2013 13th Iranian Conference on Fuzzy Systems (IFSC)

Auto Tuning PID Controller Using Chaotic PSO

Algorithm for a Boost Converter

Esmaeil Mirzaei Hamed Mojallali

Department of Electrical Engineering Department of Electrical Engineering
University of Guilan University of Guilan
Rasht, Iran Rasht, Iran
E-mail: [email protected] E-mail: [email protected]

Abstract — Improving dynamic response of DC-DC associated with such systems is to obtain the optimal PID
converters which are nonlinear and complicate systems is an controller parameters for a satisfactory control performance,
interesting challenge in power electronics. This paper presents the which is a quite difficult task. Using a mathematical model of
design and implementation of digital PID controller using the the system is the best way for PID parameters determination
chaotic particle swarm optimization (CPSO) for a Boost and parameters can be calculated to get the desired response.
converter and proposes an approach for auto tuning converter Often, a detailed mathematical description of the system is
optimum startup response. When the system starts, processor unavailable. Therefore, experimental tuning of the PID
unit analyses the open loop startup response of converter and parameters has to be performed.
prepares optimum response during some iteration using CPSO
algorithm. Chaotic mapping using logistic map sequences Despite the huge number of proposed approaches for PID
increases its convergence rate and resulting precision. Simulation parameter tuning, most of them (as well as the Ziegler-Nichols
and Implementation results show efficiency of proposed method) occasionally yield poor performance in practice.
approach.
Particle Swarm Optimization (PSO) algorithm is a social
Index Terms - PID Controller; Chaotic Particle Swarm search algorithm which is inspired by collective behaviour of
Optimization; Auto Tuning; Boost Converter groups of birds and fishes by James Kennedy &Russell
Eberhart in 1995 [3]. By using real number coding, only a few
parameters are involved in this algorithm and no gradient
I. INTRODUCTION information is strictly required, which makes the algorithm
DC-DC voltage level converters have been one of the simple and easy to implement. The main disadvantage of the
favorite fields for researchers in recent decades and PSO algorithm is that the outcomes fall into the local minima
considerable progress has been made in improving their too easily.
performance. Due to the vast and growing requirements to Chaos is a kind of characteristic of non-linear systems,
these circuits in the prospect of electronics such as switching which is a bounded unstable dynamic behavior that exhibits
power supply, UPS, DC drivers and DC proportional actuators, sensitive dependence on initial conditions and includes infinite
the necessity of more practical researches and improving their unstable periodic motions. Due to the easy implementation and
performance quality and efficiency is determined [1]. special ability to avoid being trapped in local optima, chaos
Common methods for converters modelling are averaging has been a novel optimization technique and chaos-based
state space method, which gives the approximate LTI model to searching algorithms have aroused intense interests [4].The
the system. These methods are unable to appropriately predict application of chaotic sequences instead of random sequences
converters large signal performance, in the other hand, LTI in PSO is a powerful strategy to diversify the population of
converters achieved from averaging method depend on particles and improve the PSO’s performance in preventing
operating point of the converter. Each change in reference premature convergence to local minima.
input, load parameters, or input voltage DC may change the This paper is based on designing and implementing a PID
system condition and may affects its performance [2]. controller for nonlinear system of boost converter using a
The Proportional-Integral-Derivative (PID) controller can chaotic particle swarm optimization algorithm. Because some
be used to control any measurable variable, as long as this characteristics such as overshoot and settling time cannot be
variable can be affected by manipulating some other process shown explicitly on classical criterion functions such as ISE,
variables. Many control solutions have been used over the ITSE, and ITAE, so an appropriate criterion function is
time, but the PID controller has become the ‘industry standard’ considered, which includes all the current parameters and
due to its simplicity and good performance. Primary problem characteristics in step response of boost converter. Then by
978-1-4799-1228-5/13/$31.00 ©2013 IEEE.
 2013 13th Iranian Conference on Fuzzy Systems (IFSC)

using CPSO algorithm and minimizing fitness function, PID 1

parameters would be identified. ,
The rest of the paper is organized as follows. In section II,
a digital PID controller is introduced. Section III describes a t=nT (3)
simple boost converter. The conventional PSO and CPSO
algorithms implementation is investigated in section IV. Tthe gives
implementation of auto tuning controller and experimental
results are reported in section V. 1

(4)
II. PID CONTROLLER , (5)
The basic form of the PID controller composes from sum
of the multiplication, integration and differentiation of the where n is discrete step at time t, u is the controller output, K
error signal and each operator is multiplied by three control is the proportional gain, K is the integral gain and K is the
parameters. The closed loop system with PID controller is derivative gain, must be optimized.
shown in Fig. 1. PID controller compares the measured
process value y with a reference set point value, y0. The Commonly, there are two approaches to tuning parameters
difference or error, e, is then processed to calculate a new of PID controller. When PID parameters are adjusted in real-
process input, u. This input will try to adjust the measured time when the controller is active, it is called on-line tuning.
process value back to the desired set point. Besides, the parameters are measured then the tuned controller
is performed on a plant, which is called off-line tuning.

III. BOOST CONVERTER

DC to DC converters have different topologies to increase
or decrease DC voltage level including Buck, Boost, Flyback,
etc. Generally, they are classified in two technologies of
transformer based and transformerless. The inductor currents
and capacitor voltages contain dc components, plus switching
ripple at the switching frequency and its harmonics. In most
Fig. 1. Closed loop system with PID controller
well designed converters, the switching ripple is small in
magnitude compared to the dc components. For inductor
PID controllers can be implemented both analogue and currents, a typical value of switching ripple at maximum load
digital version. Today, digital PID controllers are being used in is 10% to 20% of the dc component of current. For an output
many large and small-scale control systems, replacing its capacitor voltage, the switching ripple is typically required to
analogue type. It is now a common practice to implement PID be much less than 1% of the dc output voltage. In both cases,
controllers in its digital version, which means that they operate the ripple magnitude is small compared with the dc
in discrete time domain and deal with analogue signals component, and can be ignored [2].
quantized in a limited number of levels. The trend toward
digital rather than analogue control is mainly due to the Boost converter is a DC-DC transformerless converter that
availability of low-cost digital processors. steps up its input voltage based on the formula given in by [2]:

In this paper we implemented digital PID controller, the (6)

integral and the differential becomes sum and difference,
respectively. The PID controller receives the error as input, Where is the output voltage of boost converter, is
given by the input voltage and D is the duty cycle of incoming signal to
the MOSFET or IGBT gates. The basic diagram of the boost
(1) converter is shown in Figure 2.
This gives u with respect to e t in time domain:

(2)

A discrete PID controller will read the error, calculate and

output the control input at a given time interval, at the sample
period T. The sample time should be less than the shortest time
constant in the system. Approximating the integral and the
derivative terms to get the discrete form using (3)
Fig. 2. Basic diagram of the boost converter
 2013 13tth Iranian Conference on Fuzzy Systems (IFSC)

In practice, the resistance of the induuctor winding is where w is inertia coefficient thatt can be fixed or variable
modeled by a resistor in series with the inducttor too. depending on the need, c1 and c2 2 are learning coefficients,
which are usually given in the range [0, 2]. r1 and r2 are
When the MOSFET is switched ON, the inductor is random numbers, which are usually y considered with uniform
directly connected to the input voltage sourcee. In this case, the
inductor current rises charging it and the innductor is storing distribution in the range [0,1]. is the best solution of ith
energy while the diode is reverse biased discoonnecting the load particle, and is the best solu
ution which is suggested by
(RL) and output capacitor (C) from the source voltage. Within all particles in the population [3].
this interval, the pre-charged capacitor assuress constant voltage onventional PSO algorithm
Implementation steps of the co
across the load terminals [2]. are:
When the MOSFET is switched OFF, thee diode is forward d evaluating it
1. Creating initial population and
biased and both the source and the chargged inductor are 2. Determining the best person nal and the best global
connected to the load. The inductor releases tthe energy stored solutions
in it. This energy is transferred to the loadd in the form of 3. Updating velocity and positionn
voltage that adds to the source voltage. Hennce, the converter 4. Going back to step 2 , if sto opping conditions are not
boosts the input voltage. The boost converterr proposed in this satisfied
paper is a 5 volts input voltage to 12volts at thhe output, and the 5. Ending
switching frequency is 15 KHz. The circuuit parameters of
converter are listed in Table I. To enrich the searching behaavior and to avoid being
trapped into local minima, chaoticc dynamics is incorporated
TABLE I. Circuit Parameters of the Boost C
Converter
into the above PSO. Different typees of equations have been
Parameter Value Units considered in literature for app plications in optimization
Inductance, L 250 µH methods. The logistic equation and other
o equations, such as ten
Capacitance, C 1000 µF
map, Gauss map, Lozi map, siinusoidal iterator, Chua’s
oscillator, Mackey–Glass system, Lorenz
L system, Ikeda map,
Load resistance, RL 25 Ω
and others, have been adopted insstead of random ones and
Inductance resistance, Rl 20 mΩ very interesting results [5-10]. In this
t paper, the well-known
logistic equation [11], which exhibitts the sensitive dependence
on initial conditions, is employed for constructing hybrid PSO.
IV. CPSO ALGORITHM
In PSO algorithm, each proposed solutioon to investigate del of population biology is
The logistic map which is a mod
problem is called a particle, which is as a pooint in the search frequently used with PSO. The logiistic equation is defined as
space. Each particle in population does whhat it thinks it is follows.
correct according to its own action, but for final decision, it μ. 1 , 0 1 (9)
considers the results of its and others' deccisions. Thus, an
informational process occurs among particlees in population, Where µ is the control parameteer, is a variable and n= 0,
which would follow a more favourable final reesult [3]. 1, 2, …
In each iteration, the proposed solutionss change, and the
algorithm goes forward better results. Primarry information of
the population is created randomly. Then, infformation of each
iteration is created by use of previous iteratioon. In addition to
position, each particle has another fundamentaal property called
velocity. The velocity of each particle which is represented by
a vector, determines particle's direction and range of
movement in each iteration. Position of eachh particle in j+1th
iteration is calculated as [3]:
(7)

where is velocity vector of ith particle inn j+1 iteration. In

each iteration, each particle is displaced as muuch as its velocity
vector in the previous iteration. The decisioon made by each
particle for its next iterations influences thee velocity of that
particle. Thus, velocity of each particle in j+11 iteration can be Fig. 3. Bifurcation diagram of Logistic Map
calculated as: Although the above equation is deterministic, it exhibits
chaotic dynamics when µ=4 and 0,0.25,0.5, 0.75,1 .
. 1. c1.
That is, it exhibits the sensitiv ve dependence on initial
2. 2. (8) conditions, which is the basic characcteristic of chaos. A minute
difference in the initial value of the chaotic variable would
 2013 13th Iranian Conference on Fuzzy Systems (IFSC)

result in a considerable difference in its long time behavior. F( , 1 2 (15)

The track of chaotic variable can travel ergodically over the
whole search space. In general, the above chaotic variable has where w1 and w2 are weighting factors of two key terms that
special characters, i.e. ergodicity, pseudo-randomness and are tuned by the start-up response analyse and
irregularity [4]. Overshoot:
The process of the chaotic local search could be defined Settling time: , 2%
through the following equation:
(16)
4 1 , 1, 2, … , (10)
where is the output voltage of the Boost converter and
where is the i th
chaotic variable and k is denotes the setpoint is DC requirement value for is equal to 12volts in
iteration number. this study. The open loop start-up response of Boost converter
with parameters within Table1 has shown in Figure 4.
The procedures of chaotic local search can be illustrated as
follows [4]:
Step1: Setting k=0, and mapping the decision
variables , 1, 2, … , among the intervals
( , , , ) to chaotic variables located in the interval
(0, 1) using the following equation.
,
, 1, 2, … , (11)
, ,

Step2: Determining the chaotic variables for the

next iteration using the logistic equation according to .
Step3: Converting the chaotic variables to decision
variables using the following equation.

, , ,
Fig. 4. Start-up response of the open loop of boost converter
1, 2, … , (12)
Besides, having a good start-up response, a PID controller
Step4: Evaluating the new solution with decision must have a good dynamic response to compensate the
variables . changes of load and in boost converter. Therefore, the
controller is considered in both of start-up and dynamic
Step5: If the new solution is better than the non-chaos response. The result of implementation of PID controller using
search one, or the predefined maximum iteration is reached, CPSO algorithm for a boost converter in MATLAB is shown
output the new solution as the result of the chaos search, in Fig. 5.
otherwise, let k = k + 1 and go back to Step2.
If the search space extends in a wide area, the searching
cannot be completed in the optimal area in a short time. Hence,
in order to obtain high performance, the chaotic search is run
in a small range. This search area is changed using the
following equations [4].

, , , , , (13)

, , , , , (14)

where is the global best position so far and r is the variable

between 0 and 1 value. Decreasing the range of the chaotic
search will increase the searching efficiency.

V. IMPLEMENTATION OF CONTROLLER
Design of PID controller using CPSO algorithm for a boost
converter primarily needs to define a proper cost function, and
then decreasing it. Commonly, a good startup response is Fig. 5. Start-up Response of coost converter using CPSO Algorithm in
characterized by low overshoot and settling time, so for this MATLAB
section cost function is:
 2013 13th Iranian Conference on Fuzzy Systems (IFSC)

As shown in Table II, the average results of objective In each iteration, CPSO generates new solution ( , , )
functions (settling time & overshoot) after the 100 times run and evaluates its cost. If the cost function values is better than
show the efficiency of CPSO compared to conventional PSO previous value, the PID parameters are replaced by new
in trapping to local optima and more speed to reach the values. Therefore, the process continues until the start-up
optimum points. response conditions are satisfied. The experimental results of
proposed circuit implementation are shown in Fig. 7.
TABLE II. Average Results of Objective Functions
Any change in load or input is measured by (1) and is
NUMBER PSO CPSO compensated by tuned PID controller in circuit. For example,
OF when the circuit tuned the PID controller coefficients, load
ITERATIONS Settling time Overshoot Settling time Overshoot transient response for 50% load decrease (from 0.48 to 0.24 A)
(sec) (sec)
(%) (%) and 100% load increase (from 0.24 to 0.48 A) were applied to
it. Fig. 7 shows the performance of close loop circuit for error
5 0.0225 16.3 0.0124 3.6
compensation.
15 0.0125 3.7 0.0106 1.1
Because CPSO algorithm takes a little more time rather
25 0.0106 0.7 0.0098 0.3 than conventional PSO, there is solicitude for selecting
35 0.0096 0.4 0.0095 0.2
appropriate processor, but the ASC period eliminates this
consideration. For this implementation, AD574AJ is used as
45 0.0094 0.3 0.0095 0.2 A/D converter unit which is a complete 12-bit ADC with
Reference and Clock. Also, dsPIC30F4013 and IRFP250
The overall diagram of proposed auto tuning circuit is MOSFET are used as processing unit and switching of Boost
shown in Fig. 6. First, the system starts with initial values and circuit, respectively. Experimental result of Boost circuit
tries to accede to setpoint value in output by increasing duty implementation using CPSO algorithm is shown in Fig. 7.
cycle of PWM signal. Then processing unit analyses output
response and calculates overshoot, settling time and fitness
function values.

Fig. 6. The overall diagram of proposed auto-tuning Circuit

Besides Auto-tuning Control Signal (ACS) is tuned based

on . It means that ACS is a timing signal of input voltage
where the duty cycle of ACS is specified equal to settling time
value in open loop response at least. When the duty cycle of
ASC is equal to 3* , this ensures reaching to steady state
response. Also, it takes time recharge the energy saving
elements in circuit again which is determined by:
4 (17)
This ensures reaching to initial state. Therefore the period
of ACS is specified by:
4 3 (18)
Fig. 7. Experimental results of circuit implementation, startup & dynamic
In this study, the ACS period is equal to 150ms and its duty response. (a) Load decrease: 50%, (b) load increase: 100%.
cycle is equal to 30%.
 2013 13th Iranian Conference on Fuzzy Systems (IFSC)

[11] S. M. R. Rafiei, A. Amirahmadi, G. Griva, “Chaos Rejection and Optimal

Dynamic Response for Boost Converter Using SPEA Multi-Objective
Optimization Approach,” in proc. Annual Conference Industrial
Electronics, 2009, pp. 3315-3322.

Fig. 8. Experimental setup

VI. CONCLUSION
This paper is applied the chaotic particle swarm
optimization for PID controller design of a boost converter. A
multi objective function including dynamic response of
converter has been developed and optimized. An auto tuning
approach based on CPSO algorithm has been presented. First,
open loop response of system is analyzed. Then, optimum
coefficients for PID controller are prepared during some
iterations to have an appropriate startup and dynamic response.

REFRENCES
[1] Robert W. Erickson, Fundamentals of Power Electronics, New York:
Chapman and Hall, 1997.
[2] N. Mohan, T. Undeland, and W. Robbins, Power Electronics: Converters,
Applications and Design, 2nd ed., New York: John Wiley & Sons, 1995.
[3] J. Kennedy, R. Eberhart, “Particle Swarm Optimization,” in Proc. IEEE
International Conference on Neural Networks, vol.4, 1995, pp.1942-1948.
[4] B. Liu, L. Wang, Y.H. Jin, F. Tang, C.X. Huang, “Improved particle
swarm optimization combined with chaos,” Chaos, Solitons and Fractals,
vol. 25, pp. 1261-1271, 2005.
[5] L. D. S. Coelho, H. S. Lopes, “Supply chain optimization using chaotic
differential evolution method,” in Proc. IEEE international conference on
systems, man, and cybernetics, 2006, pp. 3114-3119.
[6] B. Liu, L. Wang, Y.H. Jin, F. Tang, D.X. Huang, “Directing orbits of
chaotic systems by particle swarm optimization,” Chaos, Solitons &
Fractals, vol. 29, no. 2, pp. 454-461, 2006.
[7] B. Liu, L. Wang, Y. H. Jin, F. Tang, D. X. Huang, “Improved particle
swam optimization combined with chaos,” Chaos, Solitons & Fractals ,
vol. 25, no. 5, pp. 1261-1271, 2005.
[8] R. Caponetto, L. Fortuna, S. Fazzino, M. G. Xibilia, “Chaotic sequences
to improve the performance of evolutionary algorithms,” IEEE
Transactions on Evolutionary Computations, vol. 7, no. 3, pp. 289-304,
2003.
[9] R. S. Guerra, H. J. Gon_calves J_unior, W. F. Lages., “A Digital PID
Controller Using RTAI”, 4th RTL Workshop, Linux for devices articles,
1997.
[10] L. Guo, J. Y. Hung, R.M. Nelms, “Comparative evaluation of sliding
mode fuzzy controller and PID controller for a boost converter,” Electric
Power Systems Research, vol. 81, no. 1, pp. 99-106, 2011.