INTERNATIONAL JOURNAL OF ELECTRICAL AND ELECTRONIC SYSTEMS RESEARCH

Implementation of Particle Swarm Optimization

for tuning of PID controller in Arduino Nano for
Solar MPPT system
Muhammad Iqbal Mohd Zakki, Mohd Najib Mohd Hussain, and Nawawi Seroji

The control plant of the system, G(S), includes PV module,

Abstract—This paper presents the implementation particle DC-DC converter, and resistive load. The control input R(s) of
swarm optimization (PSO) embedded in Arduino Nano for the process loop is the voltage reference Vref fed by MPPT
proportional-integral-derivative controller (PID) tuning as a controller, whereas the feedback signal H(s) is the
validation for real-time hardware application on the photovoltaic
maximum power point tracking (MPPT) system. The Arduino is
instantaneous voltage of PV module VPV. In this control
interfaced to the MATLAB/Simulink via serial communication scheme, the input of the PID controller is the error between
device to provide duplex communication for PID tuning. The Vref, and VPV.
performance of PID tuning using PSO in Arduino was compared There is numerous method of PID tuning. Some of them are
with the performance of MATLAB PID tuner toolbox based on by using the Ziegler-Nichols’ (Z-N) technique, the Cohen-
simulation result via Simulink. The embedded PSO-PID provides Coon technique, and gain-phase margin technique [7].
better performance of 10 times higher compared to MATLAB
PID tuner.
Nevertheless, the PID controller can be also tuned using
computation intelligence (CI) techniques by either self-
Index Terms—PV, MPPT, PSO, PID, Arduino Nano. adaptive fuzzy, or evolutionary computation (EC) [8].

I. INTRODUCTION

P ROPORTIONAL-INTEGRAL-DERIVATIVE (PID) controller are

widely used in various industrial application for the
controls of machines, and instruments. According to [1],
the PID controller was first introduced in 1910. On some
occasion, either the proportional P, proportional-integral PI, or
proportional-derivative PD controllers is used instead of the Fig. 1. General block diagram for the closed-loop PV-MPPT system
PID, which depending on the application, and design. The The PID tuning method proposed in this paper is by using
controller is a mathematical based controller which is widely the particle swarm optimization (PSO) algorithm [9] due to the
used in closed-loop system. They can offer control scheme fact that this method independent from human expertise in
such as reference input tracking, and disturbance rejection in tuning process, while can be easily applied to different PV
the system’s plant. system [10]. Furthermore, PSO algorithm is preferred for PID
Application of PID controller in maximum power point optimization process instead other algorithms in evolutionary
tracking (MPPT) for photovoltaic (PV) system is to control the algorithm (EA) family because the PSO have only three steps
switching devices in the power converter. The absence of in the algorithm [11]. Due to low requirement of memory
controllers such as lag-lead compensator, PID controller, or capacity for data processing, thus, PSO is more viable for the
fuzzy logic controller in PV-MPPT system will causes the implementation on low-cost real-time hardware.
power optimization process for solar generator become slow,
and prone to high steady-state error [2]–[6]. This will reduce II. PHOTOVOLTAIC MPPT SYSTEM
the overall efficiency of PV solar energy harvesting. PV system is an electrical generation system that can be
Shown in Fig. 1 is the block diagram the control system. built from integration of PV modules; that make-up from PV
cells. PV cells are made of semiconductor materials, with
similar working principles to other semiconductor devices
Submitted on 28th of February 2018 and accepted on 28th August 2019. The
author would like to thank Universiti Teknologi Mara Malaysia and Research such as photo-transistor, photo-resistors, photo-detectors, and
Management Institute for the financial support through Fundamental Research others. As a result, PV cells also inherit the non-linear
Grant Scheme (FRGS/1/2016/TK10/UiTM/62/2) for this study. electrical current-to-voltage (I-V) relationship that are rely on
Muhammad Iqbal Mohd Zakki is with Universiti Teknologi Mara Pulau
Pinang, Permatang Pauh, Jalan Permatang Pauh, 13500 Permatang Pauh, the ambient environment such as the intensity of incidence
Pulau Pinang, Malaysia (e-mail: muhammadiqbal.mohdzakki@live.com). light (solar irradiance ψ), and the temperature of the solar cell
Dr. Mohd Najib Mohd Hussain is with Universiti Teknologi Mara Pulau Tcell.
Pinang, Permatang Pauh, Jalan Permatang Pauh, 13500 Permatang Pauh,
Pulau Pinang, Malaysia. These two factors produces different I-V curve and their
 INTERNATIONAL JOURNAL OF ELECTRICAL AND ELECTRONIC SYSTEMS RESEARCH, VOL. 13 DEC 2018

corresponding power-to-voltage (P-V) curve as shown in Fig. for real-time application by comparing the performance of the
2. A PV system that directly connected to the electrical load external hardware-embedded PSO-PID tuning with the
will operate at non-optimum parameters because the operating internal MATLAB PID tuner. Both tuning method are applied
points of PV cells are impedance-dependent. Moreover, the on the PV-MPPT system model in the Simulink. The
optimum operating parameters, or the maximum power point performance of these tuning method are compared in terms of
(MPP) changes dynamically with insolation, and heat acts on accumulative mean error squared (MSE).
the solar cell. Hence, the PV system requires devices to
optimize the operating parameters to the MPP by employs III. PID TUNING METHODS
MPPT system. It is worth to mention that Simulink is used as an intrinsic
control plant for performance analysis of the PID tuning as a
method to eliminate influence of external disturbances as
founded in the real hardware environment. This evaluation
approach is carried out so that the C++ coding of embedded
PSO-PID algorithm can be tested, debugged, and verified. On
the other hand, a fair comparison of performance with the
MATLAB PID tuner can be made via Simulink environment.
Shown in Fig. 4 is comprehensive Simulink blocks of PV
system comprise of PV module block model, MPPT, MPPT
control system, voltage and current sensors, and the buck-
boost converter. The system plant consist of PV panel, MPPT
controlled DC-DC converter, and load.

Fig. 2. Operating environment effect on PV cell

Generally, MPPT system consist of intermediary power

converter, MPPT controller, and control loop as exemplified
Fig. 4. Simulink PV-MPPT system
in Fig. 3. The DC-DC chopper (buck-boost) is controlled by
PID controller, based on the command from MPPT. In this Fig. 5 shows the internal components of MPPT subsystem.
case, the power converter is controlled by PID controller, According to Fig. 5, the MPPT controller receives input
which receives inputs from MPPT reference voltage, Vref, and measurements of VPV, and IPV, while process the tracking
instantaneous PV voltage, VPV. reference Vref as command input to the control loop. PID
controller receives error signal from the comparison of Vref and
VPV on the summing junction. The MPPT also controls the
input and output of PID controller via ‘Enable’ port.

Fig. 5. MPPT controller with PID closed-loop

Fig. 6 displays the MPPT inverted buck-boost DC-DC

chopper driven by the control loop as presented in Fig. 5.
Fig. 3. Detailed closed loop control of PV-MPPT system

The prime focus of this paper is to verify the feasibility of

PSO implementation into low-cost microcontroller hardware
Zakki et. al.: Implementation of Particle Swarm Optimization for tuning of PID controller in Arduino Nano for Solar MPPT system

Fundamentally, PSO algorithm searches the optimum

parameters by iteratively updates the particles’ velocity and
position until the fitness, and objective of the optimization are
met.
According to Fig. 8, after the initialization, the algorithm
waits for iterative fitness computation from individual particle
Fig. 6. DC-DC buck-boost converter
position. Subsequently, the current best particle position pbest
is defined by determine the lowest MSE in the swarm fitness.
A. Arduino PSO-PID Tuning
PSO-PID tuning via Arduino Nano hardware is carried out
by interconnecting the MATLAB/Simulink with Arduino
board by Serial-to USB communication as illustrated in Fig. 7.
The PSO-PID tuning algorithm was embedded into the
microcontroller to provide PID gain parameter, while the
MATLAB/Simulink accomplishes the simulation process to
obtain the fitness for each particles position.

Fig. 7. Block diagram illustration for Arduino PSO-PID tuning

Fig. 8. PSO-PID algorithm flowchart
1) The Particle Swarm Optimization (PSO)
The particle swarm optimization algorithm was introduced After the fitness of all particle in the population have been
in 1995 by Kennedy, and Eberhart [9]. The study presents a testified individually, the current best fitness f(pbest) of the
simple algorithm derived from the social interaction concept swarm is compared to the global best fitness f(gbest). The
of birds’ flock that can be used to find solutions for wide fitness functions for each particle were evaluated as mean
range of non-linear problems by using metaheuristics search error squared, MSE of input voltage reference, Vref and the
technique. Conceptually, it has the similarities with genetic actual instantaneous voltage, VPV. The fitness function for PSO
algorithm (GA) since it uses the notation of “fitness” like is shown in (1).
other family members of evolutionary algorithm (EA) in
computation intelligence (CI) to represent the performance of 1 T
(1)
a solution from the population.
f ( s ik ) =
T
∑ (V
i =1
ref (t ) − V PV (t )) 2

According to [11], PSO uses less effort in computation to where f ( s ik ) is the fitness of the current particle, and T is the
achieve the same high quality solutions compared to GA. The total simulation data point.
study present that PSO is proven its advantage with 99% The comparison to determine the pbest, and gbest depends on
statistical confidence. Therefore, the PSO was likely more the “objective function” of the optimization process. In the
efficient to be employed into low-cost hardware systems for case of PID tuning, it is necessary for the control system to
online system application such as PV-MPPT system as this have the lowest error as possible, thus, the resultant “objective
study. For that reason, PSO was implemented as PID tuner in function” is to determine pbest, and gbest of the lowest MSE as
this study. shown in Fig. 8.
 INTERNATIONAL JOURNAL OF ELECTRICAL AND ELECTRONIC SYSTEMS RESEARCH, VOL. 13 DEC 2018

Essentially, the PSO uses equation (2) and (3) as the

respective velocity and position update function for its particle
movement. The influence of the swarm and the global optima
position on the particle’s position is illustrated as in Fig. 9.
Fig. 11. USB-Serial FTDI FT232RL communication chip breakout board.
vik +1 = wvik + c1r1 ( pbest − sik ) + c2 r2 ( gbest − sik ) (2)
sik +1 = sik + vik +1 (3) Fig. 12 shows the program flowchart for PSO-PID tuning
where, via hardware-to-MATLAB interface. In the beginning, the
PSO parameters initialize in the hardware memory, and wait
k is the kth iteration;
for MATLAB slave acknowledgement before begin
i is the particle number of a given swarm; transmitting the PID parameter. The PID gains from the
r1 and r2 is random number as learning factor for individual Arduino will be parsed to Simulink. Subsequently, the MSE
particle to prevent entrapment of particles in local optima fitness data from the simulation run will be sent back to the
master hardware. The optimization process carried out by the
vik is the current particle velocity;
master device iteratively until the fitness of the population
sik is the current particle position; converges.
w is the swarm weight, or inertia factor;
c1, and c2 are the particles, and swarm confidence factor;
pbest is particle position with the best fitness in the swarm, or
iteration;
gbest is the global best particle position.

Fig. 9. Movement of particle under swarm influence

2) Hardware Implementations
The PSO-PID tuning was carried out using hardware as
shown in Fig. 10. This tuning procedure employs master/slave
programming; hardware embedded program (master), and
MATLAB m-file (slave). The PSO-PID tuning program is
embedded in microcontroller, while the m-file program in
MATLAB is used to parse parameters between hardware, and
Simulink, back and forth. The device for hardware-to-software
interfacing is shown in Fig. 11.

Fig. 12. Flowchart of embedded PSO-PID program (master)

Shown in Fig. 13 is the output verbose of the code

Fig. 10. Hardware for PSO-PID tuning on Simulink model
compiling for PSO-PID MATLAB tuner embedded on
Arduino Nano. The memory consumption for the program and
variables are 47% of 30720 byte, and 55% of 2048 byte,
respectively.
Zakki et. al.: Implementation of Particle Swarm Optimization for tuning of PID controller in Arduino Nano for Solar MPPT system

Fig. 13. PSO-PID MATLAB tuner code compilation output verbose

Fig. 14 and Fig. 15 shows snippet of PSO-PID coding for

velocity, and position update function.

Fig. 14. PSO velocity update function

Fig. 17. Simulation runs for PSO-PID tuning

Fig. 15. PSO position update function The iteration of the PSO stopped after the fitness between
particles in the swarm converges with the standard deviation
The Fig. 16 shows the running process of PSO-PID tuning lower than 0.1. The optimized PID parameters were returned
in the MATLAB command window, while Fig. 17 shows the by the gbest position as follows;
simulation run to find the particle fitness. The fitness of the • Proportional gain, Kp = 6.4391
particle is determined by the mean sum of squared error • Integral gain, Ki = 9.9361
between the reference signal Vref and the instantaneous PV • Derivative gain, Kd = 8.6265
voltage, VPV. • Derivative filter, N = 100

B. MATLAB PID Tuner Toolbox

Since the PV system consist of non-linear system response
due to the PV module, and non-linear switching response of
the power converters, the plant of the system must be
linearized by simplify the plant with mathematical modelling
via system identification process.
As shown in Fig. 18, the process initiates from input-output
(I/O) response data sampling. Subsequently, the plant is
estimated and the fitness of the model output is compared with
the sampled data. If the fitness of the model is less than 80%
of the I/O sample, the plant will be re-modelled with new
structure. Otherwise, the linearized model will be applied into
PID tuner toolbox for tuning. Nonetheless, this linearized
model only used in the PID tuner toolbox to simplify the non-
linearity of the system plant.
The performance of the newly acquired PID parameters will
be tested with the simulation blocks as demonstrated in Fig. 4,
by replacing the MPPT with a step input, purposely to
compare with the performance PSO tuned PID controller.
Fig. 16. PSO-PID tuning process in MATLAB command window
 INTERNATIONAL JOURNAL OF ELECTRICAL AND ELECTRONIC SYSTEMS RESEARCH, VOL. 13 DEC 2018

2) Plant Identification
After the data sample is obtained, the model is then
estimated via system identification approach by using
structures by either of one-pole, two real-poles, underdamped
pair, and underdamped with real-pole transfer function, and
state-space. All of the mentioned structures are tested while
their fitness was compared to each other. In conclusion,
underdamped pair have the highest fitness value besides
having the simpler form of equation compared to state-space.
Exhibited in Fig. 20 is the linear plant modelling using
underdamped pair transfer function. The angular frequency,
Tω, and damping ratio, ζ are manually adjusted so that the
identified data intersect as close as possible on the
identification data obtained from the I/O sampling.

Fig. 18. Process flows for PID tuning using MATLAB PID tuner toolbox

1) Input-Output Sampling
To begin with, the I/O data of the system was sampled by
repeatedly inject the step input directly into the input of the
PID controller, while bypassing the input from MPPT and the
feedback loop. Thus, the control system operates as open-loop
control. The PID controller is set to P-only mode with
proportional gain, Kp = 1 for linear modelling purpose.
Fig. 19 shows present I/O sampling result using step input
with rise time at 0.0005s, until 0.01s. The sample time Ts for
I/O data sample is 1x10-6s per iteration.

Fig. 20. Linear model plant estimation

The identified model is further fined-tuned by using the

auto estimate function of the toolbox in order to achieved
fitness of 87.37% as exemplified in Fig. 21.

Fig. 19. I/O response data sampling for plant identification

Zakki et. al.: Implementation of Particle Swarm Optimization for tuning of PID controller in Arduino Nano for Solar MPPT system

Fig. 23. MATLAB PID parameter and performance

Fig. 21. Plant identification for linear model
It is worth noting that, the PID tuning by the toolbox uses
The yield of the modelling process for the underdamped the linearized plant model, therefore, the performance might
pair transfer function (4) as below; be different when the tuned PID parameter applied into the
non-linear PV system.
K (4)
G ( s) = 2
Tω s 2 + 2ζTω s + 1
IV. PERFORMANCE COMPARISON
where the value of gain, K, angular frequency, Tω, and
damping ratio, ζ are 21.551, 0.00021141, and 0.255 By utilizing both methods to obtain PID gain parameters,
respectively. Thus, the transfer function for linear model is the performance of the controller is evaluated in simulation
simplified as in (5). stage from Simulink block as shown in the Fig. 4. The system
block is injected with unit step input with amplitude of 15V
21.551 reference voltage, Vref. The parameters for both tuning method
G(s) = (5) are assessed in Table 1. Fig. 24 and Fig. 25 show the step
0.000000044694s 2 + 0.00010782s + 1
response of closed loop PID control response using MATLAB
PID tuner toolbox, and Arduino PSO-PID tuning, respectively.
3) PID Tuning
Fig. 22 and Fig. 23 are the PID tuner window, and PID TABLE I
controller specifications, correspondingly. The tuner can SIMULATION PARAMETERS OF PID TUNING METHOD
simply be used by adjust the control slider for the response MATLAB PID Tuner Arduino PSO-PID
time, and the transient behaviours. It is desirable that the PID tuning method
Toolbox Tuner
controller to be able to provide the response time, oscillations, Proportional gain, Kp 9.5594 6.4391
and errors as minimal as possible. From the tuned parameters, Integral gain, Ki 6039.1658 9.9361
the achieved rise time is 0.00002s, with 0% overshoot. Derivative gain, Kd 0.003621 8.6265
Derivative filter, N 12375568.5142 100
Settling time (seconds) 0.003501 0.000203
Mean squared error,
1.009891 0.1025
MSE

Fig. 24. Step response of closed loop PID control by MATLAB PID Tuner
Fig. 22. MATLAB PID tuner tool
 INTERNATIONAL JOURNAL OF ELECTRICAL AND ELECTRONIC SYSTEMS RESEARCH, VOL. 13 DEC 2018

[6] R. Pradhan and B. Subudhi, “Design and real-time implementation

of a new auto-tuned adaptive MPPT control for a photovoltaic
system,” Int. J. Electr. Power Energy Syst., vol. 64, pp. 792–803,
2015.
[7] A. Bagis, “Determination of the PID controller parameters by
modified genetic algorithm for improved performance,” J. Inf. Sci.
Eng., vol. 23, no. 5, pp. 1469–1480, 2007.
[8] A. A. R. Coelho, “PID techniques in intelligent and adaptive
algorithms,” in 38th Midwest Symposium on Circuits and Systems.
Proceedings, 1996, vol. 1, pp. 409–412.
[9] J. Kennedy and R. Eberhart, “Particle Swarm Optimization,” pp.
1942–1948, 1995.
[10] G. Reynoso-Meza, J. Sanchis, J. M. Herrero, and C. Ramos,
Fig. 25. Step response of closed loop PID control by PSO-PID tuning
“Evolutionary auto-tuning algorithm for PID controllers,” IFAC
Proc. Vol., vol. 2, no. PART 1, pp. 631–636, 2012.
From the step response test, it is observed that PSO-PID [11] R. Hassan and B. Cohanim, “A comparison of particle swarm
method has the fastest response time compared to the optimization and the genetic algorithm,” 1st AIAA Multidiscip. Des.
Optim. Spec. Conf., pp. 1–13, 2005.
MATLAB-tuned controller. The PSO-PID method achieves
steady-state at t = 0.703ms. Thus, the time taken for the plant
output to settle to step input tss is 0.203ms (note that unit step
rise time tr = 0.5ms). In contrast, MATLAB-tuned PID takes Muhammad Iqbal, Mohd Zakki received Diploma
time to settle for 3.501ms. The recorded MSE for MATLAB- in Electrical Engineering (Power) from Universiti
Teknologi Mara Pulau Pinang, in 2012, and B.Sc in
tuned PID is 1.009891. While for PSO-PID method, MSE Engineering (Electrical and Electronics) from the
successively reduced to 0.1025. same university, in 2017. He is currently pursuing
Since the embedded PSO-PID tuning only takes about one- the M.Sc in Electrical Engineering by research in the
field of renewable energy, specifically in
tenth of the response time of MATLAB-tuner method, optimization of solar photovoltaics system. His
therefore, the PSO-PID shows its advantage over the research interest include development of PV-MPPT
counterpart. system, embedded electronics, power electronics and energy conversion,
artificial intelligence, and system modelling.
Currently, he works as research assistant in the university. In 2015, he used to
V. CONCLUSION work as an intern at Sungai Petani Town Municipal Council (Majlis
Perbandaran Sungai Petani Kedah, MPSPK) in Engineering Department.
From the following results, the PSO-PID algorithm Mr. Muhammad Iqbal is a graduate member of Board of Engineers Malaysia
embedded in the 8–bit microcontroller proves its feasibility, (BEM), and the Institutions of Engineers Malaysia (IEM).
and effectiveness by successfully optimize the PID controller
by provide about 10 times improvement compared to
MATLAB PID tuner in terms of MSE, and response time.
Moreover, PSO-PID requires less step in tuning procedure
since it does not require collection of data for off-line system
identification. Besides that, the tuning process is also reliable
and independent from human supervision. Therefore, the PSO-
PID can be identified as an adaptive tuning method that can
directly embedded to the real PV-MPPT hardware of any
setup. Furthermore, the proposed PSO-PID tuning method
consumes low capacity of flash memory, and processing
power. Hence, PSO is viable for the implementation in low-
cost microcontroller hardware.

VI. REFERENCES
[1] M. Amirinejad, M. Eslami, and A. Noori, “Automatic PID
Controller Parameter Tuning Using Bees Algorithm,” vol. 5, no. 8,
pp. 24–28, 2014.
[2] N. Femia, G. Petrone, G. Spagnuolo, and M. Vitelli, Power
electronics and control techniques for maximum energy harvesting
in photovoltaic systems. 2013.
[3] M. H. (Muhammad H. Rashid, Power electronics handbook :
devices, circuits, and applications. Butterworth-Heinemann, 2011.
[4] Y. Wang, L. Ding, and N. Li, “The application of fuzzy parameters
self-tuning PID controller in MPPT of photovoltaic power system,”
Proc. 2011 Int. Conf. Transp. Mech. Electr. Eng. TMEE 2011, no.
Figure 2, pp. 1129–1132, 2011.
[5] A. El Khateb, N. A. Rahim, and J. Selvaraj, “Optimized PID
controller for both single phase inverter and MPPT SEPIC DC/DC
converter of PV module,” 2011 IEEE Int. Electr. Mach. Drives
Conf. IEMDC 2011, pp. 1036–1041, 2011.