Chinese Journal of Electrical Engineering, Vol.7, No.

4, December 2021

Single-phase Grid-connected PV System with

Golden Section Search-based MPPT Algorithm*
Shuang Xu1*, Riming Shao2, Bo Cao3* and Liuchen Chang3
(1. School of Electrical and Control Engineering, North China University of Technology, Beijing 100144, China;
2. R&D Department, Alpha Technologies Inc., Bellingham 98226, USA;
3. Emera and NB Power Research Centre for Smart Grid Technologies, University of New Brunswick,
Fredericton E3B5A3, Canada)

Abstract: Maximum power point tracking (MPPT) is a technique employed for with variable-power sources, such as solar, wind,
and ocean, to maximize energy extraction under all conditions. The commonly used perturb and observe (P&O) and incremental
conductance (INC) methods have advantages such as ease of implementation, but they also have the challenge of selecting the most
optimized perturbation step or increment size while considering the trade-off between convergence time and oscillation. To address
these issues, an MPPT solution for grid-connected photovoltaic (PV) systems is proposed that combines the golden section search
(GSS), P&O, and INC methods to simultaneously achieve faster convergence and smaller oscillation, converging to the MPP by
repeatedly narrowing the width of the interval at the rate of the golden ratio. The proposed MPPT technique was applied to a PV
system consisting of a PV array, boost chopper, and inverter. Simulation and experimental results verify the feasibility and
effectiveness of the proposed MPPT technique, by which the system is able to locate the MPP in 36 ms and regain a drifting MPP in
approximately 30 ms under transient performance. The overall MPPT efficiency is 98.99%.
Keywords: Grid-connected system, maximum power point tracking (MPPT), photovoltaic (PV) system, single-phase inverter

Due to the nonlinear current-voltage (I-V)

1 Introduction1
characteristics, there is a unique maximum power
The U.S. Energy Information Administration point (MPP) at a given irradiance and temperature in a
[5]
(EIA) estimates that photovoltaic (PV) generation PV system . Significant work has been conducted on
increased from approximately 5 million kW·h in 1984 MPPT with grid-connected and standalone PV systems
[1] [6-7]
to approximately 107 billion kW·h in 2019 . In a to maximize power extraction under all conditions ,
grid-connected PV system, a dc-ac inverter is among which the perturb and observe (P&O), or
[8-9]
modulated by pulse-width modulation (PWM) to hill-climbing method, is most widely used . In the
convert the PV-generated dc power into grid-side ac P&O method, the controller periodically adjusts the
power. The efficiency of power generation of a PV voltage or duty cycle by a small amount from the PV
array is optimized by maximum power point tracking system and measures the power. If the power increases,
[2]
(MPPT) technology . In a single-phase inverter further adjustments are made in that direction until the
system, the grid-side power pulsates at twice the grid power no longer increases, and vice versa. This
frequency due to the multiplication of grid voltage and method is easy to implement, but at steady state, the
current, which causes suboptimal power generation operating point still oscillates around the MPP, which
[3]
from the PV array . Therefore, an additional dc-dc causes energy wastage. Moreover, it is difficult to
converter is usually added to regulate the voltage and select a proper perturbation step considering the
[4] [8]
control the power drawn from the PV array . trade-off between convergence time and oscillation .
Under some rapidly changing atmospheric conditions,
it is well known that the P&O algorithm can face
Manuscript received July 16, 2021; revised October 15, 2021; accepted
[10]
December 5, 2021. Date of publication December 31, 2021; date of current difficulty during such time intervals .
version December 16, 2021.
* Corresponding Author, E-mail: [email protected] and E-mail: [email protected] Regarding these problems with P&O, the
* Supported in part by the Natural Sciences and Engineering Research Council
of Canada, and in part by the Atlantic Innovation Fund. incremental conductance (INC) algorithm was
Digital Object Identifier: 10.23919/CJEE.2021.000035
 26 Chinese Journal of Electrical Engineering, Vol.7, No.4, December 2021

introduced in Ref. [11]. In the INC method, the First, micro-inverters have been developed to be
controller tracks the MPP by comparing the derivative installed in each PV module, giving each PV module
of the array power with respect to the array voltage. If its own MPPT that does not affect the global MPPT of
the derivative is equal to zero, the power point is at the the entire PV array. Second, a bypass diode is
MPP; otherwise, it is at the left side of the MPP when connected in parallel with each PV module to provide
[24]
the derivative is positive and at the right side when the an alternative path during partial shading , and in
derivative is negative. Hence, the INC algorithm is this manner, MPPT techniques considering the partial
superior to P&O when determining the best direction shading condition have been proposed, including the
of perturbation and identifying whether the MPP has memetic salp swarm algorithm (MSSA) and dynamic
[10-12]
been reached . However, similar to P&O, there is leader collective intelligence (DLCI) method. Under
a tradeoff between convergence time and oscillation high penetration of distributed energy resources
when choosing the increment (step) size. As such, (DERs), the risk of grid instability increases due to the
P&O and INC strategies with variable step size were intermittent and stochastic nature of renewable energy,
proposed in Refs. [9, 13-14]. Regarding tracking and modern grid codes have been updated to include
accuracy and rate, MPPT algorithms with variable step more renewable-related rules for DER inverters to
size show better performance than those with fixed step provide power system support functions, such as
size. However, they require deliberate tuning for voltage and frequency regulations. To include
parameter optimization because the increment or step is frequency regulation capability, DERs usually have
not a fixed value. These parameters are dependent on pre-curtailment for active power reserve or headroom
system characteristics. Instructions and principles for in the case of frequency contingency. Various
tuning the optimal values are provided in Refs. [8, 15]. power curtailment methods have been discussed in
Other MPPT techniques exist, such as the ripple Refs. [25-26], shifting the operating point to the left or
correlation control method, which observes the phase right side of the MPPT curves so that the DERs are
commutation between the current and PV power to operated at a lower power instead of the maximum
track the MPP quickly; however, it is difficult to power, but such methods are outside the scope of this
measure the incremental PV power and phase study.
[7]
commutation . Some modern control strategies, such Regarding the drawbacks of the P&O and INC
as fuzzy logic control and neural networks, have also algorithms and the complexity of existing intelligent
been widely adopted as alternatives to P&O and INC control algorithms, this paper proposes a multi-stage
due to their inherent nature of dealing with non-linear MPPT algorithm that includes a golden section search
[16-17]
control objectives . However, the effectiveness of (GSS) method as a straightforward optimization
fuzzy control strategies greatly depends on expertise in approach targeting small-scale PV systems. In this
choosing the correct error computation. Similarly, method, an initial interval is first selected in the PV
neural network-based MPPT performance greatly voltage range, which brackets the MPP. Then, the GSS
depends on the design of the neuron layers and training method begins narrowing the width of the interval at
[27-28]
weights, which limits its versality and flexibility. There the rate of the golden section number (φ=0.618) .
are also some wind turbine and thermophotovoltaic Therefore, the method will converge to the MPP with a
[18-19]
system-specific MPPT algorithms . Some shrinking rate of 0.618 for each round. As for locating
comparisons have also been made among different the initial interval where the MPP search starts, the
[20-23]
MPPT techniques , but there is no de facto P&O method is applied to locate the vicinity of the
superior MPPT approach. MPP as the initial interval for GSS, and the
Nowadays, the partial shading condition is a perturbation used in P&O can be set as relatively large
critical issue in PV systems, especially in large-scale to speed up the search. As a result, the proposed
centralized PV systems, where the shaded PV modules multi-stage MPPT involves P&O in the first stage and
must absorb power from non-shaded PV modules. GSS in the second stage, and the INC technique is
There are multiple ways to deal with this problem. used to verify the located MPP in the third stage, as
 Shuang Xu et al.: Single-phase Grid-connected PV System with Golden Section Search-based MPPT Algorithm 27

the system must know if the MPP has drifted when the uses only the dc-link voltage as a control variable and
conditions change. Therefore, the proposed multi-stage requires nothing from the boost chopper for its
MPPT technique does necessitate a compromise operation. The boost chopper and inverter are thus
between the convergence time and oscillation (as P&O mutually independent, and the corresponding control
and INC do) while maintaining the advantages of P&O is greatly simplified.
and INC, such as ease of implementation and little
2.1 GSS method
search information requirement compared with other
complicated algorithms. The GSS method is used for locating the
The operating principle of the proposed maximum or minimum of a strictly unimodal function
GSS-based MPPT algorithm is described in Section 2. inside a range by repeatedly narrowing the width of
The system configuration and modeling of the the range. For a unimodal function, as shown in Fig. 2,
single-phase grid-connected PV system with MPPT it is recognized that f(x3)>f(x4) once the maximum
are described in Section 3. Simulation and value of the unimodal function lies on the left-hand
experimental results that validate the GSS-based side of x3, as indicated by the dashed line, and
MPPT algorithm are presented and discussed in f(x3)<f(x4) once the maximum value lies on the
Section 4. Finally, the contributions of this study are right-hand side of x4, as indicated by the straight line,
summarized in Section 5. and vice versa. As such, the maximum point is located
within one of the two reduced intervals (x1, x4) and (x3,
2 GSS-based MPPT algorithm
x2) after comparison between f(x4) and f(x3). Then, a
The PV system topology, which consists of a PV new point subdivides the interval into two parts, and
array, boost chopper, dc-link capacitor, and the wider range is used as the new search interval for
single-phase grid-connected inverter, is illustrated in the next iteration. When the number of iterations
Fig. 1. The boost chopper and inverter are connected reaches the limit or the range shrinks to a certain value,
in series through the dc-link capacitor. the point inserted most recently can be considered as
[28]
the maximum point .

Fig. 1 Circuit of single-phase grid-connected PV system

The MPPT mechanism is integrated with the

boost chopper. By decoupling the PV array from the
inverter, it can extract maximum power from the PV
Fig. 2 GSS scheme for locating the MPP
array by regulating the proper PV output. Because the
power injected into the grid by the single-phase As shown in Fig. 2, the main objective is to locate
inverter pulsates at twice the grid frequency, while a the maximum functional value within the range of (x1,
constant dc power is desirable at the PV output, a large x2). In each iteration, two points are selected such that
electrolytic capacitor is paralleled at the dc-link to each point subdivides the range into two parts that
absorb the power imbalance. The instantaneous dc-link meet the golden section, which means the potential
voltage also pulsates at twice the grid frequency search section (x3, x2) or (x1, x4) in the next iteration
because of the second-order power imbalance, but the has a width of 0.618× that in the previous iteration.
average value of the dc-link voltage stays constant. The widths of the potential sections must be equal
In this case, the balanced power flow control because the search speed will rapidly decrease if wider
strategy for the inverter can be as simple as using an sections are chosen more frequently in some cases.
averaged dc-link voltage control. Then, the inverter Therefore, the inserted point x4 must satisfy the
 28 Chinese Journal of Electrical Engineering, Vol.7, No.4, December 2021

following equation This is because the GSS does not require any
a=c (1) derivative computation that is easily disturbed by
In this manner, the sections have the same proportion fluctuating signals and noises, and it can automatically
of spacing between the three points such that the change the search direction as the interval shrinks.
algorithm converges at a constant and optimal speed. Fig. 3 illustrates the principle of the GSS-based
As shown in Fig. 2, section (x1, x2) and the potential MPPT algorithm in single-phase grid-connection PV
section (x1, x4) or (x3, x2) have a relationship between applications, where V, I, and P represent the PV
larger fraction and smaller fraction as voltage, current, and power, respectively, and n is the
a = c = b+c =ϕ number of iterations. The control objective of the GSS
(2)
b b a algorithm is the PV voltage V, so it is regulated
If we combine Eqs. (1) and (2), φ can be calculated as towards the MPPT voltage (VMPPT) under each
iteration. When the width of the interval shrinks to a
ϕ = 1 + 5 = 1.618 033 98 (3)
2 certain value, called the tolerance ε, it is assumed that
which is the golden ratio, as indicated by the name VMPPT has been found, and the GSS process ends.
GSS. After one iteration, the section (x1, x2) shrinks
from (a+b+c) to either (a+b) or (b+c), and

a + b = b + c = 0.618 ( a + b + c ) (4)

Thus, the convergence rate R of the GSS is

R = ϕn (5)

where n is the number of iterations.

Based on Eq. (5), only 15 iterations are required
for the GSS to shrink the search interval to less than
0.1% of its original length. This means that for a
digital control system with 0.1% resolution
corresponding to a 10-bit analog-to-digital converter
(ADC), no more than 15 iterations are required to
obtain the best applicable solution, indicating that the
GSS method converges quickly with sufficient
accuracy. For example, if there are 1 000 points within
an interval, the GSS will find the MPP in no more than
15 iterations, while P&O will require an average of
250 to achieve the same. ε is very similar to the step
size in the P&O method. In the voltage range of 0-180
V, the step size can be set as 0.1% of the whole range,
i.e., 0.18 V. P&O will require 250 iterations on
average to find the MPP, and it will continue
oscillating around the MPP with 0.18 V step size. With
GSS, it will take no more than 15 iterations to find the
MPP within 0.18 V range.
In addition to its fast search speed, the GSS has
strengths such as noise-resistive capability and
tolerance for measurement errors due to the inherent
noises and power ripples in switching-mode converters. Fig. 3 Principle of GSS-based MPPT algorithm
 Shuang Xu et al.: Single-phase Grid-connected PV System with Golden Section Search-based MPPT Algorithm 29

Vn−2) into two parts that meet the golden section ratio,
2.2 GSS-based MPPT with P&O and INC
and (Vn−3, Vn−1) has a width that is 0.618× that of (Vn−1,
Besides the GSS process, P&O and INC are Vn−2).
integrated into the multi-stage MPPT algorithm for ② Insert another point Vn that divides section
single-phase grid-connected PV systems. The MPPT (Vn−3, Vn−2) into two parts that meet the golden section
strategy consists of three stages, each regulated by one ratio, and (Vn−3, Vn) has a width that is 1.618× that of
of the methods. (Vn, Vn−2).
In the first stage, P&O with relatively large ③ Compare the power values at Vn−1 and Vn. If
perturbation steps is used to achieve a fast search Pn>Pn−1, the search interval will be shrunk towards the
speed with the compromise of accuracy. Once VMPPT is right-hand side, where the point Vn−2 remains the same,
located in a narrower range with the relatively simple the initial point of Vn−3 will be updated by the value of
P&O method, the MPPT enters the second stage, Vn−1, and the initial point of Vn−1 will be updated by the
which uses GSS, where the width of the range will be value of Vn. If Pn<Pn−1, the search interval will be
further narrowed at the rate of the golden section ratio. shrunk towards the left-hand side, where the point Vn−1
Finally, when the interval width reaches the tolerance remains the same, the initial point of Vn−3 will be
value, INC is used to verify the MPP. If the obtained updated by the value of Vn, and the initial point of Vn−2
MPP does not drift, the system will operate at the MPP will be updated by the value of Vn−3 to ensure that
steadily; otherwise, the system must revert back to the (Vn−3, Vn−1) has a width that is 0.618× that of (Vn−1, Vn−2).
first stage. ④ Insert a new point Vn and repeat Steps ②
The detailed procedure is as follows. and ③.
(1) P&O: The perturbation steps are counted by ⑤ Once the difference between Vn and Vn−1
number n = 0, 1, 2, …. Starting with n = 0, the initial reaches the tolerance value, the MPP is assumed to be
PV power P0, and the initial PV voltage V0, the P&O found.
search steps are as follows. (3) INC: An MPP has the mathematical
① Perturb the system with an increment ΔV to characteristic that dP/dV=0. In Stage (3), INC aims to
V0, i.e., V1=V0+ΔV. ΔV can be positive or negative. validate the MPP obtained in Stage (2) and detect the
② Observe the output power P1 under steady drift of the MPP under varying weather conditions.
state at the new operating point after a time step. The INC procedure is as follows.
③ Determine whether P1>P0. If so, maintain the ① Calculate the slope at the MPP by applying a
sign of ΔV; otherwise, change the sign of ΔV and swap voltage increment greater than dVmin to minimize the
points 0 and 1 to reestablish P1>P0. Repeat Step ① effect of equipment noise and sampling tolerance.
and update Vn=Vn−1+ΔV. ② Obtain the power increment dP and calculate
④ Observe Pn. If Pn>Pn−1, repeat P&O by adding the slope dP/dV.
1 to n. P&O continues as long as Pn>Pn−1. Therefore, ③ Determine if the absolute value |dP/dV| is
the power values exhibit an increasing order with approximately equal to zero. Assume a small positive
Pn>Pn−1>Pn−2> … >P0. tolerance value ε; if |dP/dV|>ε, then the MPP obtained
Once Pn<Pn−1, the P&O process ends, and the in Stage (2) is not correct or has drifted, and the MPPT
MPP is located within the section (Vn−2, Vn). If P&O must start over from Stage (1). Otherwise, |dP/dV| can
continues to look for the MPP, the system will oscillate be regarded as approximately equal to zero, and the
around the MPP with large time step and low accuracy. MPPT has been verified.
Therefore, the GSS must start as the next stage. Whenever it recognizes the MPP as not correct or
(2) GSS: With the GSS, the starting interval is having drifted, the GSS-based multi-stage MPPT
now updated as (Vn−3, Vn−2). The GSS procedure is as procedure will restart, and the three stages will be
follows. repeated. The full procedure of the multi-stage MPPT
① Insert a point Vn−1 that divides section (Vn−3, strategy is illustrated in Fig. 4.
 30 Chinese Journal of Electrical Engineering, Vol.7, No.4, December 2021

Fig. 5 Control block diagram of the single-phase

grid-connected PV system

chopper decreases, while the dc-ac inverter continues

to inject the same amount of power into the grid. Thus,
the dc-link capacitor must discharge, and the dc-link
voltage will decrease. To maintain the dc-link voltage,
the dc-ac inverter must reduce the power injected into
the grid until it reaches the new MPPT power.
The switching frequency of the boost chopper
and the bridge inverter can be different. From the
control perspective, it means the front end and
following stages are decoupled through the dc-link
capacitor. The decoupling control strategy is also
[29]
suited to multi-input hybrid systems , where several
input PV modules or wind sources share a common dc
bus and inverter.
Fig. 5 shows a control block diagram of the PV
system with the GSS-based MPPT algorithm. The PV
array block represents the model of the PV
Fig. 4 Procedure of the GSS-based MPPT with P&O and INC characteristics. The voltage controller of the boost
chopper and the voltage controller of the inverter are
3 System configuration and modeling both PI compensators. The reference of the voltage
controller for the chopper is given by the MPPT
The single-phase grid-connected inverter system
controller. However, the reference voltage of the
contains a boost chopper and a dc-ac inverter, where
the boost chopper operates in the voltage control mode. voltage controller for the inverter is set based on the
The GSS-based multi-stage MPPT strategy is applied amplitude of the grid voltage. The predictive control
to the boost chopper stage, where the output voltage of algorithm is employed here in the current controller
the PV array is controlled by a proportional-integral for the inverter to achieve high-quality grid current
(PI) controller to trace the reference generated by the injection.
MPPT controller. Both the PI and MPPT controllers The installed roof PV array consists of 12
are included in the MPPT algorithm block of Fig. 5. modules in total and 3 strings in parallel, where each
While the boost chopper is responsible for tracking the string has 4 modules in series. The parameters of each
MPP, the inverter is then used to control the dc-link module and the whole array are listed in Tab. 1 under
2
voltage, which indirectly transfers the MPPT power normal conditions of 1 000 W/m and 25 ℃. All
from boost chopper to the grid. When there is a cloud modules are assumed as identical with the same solar
over the PV array, the power extracted by the boost insolation.
 Shuang Xu et al.: Single-phase Grid-connected PV System with Golden Section Search-based MPPT Algorithm 31

Tab. 1 Parameters of PV modules and array

Quantity Module Array

Open circuit voltage VOC/V 43.2 172.8

Short circuit current ISC/A 4.7 14.1

Maximum power voltage VMPP/V 34.6 138.4

Maximum power current IMPP/A 4.35 13.05

Maximum power PMPP/W 150 1 800

Mathematical model of the PV array is given by Fig. 7 P-V characteristics of the PV array
the equation as under different conditions
⎛ q(V + IR ) ⎞ The function of the characteristic curves in Eq. (7)
I =I ph − I D =I ph − I o ⎜ exp( a ) − 1⎟ (6)
⎝ πkTa ⎠ is embedded in the Matlab/Simulink model in Fig. 8.
where Iph is photocurrent, Io is saturation current of the With the given solar insolation and temperature, the
diode, q is elementary charge (1.602 176 46×10−19 C), characteristic curves can determine the relationship
n is the diode ideality factor, k is Boltzmann constant between the voltage V and current I at the PV terminal.
−23
(1.380 650 3×10 J/K), and Ta is absolute temperature. With a specific load, there exists a unique operating
Following the I-V characteristic equation of the equilibrium point, which can be automatically
PV array from Eq. (6), the nonlinear relationship can obtained then this block is inserted model into the
be represented by Eq. (7). The values and parameters Matlab/Simulink model.
are all given by the handbook and manual of the
manufacturers.
The output current is represented by
I = f (V , S , T ) (7)
where I and V are the output current and output
voltage of the PV array, respectively, S is the solar
irradiance level, and T is the temperature.
Fig. 8 Simulink model for PV array
With this equation, the model can be used in the
simulation to emulate the practical PV array. Fig. 6 To verify the GSS-based multi-stage MPPT
shows both the fitted and the manufacturer’s I-V and algorithm, the system shown in Fig. 9 was built in
P-V characteristics. Fig. 7 shows both the simulated Matlab/Simulink. The PV array as described by Fig. 8
and measured P-V curves under various conditions. It is represented by the block “PV Array” in Fig. 9, and
shows the fitted characteristic curves match the the GSS-based multi-stage MPPT algorithm is
manufacturer’s handbook and measured characteristic implemented as an S-function module written in C
curves, which validates the effectiveness of the language. The system parameters in simulation are
simulated model. configured to match the real system. For simplicity, a
200 V battery is connected as the load of the boost
chopper in the simulation for the MPPT performance.

Fig. 6 I-V and P-V characteristics of PV system with given Fig. 9 Simulink diagram for the single-phase grid-connected
temperature and insolation inverter system
 32 Chinese Journal of Electrical Engineering, Vol.7, No.4, December 2021

successfully locate the vicinity area. Secondly, GSS

4 Simulation and experimental results
stage takes another 4 search steps to reach the MPP.
Therefore, with 3 ms for each time interval, the
4.1 Simulation results
search time in total is 36 ms. Finally, INC stage
The simulation of the proposed MPPT technique verifies the located MPP, and the system reaches
was run in Matlab/Simulink with the ode23t steady state. The steady-state output power under
solver and sampling step size 1×10−7. Figs. 10 and 11 steady state is 397.7 W compared with the
show the simulation results of the proposed maximum PV output power of 397.9 W. The MPPT
multi-stage MPPT and P&O/INC under different efficiency is as high as 99.95%. In contrast, the
solar insolation and temperatures, which include the simulation results for the system with P&O/INC are
boost chopper voltage reference Vref, PV terminal shown in Fig. 10b. The perturbation step size was
voltage V, current I, and output power P. The time chosen optimally as 4.5 V for P&O/INC to achieve
interval of the search step is set as 3 ms. The search almost the same search speed as the proposed
step should be greater than the settling time to ensure multi-stage MPPT technique. In this manner, there
that the system is able to track the updated MPPT exists oscillation under steady state, the average
during each time interval. output power is reduced to 395.3 W, and the MPPT
efficiency is reduced to 99.35%.
Under a different solar insulation and temperature
2
of S=930 W/m , T=37.5 ℃, Fig. 11 shows the similar
results. From Fig. 11a, firstly, P&O stage of the
proposed MPPT technique has 7 search steps to
successfully locate the vicinity area. Secondly, GSS
stage takes another 5 search steps to reach the MPP.
Therefore, with 3 ms for each time interval, the search
time in total is 36 ms. Finally, INC stage verifies the
located MPP, and the system reaches steady state. The
steady-state output power and search accuracy are
1 438.6 W and 99.94%, respectively, as been put in
Tab. 2. In contrast, the simulation results for the
system with P&O/INC are shown in Fig. 11b. The
search time of P&O/INC is close to 36 ms, but the
output power drops to 1 437.3 W and the MPPT
efficiency becomes 99.85%.

Fig. 10 Simulation results under the conditions of S=237

W/m2, T=13.9 ℃

From Fig. 10a, the GSS-based multi-stage

MPPT has 3 stages. Firstly, P&O stage of the
proposed MPPT technique has 8 search steps to
 Shuang Xu et al.: Single-phase Grid-connected PV System with Golden Section Search-based MPPT Algorithm 33

algorithm. The following figures show the data

obtained from the experiment and plotted in Matlab.

Fig. 11 Simulation results under the conditions of S=930 Fig. 12 Prototype of hybrid PV and wind converter
W/m2, T=37.5 ℃
Fig. 13 plots the GSS-based multi-stage MPPT
Tab. 2 Search time and accuracy results for different performance under certain weather conditions. The
MPPT methods experimental results are consistent with the simulation
Search time/ Search results in Figs. 10 and 11 for the same weather
MPPT methods
ms accuracy (%) conditions. Fig. 14 shows the system behavior under
Proposed multi-stage MPPT with minor environmental condition changes. The real
36 99.94
GSS system can catch up with condition changes within 30
P&O [9] 28 500 98.28 ms, which is also consistent with the simulation
Modified adaptive hill climbing [9]
23 500 98.45 results.
[14]
Fixed step size INC 7 500 98.90
[14]
Variable step size INC 1 500 99.20

Some facts should be noted from the simulation

results, for example, the ripples of the voltage are the
most prominent because the search continues even
after MPP has been found. In other words, the MPPT
algorithm still perturbs the located voltage to verify if
the located MPP is the real MPP. According to the P-V
characteristic shown in Fig. 7, around the MPP, the
voltage oscillation does not necessarily lead to power Fig. 13 Experimental waveforms of GSS-based MPPT with
oscillation. insolation of 930 W/m2 and temperature of 37.5 ℃

4.2 Experimental results

In addition to the simulations, verification

experiments were conducted on the real system. The
experimental setup was based on a wind/solar hybrid
inverter system, where the wind has a power rating of
10 kW and the PV has a power rating of 2 kW. The
wind power system is paralleled with the PV system at
the dc-link through a diode rectifier and boost
converter. The experimental setup is shown in Fig. 12,
where a DSP TMS320F28335 was used as the Fig. 14 Transient response of the GSS-based MPPT algorithm
microcontroller to implement the GSS-based MPPT with disturbance of solar insolation
 34 Chinese Journal of Electrical Engineering, Vol.7, No.4, December 2021

Fig. 15 plots the steady state waveforms of the experimental results. The simulation and experimental
grid voltage, current, and dc-link voltage. The grid results verified the feasibility and effectiveness of the
current must be in phase with the grid voltage for a proposed MPPT technique.
unity power factor. The dc-link voltage was set to 200 The proposed MPPT technique mainly targets the
V. As discussed previously, 120 Hz ripples exist in the MPP of a unimodal function, i.e., the curve has only
dc-link voltage. However, the ripples barely affect the one extremum, but it is not suited to multimodal
quality of the output current of the inverter. In this functions, i.e., the curve has multiple extrema when
case, the total harmonic distortion (THD) of the partial shading happens. Thus, to apply the proposed
current is 1.8%, and the power factor is unity. The MPPT technique to partial shading conditions, more
search time and accuracy of the proposed multi-stage points must be inserted into the GSS scheme to locate
MPPT method with GSS are compared with other the MPP. The application of the proposed MPPT
MPPT methods in Tab. 2, where the proposed MPPT technique under the partial shading condition remains
method improves the search time significantly as future work.
compared with P&O and INC under grid-connected
References
operation. The search times of P&O and INC are
highly dependent on the step sizes of the [1] U.S. Energy Information Administration (EIA), Where
corresponding methods. The search time of P&O and solar is found and used. [2020-09-05]. https://
INC can be significantly improved by sacrificing www.eia.gov/energyexplained/solar/where-solar-is-found.
search accuracy. php.
[2] L Zhang, F K Jiang, D W Xu, et al. Two-stage
transformerless dual-buck PV grid-connected inverters
with high efficiency. Chinese Journal of Electrical
Engineering, 2018, 4(2): 36-42.
[3] S Xu, L Chang, R Shao. Evolution of single-phase power
converter topologies underlining power decoupling.
Chinese Journal of Electrical Engineering, 2016, 2(1):
24-39.
[4] R W Erickson, D Maksimovic. Fundamentals of power
electronics. Boston: Springer, 2001.
[5] P K Bonthagorla, S Mikkili. Performance investigation of
Fig. 15 Experimental waveforms of the single-phase
hybrid and conventional PV array configurations for
grid-connected inverter
grid-connected/standalone PV systems. CSEE J. Power
Energy Syst., 2020: 1-16.
5 Conclusions [6] C Schaef, J T Stauth. Multilevel power point tracking for
In this paper, a GSS-based multi-stage MPPT partial power processing photovoltaic converters. IEEE J.

technique that combines GSS with P&O and INC was Emerg. Sel. Top. Power Electron., 2014, 2(4): 859-869.

proposed to achieve fast convergence to the MPP with [7] F Li, C B Li, K Sun, et al. Capacity configuration of
hybrid CSP/PV plant for economical application of solar
small oscillation. The proposed multi-stage MPPT
energy. Chinese Journal of Electrical Engineering, 2020,
technique starts by searching the vicinity of the MPP
6(2): 19-29.
by P&O in the first stage. Then, it quickly narrows the
[8] N Femia, G Petrone, G Spagnuolo, et al. Optimization of
width of the interval with the golden ratio in the
perturb and observe maximum power point tracking
second stage, and the located MPP by is verified by method. IEEE Trans. Power Electron., 2005, 20(4):
the INC method in the third stage. The advantages of 963-973.
P&O, GSS, and INC are exploited, while their [9] W D Xiao, W G Dunford. A modified adaptive hill
disadvantages are avoided. Therefore, this MPPT climbing MPPT method for photovoltaic power systems.
algorithm provides fast response and high tracking in 2004 IEEE 35th Annual Power Electronics Specialists
efficiency, as shown through the simulation and Conference (IEEE Cat. No.04CH37551), 2004, 3(3):
 Shuang Xu et al.: Single-phase Grid-connected PV System with Golden Section Search-based MPPT Algorithm 35

1957-1963. algorithms for PV arrays. in 2020 6th IEEE International

[10] M Dehghani, M Taghipour, G B Gharehpetian, et al. Energy Conference (ENERGYCon), 2020: 819-823.
Optimized fuzzy controller for MPPT of grid-connected [23] B Mbarki, F Farhani, A Zaafouri. Comparative study of
PV systems in rapidly changing atmospheric conditions. J. some maximum power point tracking algorithms. in 2019
Mod. Power Syst. Clean Energy), 2020: 1-8. International Conference on Signal, Control and
[11] A Safari, S Mekhilef. Simulation and hardware Communication (SCC), 2019: 145-149.
implementation of incremental conductance MPPT with [24] A Ibrahim, MB Shafik, Min Ding, et al. PV maximum
direct control method using Cuk converter. IEEE Trans. power-point tracking using modified particle swarm
Ind. Electron., 2011, 58(4): 1154-1161. optimization under partial shading conditions. Chinese
[12] E Roman, R Alonso, P Ibanez, et al. Intelligent PV Journal of Electrical Engineering, 2020, 6(4): 106-121.
module for grid-connected PV systems. IEEE Trans. Ind. [25] Z Yi, Y Xu, W Gu, et al. Distributed model predictive
Electron., 2006, 53(4): 1066-1073. control based secondary frequency regulation for a
[13] A Pandey, N Dasgupta, A K Mukerjee. Design issues in microgrid with massive distributed resources. IEEE
implementing MPPT for improved tracking and dynamic Transactions on Sustainable Energy, 2021, 12(2):
performance. in IECON 2006 - 32nd Annual Conference 1078-1089.
on IEEE Industrial Electronics, 2006: 4387-4391. [26] E I Batzelis, S A Papathanassiou, B C Pal. PV system
[14] F Liu, S Duan, F Liu, et al. A variable step size INC control to provide active power reserves under partial
MPPT method for PV systems. IEEE Trans. Ind. Electron., shading conditions. IEEE Transactions on Power
2008, 55(7): 2622-2628. Electronics, 2018, 33(11): 9163-9175.
[15] N Femia, G Petrone, G Spagnuolo, et al. A technique for [27] P E Gill, W Murray, M H Wright. Practical optimization.
improving P&O MPPT performances of double-stage Pittsburgh: Academic Press, 1981.
grid-connected photovoltaic systems. IEEE Trans. Ind. [28] R M Shao, L C Chang. A new maximum power point
Electron., 2009, 56(11): 4473-4482. tracking method for photovoltaic arrays using golden
[16] H Rezk, M Aly, M Al-Dhaifallah, et al. Design and section search algorithm. in 2008 Canadian Conference
hardware implementation of new adaptive fuzzy on Electrical and Computer Engineering, 2008:
logic-based MPPT control method for photovoltaic 000619-000622.
applications. IEEE Access, 2019(7): 106427-106438. [29] Y Chen, Y Liu, S Hung, et al. Multi-input inverter for
[17] F Lin, K Lu, B Yang. Recurrent fuzzy cerebellar model grid-connected hybrid PV/wind power system. IEEE
articulation neural network based power control of a Trans. Power Electron., 2007, 22(3): 1070-1077.
single-stage three-phase grid-connected photovoltaic
Shuang Xu (S’15-M’18) received the
system during grid faults. IEEE Transactions on
B.Sc.E.E. in 2012 from Hefei University of
Industrial Electronics, 2017, 64(2): 1258-1268.
Technology, Hefei, China, and the Ph. D. in
[18] F Fateh, W N White, D Gruenbacher. A maximum power Electrical Engineering in 2018 at the
tracking technique for grid-connected DFIG-based wind University of New Brunswick (UNB),
turbines. IEEE J. Emerg. Sel. Top. Power Electron., 2015, Fredericton, Canada.
3(4): 957-966. He was a Post-Doctoral Fellow with the

[19] R C N Pilawa-Podgurski, W Li, I Celanovic, et al. Emera and NB Power Research Centre for
Smart Grid Technologies at UNB and a Post-Doctoral Associate at
Integrated CMOS energy harvesting converter with digital
Western University, London, Canada, from 2018 to 2021. He is
maximum power point tracking for a portable
currently a Guest Professor in the School of Electrical and Control
thermophotovoltaic power generator. IEEE J. Emerg. Sel. Engineering at North China University of Technology, Beijing, China.
Top. Power Electron., 2015, 3(4): 1021-1035. His research interests include renewable energy systems, energy storage
[20] B Subudhi, R Pradhan. A comparative study on maximum technologies, power electronics, and power system support functions
power point tracking techniques for photovoltaic power for distributed energy resources.

systems. IEEE Trans. Sustain. Energy, 2013, 4(1): 89-98. Dr. Xu was a recipient of the IEEE Applied Power Electronics
Conference Outstanding Presentation Award in 2017. He was the
[21] T Esram, P L Chapman. Comparison of photovoltaic array
Tutorial Speaker in IEEE PEDG 2019, 2021, IEEE CPERE 2019, IEEE
maximum power point tracking techniques. IEEE Trans.
HVDC 2020, and Session Chair of the IEEE Symposium on Power
Energy Convers., 2007, 22(2): 439-449. Electronics for Distributed Generation Systems (PEDG 2019). He also
[22] N C Swanepoel, C G Richards, A F Nnachi, et al. A serves as the Treasurer in 2018 and the Chair in 2019 of IEEE UNB
comparative study of maximum power point tracking Power Electronics Society Student Branch Chapter.
 36 Chinese Journal of Electrical Engineering, Vol.7, No.4, December 2021

Riming Shao (M’08) received the B.Sc.E.E. in Liuchen Chang (S’80-M’92-SM’99) received
1994 and M.Sc. in 1997 from Tongji B.S.E.E. from Northern Jiaotong University in
University, Shanghai, China; and the Ph.D.
1982, M.Sc. from China Academy of Railway
degree in 2010 from the University of New
Brunswick, Fredericton, Canada. He currently Sciences in 1984, and Ph.D. from Queen’s
works as a Project Engineer with Alpha University in 1991. He joined the University of
Technologies at Bellingham, Washington State, New Brunswick in 1992 and is a Professor in
USA. His research interests include smart grids, Electrical and Computer Engineering. He was
power converters, renewable energy systems,
the NSERC Chair in Environmental Design
and distributed power generation systems.
Dr. Shao is a registered Professional Engineer of Engineers and Engineering during 2001-2007, and was the Principal Investigator of
Geoscientists New Brunswick, Canada. Canadian Wind Energy Strategic Network (WESNet) during 2008-2014.
He is currently the President of the IEEE Power Electronics Society.
Bo Cao (S’08-M’15) received the B.Sc.E.
Dr. Chang was a recipient of CanWEA R.J. Templin Award in
degree from East China University of
2010 for his contribution in the development of wind energy
Science and Technology, Shanghai, China,
in 2005; and the Ph.D. degree from the technologies, and the Innovation Award for Excellence in Applied
University of New Brunswick, Fredericton, Research in New Brunswick in 2016. He is a Fellow of Canadian
NB, Canada, in 2015. He is currently a
Academy of Engineering (FCAE). He has published more than 340
Research Associate with the Emera and NB
Power Research Centre for Smart Grid refereed papers in journals and conference proceedings. Dr. Chang has
Technologies at the University of New focused on research, development, demonstration and deployment of
Brunswick. His principle research interests include power converter renewable energy based distributed generation systems and direct load
design, grid-integration technology, distributed generation system
control systems.
and smart grid techniques.