with many applications in physics,
engineering, mathematical biology, finance, life sciences, and
I. INTRODUCTION optimal control. The first appearance of fractional calculus is
One of the oldest control strategies is the PID control. It has before three centuries ago. In 1695, the derivative of the order
been widely used in the industrial control field because of its α=1/2 was described by Leibniz and L. Euler (1730) [7], it has
simplicity of design, good performance including low been developed up to nowadays. A list of mathematicians,
percentage overshoot and small settling time for slow whom have provided important contributions up to the middle
industrial processes [1,2]. Therefore, it is worth the care to of 20 century, includes P.S. Laplace (1812), J.B.J. Fourier
improve their quality and robustness. One of the methods that (1822), N.H. Abel (1823-1826), J. Liouville (1832-1873), B.
are used to improve conventional PID controller is to employ Riemann (1847), H. Holmgren (1865-67), A.K. Grunwald
fractional order controller with fractional order of I and D. In (1867-1872), A.V. Letnikov (1868-1872), H. Laurent (1884),
recent years, fractional calculus has been applied in the P.A. Nekrassov (1888), A. Krug (1890), J. Hadamard (1892),
modeling and control of various kinds of physical systems, as O. Heaviside (1892-1912), S. Pincherle (1902), G.H. Hardy
is well known and documented in many control theory or in and J.E. Littlewood (1917-1928), H. Weyl (1917), P. Levy
the literatures of applications. In FOPID controller beside the (1923), A. Marchaud (1927),H.T. Davis (1924-1936), A.
proportional, integral and derivative parameters (Kp, Ki, and Zygmund (1935-1945), E.R. Love (1938-1996), A. Erdelyi
Kd) it have two additional parameters the order of fractional (1939-1965), H. Kober (1940), D.V. Widder (1941), M. Riesz
integration λ and the order of fractional derivative δ. (1949)[8].
Therefore, it has five parameters that are made the FOPID Fractional calculus is a successful tool for describing
more flexible [3-5]. complex quantum field dynamical systems, dissipation, and
To get optimal controller, it must be found an optimal set of long-range phenomena that cannot be well illustrated using
values for Kp, Ki, Kd, λ and δ. In fact, there are many ordinary differential and integral operators [7].There are
powerful intelligent searching methods to find any optimal several different ways to define fractional order differential
The Second Engineering Conference of Control, Computers and Mechatronics Engineering (ECCCM2, 2014)
and fractional order integral. The generalized differintegrator 1 𝑥
aDx
-α
f(x) = 𝑥−𝑡 α -1
f(t) dt (9)
operator may be as: 𝛤(α) a
a a
𝑓 𝑥1 , 𝑡 𝑑𝑡𝑑𝑥1
𝑥 𝑥 𝑥 𝑥 𝑡−𝑎
=
a t
𝑓 𝑥1 , 𝑡 𝑑𝑥1 𝑑𝑡= a
𝑓 𝑡 t
𝑑𝑥1 𝑑𝑡= Where, [ ] means the integer part, the first method is
𝑥 derived from the Grunwald-Letnikov definition, this definition
a
𝑓 𝑡 𝑥−𝑡 𝑑𝑡 (6) is come from basic definition of derivative:
By applying this property again, 𝑓 𝑡 −𝑓 𝑡−
Df(t)=lim →∞ (15)
𝑥 𝑥1 𝑥2 𝑥 𝑥1
a a a
𝑓 𝑡 𝑑𝑡𝑑𝑥2 𝑑𝑥1 = a a
𝑥1 − Let us reckon explicit expressions for higher order derivatives.
𝑥 𝑥1 The first two cases are
𝑡 𝑓 𝑡 𝑑𝑡𝑑𝑥1 = a t
𝑥1 − 𝑡 𝑓 𝑡 𝑑𝑥1 𝑑𝑡 =
𝐷𝑓 𝑡 −𝐷𝑓 𝑡−
𝑥 𝑥1 𝑥 𝑥 1 −𝑡 2 D2f(t)=lim →∞ (16)
a
𝑓 𝑡 t
𝑥1 − 𝑡 𝑑𝑥1 𝑑𝑡 = a
𝑓 𝑡 𝑑𝑡 (7)
2
𝑓 𝑡 −𝑓 𝑡− 𝑓 𝑡− −𝑓 𝑡−2
− 𝑓 𝑡 −2𝑓 𝑡− +𝑓 𝑡−2
With α order of integral then; =lim →∞
=lim →∞
2
𝑥 𝑥1 𝑥 −𝑡 𝛼 −2 3 𝐷 2 𝑓 𝑡 −𝐷 2 𝑓 𝑡−
a a
𝑓 𝑡 1𝛼 −2 ! 𝑑𝑡𝑑𝑥1 = D f(t)=lim →∞
𝑓 𝑡 −2𝑓 𝑡− +𝑓 𝑡−2 𝑓 𝑡− −2𝑓 𝑡−2 +𝑓 𝑡−3
𝑥 𝑥 𝑥 −𝑡 𝛼 −2 2 −
2
a t
𝑓(𝑡) 1𝛼 −2 ! 𝑑𝑥1 𝑑𝑡= =lim→∞
𝑥 𝑥 𝑥 −𝑡 𝛼 −2 𝑥 𝑥−𝑡 𝛼 −1 𝑓 𝑡 −3𝑓 𝑡− +3𝑓 𝑡−2 −𝑓 𝑡−3
𝑓 𝑡 t 1𝛼 −2 ! 𝑑𝑥1 𝑑𝑡 = a 𝑓 𝑡 𝑑𝑡 (8) =lim→∞ (17)
a 𝛼 −1 ! 3
This can be rewritten with Euler’s gamma function [10]
The Second Engineering Conference of Control, Computers and Mechatronics Engineering (ECCCM2, 2014)
𝑓 𝑡 − 3𝑓 𝑡 − + 3𝑓 𝑡 − 2 − 𝑓 𝑡 − 3 = Where ak (k=0,1,2,….,n) are constant coefficients of the
3 FODE; αk, (k = 0, 1, 2, ….. , n) are real numbers and
𝑗 3
−1 𝑗
𝑓(𝑡 − 𝑗) (18) αn> αn-1 > αn-2 >……α1 > α0 ≥ 0 .
𝑗 =0
By taking Laplace transform to the linear Fractional-order
The numerical calculation formula of fractional derivative can Differential Equations (FODEs):
be achieved as [10]:
𝐿
ansαny(s)+an-1sαn-1y(s)+……..+a1sα1y(s)+a0sα0y(s)=u(s) (34)
α −𝛼
t-LDt f(t)≈ 𝑇 𝑏
𝑗 =0 𝑗
𝑓(𝑡 − 𝑗) (19) The transfer function is:
𝑌(𝑠) 1
L is the length of memory. T the sampling time always 𝐺𝑝 = = (35)
𝑈(𝑠) 𝑎 𝑛 𝑠 𝛼 𝑛 +𝑎 𝑛 −1 𝑠 𝛼 𝑛 −1 +𝑎 𝑛 −2 𝑠 𝛼 𝑛 −2 +⋯+𝑎 1 𝑠 𝛼 1 +𝑎 0 𝑠 𝛼 0
replaces the time increment h during approximation. The
weighting coefficients bj can be calculated recursively by [3]:
1+α III. STABILITY OF FRACTIONAL ORDER SYSTEMS
b0=0 ,bj= 1 − 𝑏𝑗 −1 , (j≥1) (20)
𝑗
The stability is the very fundamental and critical requirement
The second method α
is binomial coefficients. during control system design. In an integer order continuous-
𝑗 time linear time-invariant system is stable, if and only if, all
α α α α−1 α−2 …(α−j+1) poles of its characteristic polynomial have negative real parts.
0
=1, 𝑗
= (21)
𝑗! In other words, the system is stable, if all poles in the left half
It can be replaced by Euler’s Gamma function [4] side of the complex s-plane, while in fractional order, the
stability is not determined by only the locations of the poles in
α 𝛤 α+1
𝑗
= (22) the left half side. It depends on the fractional order that
𝑗 !𝛤 α−𝑗 +1
becomes more complex as shown below:
Then the equation (2) becomes: The characteristic equation of a general linear fractional
𝑡−𝑎 differential equation has the form
1 𝛤 α+1 𝑗 𝛼𝑖
𝑖=0 𝑎𝑖 𝑆 =0
α 𝑗
aDt f(t) = lim →0 −1 𝑓(𝑡 − 𝑗) (23) (36)
𝛼 𝑗 !𝛤 α−𝑗 +1
𝑗 =0 Where, αi is rational, the characteristic equation can rewritten
The derivative function of f(t) can be converted to Laplace as;
𝑖
transform by : 𝑛
𝑖=0 𝑎𝑖 𝑆 𝑚 = 0 (37)
L{0Dtα f(t)}=sα F(s)-[0Dtα-1 f(t)]t=0 (24)
1
Where, m is integer, α = and ai> 0
Where F(s) is the Laplace transform of f(t). 𝑚
The integral function of f(t) can convert to Laplace transform Fig.(1.a) shows the stability region of fractional order system
by[10,13] : when 0 <α <1, the stability region larger than integer order
system. Fig.(1-b) shows the stability region of fractional order
L{0Dt- αf(t)}=s- αF(s) (25) system when 1 < α < 2, the stability region less than integer
order system [15-18].
The some properties of fractional differintegral order .n, α≥0
i- DnDαf(t)=Dn+αf(t) (26) Im
n α α n
ii- D D f(t)=D D f(t) (27)
iii- α
D [f(t)+u(t)]=D f(t)+D u(t) α α
(28) Stable area Unstable
𝛼π area
iv- JαDαf(t)=f(t) (29)
2
where Jα is fractional integration
𝛼π Re
v- D -αf(t)=Jαf(t) (30)
𝑛 𝑛
2
vi- c𝐷𝑥 𝑓 𝑥 = (𝐷𝑐+ 𝑓) 𝑥 = 𝐷𝑥𝑛−𝑐 𝑓 𝑥 =
𝑑𝑛
𝑓 𝑥 (31)
𝑑(𝑥−𝑐)𝑛
𝑛 𝑛 𝑛
vii- x𝐷𝑐 𝑓 𝑥 = (𝐷𝑐− 𝑓) 𝑥 = 𝐷𝑐−𝑥 𝑓 𝑥 =
𝑑𝑛
𝑓 𝑥 [10,12] (32)
𝑑(𝑐−𝑥)𝑛
0 <α<1
Linear Fractional-order Differential Equations (FODEs) with
n term in time domain is given by [14]: Fig.1.a
αn αn-1 α1 α0
anD y(t)+an-1D y(t)+……..+a1D y(t)+a0D y(t)=u(t) (33)
The Second Engineering Conference of Control, Computers and Mechatronics Engineering (ECCCM2, 2014)
In time domain, the control function is the sum of these three
Im actions.
𝑑𝑒 (𝑡)
𝛼π 𝑈(𝑡) = 𝐾𝑝 𝑒(𝑡) + 𝐾𝑖 𝑒 𝑡 𝑑𝑡 + 𝐾𝑑 (38)
𝑑𝑡
Unstable
Stable
2 area The transfer function of PID controller in s-domain is;
area U(S) 1 Kd s 2 + Kps + Ki
Gc(s) = = Kp + Ki + Kd s = (39)
𝑒(𝑆) 𝑠 𝑠
Re The FOPID controller at first time was proposed by Podlubny
𝛼π in 1999. It is expansion of the conventional PID controller
2 based on fractional calculus.
Kp e(t)
W + E Ki D-λ e(t) U Y
∑ Gp(s)
_
1 <α< 2 Kd Dδ e(t)
Fig.1.b feedback
Fig. 1. Region of stability for fractional order system. Fig. 3. Generic closed loop control system with the fractional order
a. when 0 < α < 1 b. when 1 < α < 2 (PIλDδ ) controller (FOPID)
IV. FOPID Controller It is same as conventional PID controller when λ=1 and
The PID controller is a flexible feedback controller for many δ=1, it is PI controller when δ=0, and it is PD controller when
applications, and some time it is called the three modes λ=0. The block diagram of a single input - single output closed
controller (three terms controller) because it has three loop control system with fractional (PIλDδ) controller (FOPID)
parameters in its building structure. The values of these is illustrated in Fig.(3) [3,4,9,19].
parameters are depending on the required performance index. The differential equation of the PIλDδ controller is described
One of these performance indices is the integral of square by;
error between a measured process variable and a desired set 𝑈 𝑡 = 𝐾𝑝 𝑒 𝑡 + 𝐾𝑖 𝐷 −𝜆 𝑒 𝑡 + 𝐾𝑑 𝐷𝛿 𝑒(𝑡) (40)
point. The developed controller tries to minimize the integral
of square error and so improve dynamic response. The values Where, λ and δ are fractional orders. By taking Laplace
of the parameters of the three actions can be interpreted in transform for the above equation, the controller equation
terms of time: ( P ) depends on the present error, ( I ) depends becomes;
on the accumulation of past errors, and ( D ) is a prediction of
U(s)=Kp e(s) +Ki s-λ e(s) + Kdsδ e(s) (41)
future errors. The PID controller is the sum of these three
λ δ
actions. Table (I) shows the effect of the parameter of each The PID controller is a special case of PI D controller,
action on time response and stability of the system [19]. where λ = δ =1.The conventional PID controller is represented
by fourpoints (P, PI , PD , PID), while fractional PI λDδ
Table I. Effects of increasing the value of each parameter of conventional controller is expanded to plane. This expansion adds more
PID controller. flexibility and accuracy, Fig.(4) displays this concept
Rise Steady-
Parameter
time
Overshoot Settling time
state error
Stability [3,4,9,18].
Kp
Ki
Decrease
Decrease
Increase small change
Increase Increase
Decrease
Eliminate
Degrade
Degrade
λ
Kd
Minor
change
Decrease Decrease
No effect in Improve
theory if Kd small (0,2) (2,2)
The block diagram of a single input-single output with unity
feedback closed loop control system is shown in Fig.(2),
where Gc(s) is the transfer function of the controller , Gp(s) is
the transfer function of the plant ,W is the required value, E is
(0,1) (1,1)
the error between the value of desired input and the value of PI PID
real output, U is the control value and Y is the real value of
output.
W + E Gc(s) U Gp(s) Y (1,0) (2,0)
_
P PD δ
feedback
Fig. 2. The feedback control loop
Fig. 4. Generalization of FOPID Controller: From point to plan
The Second Engineering Conference of Control, Computers and Mechatronics Engineering (ECCCM2, 2014)
V. Particle Swarm Optimization (PSO)
PSO is an optimization method based on stochastic technique
that is developed by Dr. Eberhart and Dr. Kennedy in 1995 Start
[20, 21]. It is inspired from intelligent of the swarm and is
based on the movement behavior of the flock of bird and fish
for research about food. The birds are searching about food in
the form of either scattered or go together. They know the Generate random
distance of the food but don’t know the location of the food. initial swarm
The birds have positions and velocities while birds fly to (position and velocity)
search for the food from one place to another. The good
information is translated immediately from one bird to the
swarm at any time. This information comes from good sense
of smell for birds which leads the swarm toward the food. The Calculate the fitness for
movement of bird from one place to another is equal to each particle (using simulink)
development of the solution in PSO [21].
In PSO the particle is same as bird in swarm; it has
position, velocity, and distance from optimal solution. The
PSO has some advantages; it is simple, easy to implement, and Calculate thePbest for each
the algorithm can be used in wide range of the fields such as particle and gbest for swarm
function optimization, the model classification, machine
study, neutral network training, the signal processing, vague
system control, automatic adaptation control and etc...[21].
PSO algorithm consists of 'n' particles, each particle existing Update the position and velocity
in a 'D' dimensional solution space. The condition of each according to Pbest and gbest
particle is changing according to three principles:
1- To keep inertia of particles.
2- To change the condition according to most optimize
position of particle.
3- To change the condition according to most optimize If
position of swarm. the maximum No
The new position of each particle is affected by the most iterations reached
optimized position of its previous positions which they passed
during its movement and the most optimized position of
particles in its surrounding.
Each particle has position and velocity that are both Yes
changing because of movement. The new position and
velocity of particle can be calculated by two equations below
[21,22]:
Stop
Output Responses
1
Example 1.
The open loop third order stable system is described as [27]; 0.8
1
𝐺(𝑠) = (44) 0.6
𝑠+3 (𝑠 2 +2𝑠+2)
Criterion Method Kp Ki Kd λ δ of
criterion size
Iteration
0.8 PID with - - -
1 1 2.125728 30 1000
GA[27] 0.65490.48230.3176
PID with - - -
0.6 ISE PSO 0.65790.48510.3149
1 1 2.124423 10 150
FOPID
- - -
0.4 with 1.124 1.379 2.0408 10 150
0.98320.38850.2406
PSO
PID with - - -
0.2 1 1 1.126229 30 1000
GA[27] 0.81170.58430.3725
PID with - - -
1 1 1.1242 10 150
0 ITSE PSO 0.81500.58950.3740
0 2 4 6 8 10 12 14 16 18 20 FOPID
Time (sec) - - -
with 0.979 1.185 1.0512 10 150
0.99240.59780.4200
Fig.7.a. ISE criterion PSO
The Second Engineering Conference of Control, Computers and Mechatronics Engineering (ECCCM2, 2014)
1.2 Table IV. The optimal PID and FOPID controller’s parameters for example 3
1 Criterion Method Kp Ki Kd λ δ
Value of Number of
FOPID
criterion Iteration
0.8
PID PID with
0.6 10 0.82352 10 1 1 0.0756874 1000
GA[27]
Output Responses
-0.8
0 2 4 6 8 10 12 14 1.4
Time (sec) FOPID
Fig.8.a. ISE criterion PID
1.2
1.2
1
1
Output Responses
0.8 FOPID
0.6 PID
0.8
0.4
Output Responses
0.2 0.6
0
-0.2 0.4
-0.4
-0.6 0.2
-0.8
-1 0
0 2 4 6 8 10 12 14
-1.2 Time (sec)
Fig.9.a ISE criterion
-1.4
0 2 4 6 8 10
Time (sec)
Fig.8.b. ITSE criterion 1.4
Fig.8. The step responses of optimal PID and FOPID controllers for FOPID
example 2. PID
1.2
Example 3.
1
Output Responses