You are on page 1of 6

Evaluation of a Particle Swarm Optimization

controller for DC-DC boost converters


J. B. L. Fermeiro, J.A.N. Pombo, M. R. A. Calado, S.J.P.S. Mariano
IT – Instituto de Telecomunicações
Department of Electromechanical Engineering
University of Beira interior
Covilhã, Portugal
d1652@ubi.pt; d1232@ubi.pt, rc@ubi.pt; sm@ubi.pt

Abstract—Particle swarm optimization has been used as a however the controller need time to train with a training
tuning algorithm for dc-dc converter controllers. In this work function to adjust the population weights, and depending on
it is proposed a novel approach where the PSO is used as the the complexity of the controller it can be time-consuming.
controller itself for a dc-dc boost converter. The PSO Optimal control is another approach [8, 9] that is very robust
controller uses a population of particles to achieve the to changes in the system parameters, however in order to
necessary compensation to stabilize the output at reference achieve the finest response from the controller, certain
voltage. The algorithm uses the gain formula to reposition the controller parameters need to be calculated, thus being time-
particles near the required position, to stabilize the output at consuming. The sliding mode control also requires time-
reference voltage, according to the disturbances in the
consuming tuning to achieve satisfactory results [10-12].
converter. Simulations have been carried on in order to
observe the controller’s response to changes in reference In order to optimize certain controller approaches
voltage and in load variance. researchers can use hybrid techniques and particle swarm
optimization (PSO). This last optimization algorithm is used
Keywords—Dc-Dc boost converter; Pulsewidth modulation to find the optimal parameters for the controllers [13, 14] and
switching control; Particle Swarm Optimization; Space state enhance its response. After some search we found that the
model boost converter. PSO is only used to tune and optimize other controller
I. INTRODUCTION approach and never as the controller itself.
Switching dc-dc converters are power electronic The optimization method PSO is a simple, robust method
converters and therefore extremely important in most with great flexibility and rapid convergence. It allows the
electronic devices, from consumer electronic to renewable control of the response timing through a velocity equation
energy systems [1, 2]. In order to achieve the high efficiency where we can give different weights for the social and
and quality required for the different applications, the control individual parameters. After the reference voltage is
and modeling of the dc-dc converters have suffered achieved from the controller the output doesn’t oscillate from
extensive research and improvement in the last few decades. that value, providing a more pure and clean voltage. With
The most common control method for dc-dc converters is the this in mind a PSO controller is proposed and tested. It was
pulsewidth modulation (PWM). Classical controllers that tested, comparing with a classic PI controller, in a simulation
take advantage of conventional proportional-integral- environment with reference voltage and load variations to
derivative (PID) approach are sensitive to the system observe its response.
parameter. They work best in systems with single input and This paper is organized as follows: in section II is
output, thus to control systems that have multiple presented a brief resume of the optimization algorithm PSO
inputs/outputs several PID controllers are required. This where the most common topologies are addressed; in section
strategy might work however is hard to implement because III is stated a study of the converter and the mathematic
of the correlation between the variables[3]. Also the plant model in space state equations of the boost converter for the
requiring the control might not behave in a linear fashion, two stages of the converter operating in continuous mode; in
meaning that the output for a given input might not exhibit a section IV the proposed approach of the PSO controller for
linear response. The plant behavior changes over time due to dc-dc converter is presented, with its characteristics and
changes on plant loads and normal wear and tear, so the system operation; in section V the results of the simulations
controller needs recalibration maintenance over time. Lastly carried out are reported, with the response of the controller
the tuning of PID controllers might not even achieve optimal PSO, comparing with a classic PI controller, proposed to
system performance [4]. three different situations, considering changes in reference
For all of these reasons researchers are searching for voltage and in load.
optimizations for dc-dc converters. Several approaches have
been published like fuzzy control, optimal control or sliding II. PARTICLE SWARM OPTIMIZATION
mode control. Controllers that use neural networks and fuzzy The particle swarm optimization algorithm is inspired in
are accurate and robust [5-7] to be used in dc-dc converters cooperation and social behavior principles. This algorithm

978-1-4799-6301-0/15/$31.00 ©2015 IEEE 179


has particle populations, where each one represents a (3)
possible solution. .
For each particle there is a velocity related to it, which is
adjusted with an update equation that consider the historic III. STATE SPACE MODEL BOOST CONVERTER
of individual and collective experiences [15]. The Boost Converter is a non-isolated converter to rise
The main idea is to change the position of each particle the voltage, and it is capable of providing output voltage
with certain velocities in order to find the best solution. In higher or equal to the input voltage. In Fig. 2 it is represented
each iteration the particle and the population performances the electric circuit and the control of the converter.
are evaluated with a fitness function and each particle
velocity is incremented or decremented in the direction of
their best solution (Pbest), as well as in the direction of the
best performance of the global best solution (gbest), Fig. 1.
pbest ,i

xik +1

xik
xik -1 gbest

Fig. 1. Graphical representation of the particle evolution.

The most important aspect for the algorithm


performance is its topology, the way that the particles
communicate between each other. There are many
topologies in the literature where the more common are: Fig. 2. Electric circuit and control of the Boost Converter.
• Star topology - where every particle
Looking closely at Fig.2, it can be stated two stages in
communicates with each other. the converter operating, in continuous mode:
• Ring topology – each particle only
communicates with a certain k number of • On the first stage where 0 the
adjacent neighbors. mosfet is in electric conduction state and the
• Cluster topology – certain particles perform the diode is in reverse polarization state, Fig. 3.a.
communication between the clusters and inside • On the second stage where 0 the
each cluster the particles communicate with mosfet is in cut off voltage state and the diode is
each other. in direct polarization state, Fig 3.b.
• Von Neumann topology – the particles are
connected in a way where the particles in one
extremity communicate with particles in the
opposite extremity.

The velocity for each particle can be defined by the


equation (1),

(1)

where

, ~ ,
(2)
Fig. 3. Boost converter electric conduction stages operating in continuous
mode.
The positive constants and are the acceleration and
the inertial weight, they should respect the equation (3) so The differential equations that characterize the first stage
that the particle velocities and the particle positions do not of conduction considering ideal semiconductors (mosfet and
diverge. diode) are:

180
0
0 (4)
(9)

If equations (4) are rewritten in state-space form, The mathematical model of the converter can be obtained
considering the state vector , it is obtained: doing an arithmetic average weighted by the modulation
parameter of the converter from the first and the second
stages, so:

(5) (10)

Resulting in,
Writing the system (5) in the matrix form it is obtained:

(6) … (11)

On the second stage, where , the


differential equations that characterize the conduction stage The component in permanent regime from the previous
considering that the semiconductors are ideal (mosfet and system is given from:
diode) are:
0
(12)
0
From the previous equations we can obtain the equation
0
of the realistic converter gain, equation (13). Considering
and equal to zero, the gain expression is given by equation
(14)

(13)
1
(7)
(14)
0

If this is written in the equation state form where the state The gain variation of the converter as function of the
vector is it is obtained: duty cycle for the system parameters used can be visualized
in Fig. 4. It is non-inverting, i. e, there is not polarity
inversion, the graphic shows a curve with higher slopes for
duty cycle above 0.5 and a more aligned behavior below that
mark.
(8) Another relevant aspect to be accounted for is that the
converter should not be designed with a nominal duty cycle
close to the maximum gain, because if there is the need to
increase the output voltage, by increasing the duty cycle, the
Writing the previous system in the matrix form it is controller will be operating in the descending part of the
obtained: curve, lowering the output voltage instead of increasing it,
and this will lead to a converter failure.

181
Fig. 6. Graphical representation of the particle positioning and
repositioning – A when submitted to a reference voltage variation and B
when submitted to load variation.

After the particles positioning it evaluate the performance


of each particle and determines the best Particle and global
performances. In the third stage it updates the position of
each particle with the equation (1), Fig. 7.
Fig. 4. Gain variation as function of the duty cycle with Rload=1000Ω,
C=200e-6F, RC=0.05 Ω, L=1.6e-3 H, RL=0.5Ω.

IV. PROPOSED APPROACH


The controller proposed provides the output voltage or
feedback voltage through mathematical formulation of the
determined duty cycle for each k moment, for a more
realistic response of the controller, Fig. 5. The algorithm has
3 particles in a star topology, which means that all the
particles communicate with each other. The algorithm in a
first stage positions the particles (converter duty cycle) as
shown in Fig.6.

Fig. 5. The general schematic of the controller system.

In case of the reference voltage change the controller


Fig. 7. Fluxogram of the controller algorithm.
repositions the particles in the following manner, Fig 6. A, if
Vref(k) is different from Vref(k-1), the central particle is
In Fig. 8 we can observe with great detail the particle
positioned according equation (14) and the other two
behavior when the controller is subjected to reference
particles are positioned in a configurable predetermined
voltage variations. In the first iteration the particles are
range, originating a small convergence time and little
positioned as mentioned before, this happens within four
overshoot. In case of the load change the controller
sampling times (with Ts = 1/25kHz), one for each particle
repositions the particles in the following manner, Fig 6. B, if
and the last one for the velocities calculation and new
the difference between Vrefk) and Vfbk(k) is higher than a
position determination, with equation (1).
configurable predetermined value, and the particles
velocities are lower than another configurable predetermined
value, the particles are repositioned. This last is required
because the particles converge to the reference voltage and
their velocities tend to zero, at this point the controller will
not respond to this kind of perturbations without
repositioning the particle positions. In this situation the
central particle is positioned again according to equation (14)
and the other two particles are positioned in a range
determined with the difference between the reference voltage
and the feedback voltage.

Fig. 8. Representation of the algorithm particle positioning.

182
V. TEST RESULTS
The proposed controller was compared to a classic PI
controller. The PI parameters used were Kp= 0.6 and Ki= 2.5.
Comparing both controllers the PSO controller revealed a
reduced convergence time and lower overshoot. However the
biggest advantage of this approach is that after the system
converge it stabilizes at reference voltage with no
oscillations. At first the controller was subjected to respond
to a single reference voltage given of 100 V, Fig. 9 and 10.
In this case the convergence time for both controllers was
very little but the PSO controller revealed lower overshoot
and no oscillation after reaching the reference voltage.

Fig. 11. Response, in output voltage, of the PSO controller and classic PI
controller to two changes in reference voltage.

Fig. 9. Response, in output voltage, of the PSO controller and classic PI


controller to a single reference voltage.

Fig. 12. Response, in duty cycle, of the PSO controller and classic PI
controller to two changes in reference voltage.

In the last trial the controller was submitted to changes in


the reference voltage, starting at 100 V from 0 s to 0.015 s
and finishing with 120 V from 0.015 s to 0.03 s, and a load
increase by a factor of 5 at 0.0075 s and then decreased back
to the initial load at 0.0225 s, Fig. 13 and 14.
To ensure the correct convergence of the particles at this
moment, the parameters used for the range of the particles
positioning might not be ideal and with the improvement of
the algorithm the controller will obtain better results. This
Fig. 10. Response, in duty cycle, of the PSO controller and classic PI influenced the performance of the PSO controller when
controller to a single reference voltage.
comparing with the classic PI controller at the moment when
In a second test the controller was submitted to changes the load is changed, showing a little more overshoot and
in reference voltage starting with 100 V from 0 s to 0.01 s, convergence time. However and as said above this behavior
then lowering to 60 V from 0.01 s to 0.02 s and finishing can be improved with the PSO parameters optimization, also
with 120 V from 0.02 s to 0.03 s, Fig. 11 and 12. Again the it is important to mention that the classic PI controller
PSO controller exhibited a better response than the classic PI responded worst when the reference voltage was changed
controller, with less overshoot, oscillation and settling time. with the new load, with higher convergence time and
increased overshoot for the same PI parameters.

183
this use. In a future work we expect to optimize the PSO
controller and test it in a real circuit to obtain the response
and performance of this controller in a real environment.
REFERENCES
[1] A. I. Bratcu, I. Munteanu, S. Bacha, D. Picault, and B. Raison,
"Cascaded dc–dc converter photovoltaic systems: Power
optimization issues," Industrial Electronics, IEEE Transactions on,
vol. 58, pp. 403-411, 2011.
[2] O. Hegazy, J. Van Mierlo, and P. Lataire, "Control and analysis of
an integrated bidirectional DC/AC and DC/DC converters for plug-
in hybrid electric vehicle applications," Journal of Power
Electronics, vol. 11, pp. 408-417, 2011.
[3] O. Garpinger, T. Hägglund, and K. J. Åström, "Performance and
robustness trade-offs in PID control," Journal of Process Control,
vol. 24, pp. 568-577, 2014.
[4] A. G. Alexandrov and M. V. Palenov, "Adaptive PID Controllers:
State of the Art and Development Prospects," Automation and
Remote Control, vol. 75, pp. 188-199, Feb 2014.
Fig. 13. Response, in output voltage, of the PSO controller and classic PI [5] K. H. Cheng, C. F. Hsu, C. M. Lin, T. T. A. Lee, and C. S. Li,
controller to a changes in reference voltage and in load. "Fuzzy-neural sliding-mode control for DC-DC converters using
asymmetric Gaussian membership functions," Ieee Transactions on
Industrial Electronics, vol. 54, pp. 1528-1536, Jun 2007.
As we can see in Fig. 14 the controllers adjusted the duty [6] R. J. Wai and L. C. Shih, "Adaptive Fuzzy-Neural-Network Design
cycle to compensate the current consumption, increasing it for Voltage Tracking Control of a DC-DC Boost Converter," Ieee
when load increases and lowering when load decreases, in Transactions on Power Electronics, vol. 27, pp. 2104-2115, Apr
order to provide the required output voltage. 2012.
[7] J. Mahdavi, M. R. Nasiri, A. Agah, and A. Emadi, "Application of
neural networks and state-space averaging to DC/DC PWM
converters in sliding-mode operation," Mechatronics, IEEE/ASME
Transactions on, vol. 10, pp. 60-67, 2005.
[8] S. J. P. S. Mariano, J. A. N. Pombo, M. R. A. Calado, and L. A. F.
M. Ferreira, "Optimal Output Control: Load Frequency Control of a
Large Power System," 2009 International Conference on Power
Engineering, Energy and Electrical Drives, pp. 358-363, 2009.
[9] F. H. F. Leung, P. K. S. Tam, and C. K. Li, "An Improved Lqr-
Based Controller for Switching Dc-Dc Converters," Ieee
Transactions on Industrial Electronics, vol. 40, pp. 521-528, Oct
1993.
[10] J. Mahdavi, A. Emadi, and H. Toliyat, "Application of state space
averaging method to sliding mode control of PWM DC/DC
converters," in Industry Applications Conference, 1997. Thirty-
Second IAS Annual Meeting, IAS'97., Conference Record of the 1997
IEEE, 1997, pp. 820-827.
[11] E. M. Navarro-López, D. Cortés, and C. Castro, "Design of practical
sliding-mode controllers with constant switching frequency for
power converters," Electric Power Systems Research, vol. 79, pp.
796-802, 2009.
[12] B. Cardoso, A. Moreira, B. Menezes, and P. Cortizo, "Analysis of
Fig. 14. Response, in duty cycle, of the PSO controller and classic PI switching frequency reduction methods applied to sliding mode
controller to changes in reference voltage and in load. controlled DC-DC converters," in Applied Power Electronics
Conference and Exposition, 1992. APEC'92. Conference
VI. CONCLUSION Proceedings 1992., Seventh Annual, 1992, pp. 403-410.
[13] C. N. Ko and C. J. Wu, "A PSO-tuning method for design of fuzzy
The particle swarm optimization is a great algorithm to PID controllers," Journal of Vibration and Control, vol. 14, pp. 375-
rapidly and accurately find the best parameters for common 395, Mar 2008.
controllers of dc-dc converters, and it is been used only with [14] M. Ranjani and P. Murugesan, "Optimal fuzzy controller parameters
that purpose. However it can be, as shown in this work, used using PSO for speed control of Quasi-Z Source DC/DC converter
itself as a controller for dc-dc boost converter. Simulations fed drive," Applied Soft Computing, vol. 27, pp. 332-356, Feb 2015.
[15] M. Ali and P. Kaelo, "Improved particle swarm algorithms for global
have been carried with component parameters in order to
optimization," Applied Mathematics and Computation, vol. 196, pp.
obtain the most reliable response from the controller. The 578-593, 2008
proposed approach is a PSO controller with a population of .
particles in a star topology, where they communicate with
each other. These are positioned and repositioned when the
controller detects reference voltage and load disturbances.
Overall the results of this approach are very promising, when
compared to a classic PI controller, the controller response is
accurate, with rapid convergence time, little overshoot and
no oscillation after convergence, proving its robustness for

184

Powered by TCPDF (www.tcpdf.org)

You might also like