Professional Documents
Culture Documents
Energy Reports
journal homepage: www.elsevier.com/locate/egyr
Research Paper
A R T I C L E I N F O A B S T R A C T
Keywords: The increasing demand for clean energy to address the looming energy crisis has led to the widespread use of
Data-driven photovoltaic-grid systems (DPGS) photovoltaic grid-connected technology, particularly in microgrids. To fully harness solar energy, this study
Maximum power point tracking (MPPT) proposes a data-driven strategy for photovoltaic maximum power point tracking with adaptive adjustment to
Online tracking
environmental dynamics. Exploiting deep learning and incremental adjustment, our data-driven photovoltaic-
Deep learning
grid systems (DPGS) upgrade the traditional perturbation and observation (P&O) MPPT to a dynamic evolu
Two-stage single-phase inverter
tionary scheme. DPGS gathers the photovoltaic panel’s output voltage and current, calculates the current power,
and then outputs the appropriate reference voltage based on the power difference. The photovoltaic voltage is
then adjusted using a data-driven strategy. In this study, a double-hidden layer deep learning network is utilized
to output the prediction control signal of the first-stage circuit while continuously modifying the weight matrix
and optimizing the tracking parameters of DPGS. Besides, a two-stage single-phase grid-connected photovoltaic
inverter is designed to handle environmental dynamics. The simulation results validate the reliability of our
suggested DPGS. DPGS often responds within 0.4 s, which is 33 % faster than conventional P&O techniques.
DPGS has a power ripple rate that is approximately 78 % greater than conventional P&O approaches, at 0.022 %.
DPGS has a quicker response time and less power fluctuation under external interference than traditional P&O
MPPT. Our study contributes to the efficiency and reliability enhancement of grid-connected photovoltaic sys
tems and has wide application in renewable energy systems.
* Corresponding author.
E-mail addresses: tianluyu@tju.edu.cn (L. Tian), dong0120@e.ntu.edu.sg (C. Dong), yunfeimu@tju.edu.cn (Y. Mu), hjjia@tju.edu.cn (H. Jia).
https://doi.org/10.1016/j.egyr.2024.01.038
Received 4 June 2023; Received in revised form 2 November 2023; Accepted 16 January 2024
Available online 30 January 2024
2352-4847/© 2024 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
L. Tian et al. Energy Reports 11 (2024) 1910–1924
Although the efficiency of photovoltaic inverters can reach around 98 % photovoltaic fault diagnosis (Yuan et al., 2022), photovoltaic power
(Chen et al., 2012), the current photoelectric conversion efficiency of prediction (Ferrero Bermejo et al., 2019), grid connection control of
commercial photovoltaic panels is typically less than 20 % (Razykov photovoltaic power generation (Aref et al., 2023), and MPPT manage
et al., 2011). This is mainly due to material constraints of mono ment of photovoltaic maximum power point tracking in solar
crystalline silicon or polycrystalline silicon solar panels that result in low grid-connected systems (Kermadi and Berkouk, 2017). Many studies use
photoelectric conversion efficiency, thereby reducing photovoltaic machine learning techniques to coordinate PV grid-connected systems
panel efficiency (Yun, 2017). Therefore, photovoltaic power generation (Lim and Yoon, 2022). For instance, a literature review presented an
must be equipped with MPPT maximum power point tracking to MPPT method of ANN combined incremental conductance method (Lee
improve power generation efficiency. MPPT control technology uses et al., 2010), which accurately searched the maximum power based on
power tracking to acquire the best energy, enabling photovoltaic panels feedback voltage and current to approach the maximum power point
to adapt to changes in environmental factors (Inthamoussou et al., quickly. Combining various PID controllers with machine learning
2012). Various MPPT algorithms have been examined so far, including techniques serves to improve the performance of the controller itself.
the simple-to-implement perturbation and observation (P&O) algorithm Artificial intelligence algorithms are also used to enhance the
(Saravanan and Babu, 2016), the incremental conductance algorithm grid-connected current from photovoltaic systems connected to the grid.
(Aldair et al., 2018), fuzzy logic control (Yap et al., 2020), and other new In the literature, a single-phase LCL photovoltaic grid-connected system
techniques (Bendib et al., 2015). For instance, to manage the maximum was built, and a neural network was employed to improve both the
power of solar under shadow conditions, some studies utilize particle photovoltaic grid-connected waveform and the grid-connected PID
swarm optimization (PSO) in conjunction with the disturbance moni controller (Qu, 2022). In addition, a PID controller for the DC-AC con
toring approach (Lian et al., 2014). However, the conventional verter’s efficiency increase was designed, which enhanced the
grid-connected photovoltaic control algorithm has fixed parameters and grid-connected photovoltaic side’s performance (Lakshmi and Reddy,
a limited capacity to respond to environmental dynamics. To increase 2022). At the same time, data-driven involves collecting and analyzing
the accuracy of MPP collection, numerous research studies have inte data (Bertsimas et al., 2018), allowing the model to adjust according to
grated MPPT algorithms with machine learning techniques. different data (Montans et al., 2019). In addition, data-driven requires a
Machine learning algorithms have distinct advantages, making them large amount of data to adjust the model or model parameters, which
frequently used in photovoltaic power-generating control. Artificial greatly reduces the system’s dependence on the model and makes it
neural networks, in particular, stand out among other approaches due to more suitable for different scenarios (Silva et al., 2018).
their ability to determine the relationship between various factors Fig. 1 shows the closed-loop regulation of voltage in MPPT control.
without relying on complex mathematical models, allowing for a more There have been several studies focusing on improving the PID
accurate determination (Li et al., 2022). Moreover, artificial neural controller for MPPT in solar energy systems. For example, (Harrag and
networks offer off-line training, nonlinear coupling, and high-speed Messalti, 2015) used a genetic algorithm (GA) to adjust PID parameters
response advantages (Dong et al., 2021). They are widely employed in and change the step size of P&O (Sibtain et al., 2021). proposed a
1911
L. Tian et al. Energy Reports 11 (2024) 1910–1924
Fig. 2. Complete schematic diagram of the photovoltaic grid-connected system with DPGS.
fractional PID controller for both photovoltaic and wind power systems. designed specifically for inaccurate photovoltaic grid-connected systems
In (Yang et al., 2018a), (Thangam and Muthuvel, 2022), and (Yang that complete MPPT without relying on precise PV models. Despite these
et al., 2018b), passive fractional controllers were designed, with each advances, there are not many matching PID controllers that can
adopting different meta-heuristic algorithms to update parameter extremely efficiently and economically achieve closed-loop control of
values. In addition, since passive PID does not require current to the voltage at the MPP, especially in the event of extreme environmental
participate in regulation, additional overcurrent protection devices may changes to achieve adaptive adjustment. Additional research is neces
be needed for the photovoltaic system. Many studies have also explored sary to address this gap in the existing literature.
fuzzy control strategies for MPPT. For instance, (Chouksey et al., 2020) To enhance the effectiveness of photovoltaic maximum power
created a fuzzy PID controller based on adaptive gain scheduling and tracking, this study proposes a data-driven photovoltaic grid-connected
used an improved artificial neural network-based particle swarm opti system that combines deep learning with photovoltaic grid-connected
mization approach to reduce steady-state error and achieve system control. The aim is to tackle the aforementioned challenges. This strat
stability. In Yang et al. (2018c) a fractional sliding mode controller was egy can be seamlessly integrated with photovoltaic grid-connected
1912
L. Tian et al. Energy Reports 11 (2024) 1910–1924
MPPT control. The following outlines the key contributions of this 1. Online regulation in MPPT control;
paper: 2. Self-adjustment of DPGS parameters;
3. Adaptability for variable environment.
1. A two-stage single-phase photovoltaic grid-connected system that
incorporates deep learning. By doing so, a data-driven grid-con The main structure of the rest part of this paper is as follows: Section
nected system is developed; 2 introduces the construction of a two-stage single-phase photovoltaic
2. A novel approach that combines deep learning and PID controller is grid-connected system and related device parameters; Section 3 in
proposed, wherein the improved PID controller is designed using a troduces the design of DPGS; Section 4 introduces the control algorithm.
deep neural network structure. This controller is capable of adjusting The simulation platform is constructed and tested in Section 5. Section 6
PID parameters adaptively for photovoltaic grid-connected systems, draws the conclusion.
accounting for changes in environmental factors and achieving ac
curate MPPT regulation; 2. Photovoltaic inverter model
3. To verify the effectiveness of the proposed strategy, a single-phase
photovoltaic grid-connected system was built, and the system was In this paper, we aim to test the effectiveness of the proposed MPPT
tested in various environments. A comparison of test results led to strategy enhanced by deep-BPNN as an auxiliary system. To achieve
conclusions on the system’s performance. this, we constructed a two-stage single-phase grid-connected photo
voltaic inverter model. The first-stage utilized the boost circuit to
Fig. 2 illustrates the structure of the entire DPGS. In this paper, the manage the MPPT control of the photovoltaic panel. In the second-
deep learning method is utilized to improve conventional photovoltaic stage, we selected the T-three-current H-bridge inverter topology to
PID by realizing the following advantages in variable environments: enhance the inverter waveform and minimize switching stress. Fig. 3
1913
L. Tian et al. Energy Reports 11 (2024) 1910–1924
circuit generates the DC bus voltage required to raise the solar voltage
for grid connection.
3. Composition of DPGS
1914
L. Tian et al. Energy Reports 11 (2024) 1910–1924
structure, and corresponding parameters. Additionally, the activation 3.3. Forward propagation in DPGS
function of each layer and the initial value of the weight coefficient
matrix must be established. In the forward propagation process, the output of the input layer of
the double-layer neural network is:
Traditional PID controllers consist of three parameters: kp, ki, and kd. Where, i = 1, 2, 3, 4, is the i-th element of the input vector I.
The control quality of the controller depends on the value of the pa The output of the input layer xout,i (k) serves as the input of hidden
(1)
output response y, and the feedback deviation e as input. The ideal kp, ki, v(1)
j1 (k) = w(1) (1) (1)
j1i xout,i = W I (6)
and kd parameters will be output by the DL-PID in accordance based on
i=1
The number of input elements is four and the number of neurons in both
2. wj2j1 is the weight coefficient matrix from the j1-th neuron to the j2-th
(2)
hidden layer 1 and hidden layer 2 is five, meanwhile, the number of
output elements is three. The structure is shown in Fig. 8. neuron in the hidden layer 1 to the hidden layer 2.
W(1) is the weight coefficient matrix of the hidden layer 1. W(2) is the The output vector of hidden layer 2 is:
weight coefficient matrix of the hidden layer 2. W(3) is the weight co ( )
efficient matrix of the output layer. Furthermore, f (1) (⋅)andf (2) (⋅)are the x(3)
out,j2 (k) = f
(2)
v(2)
j2 (k) (9)
activation function of hidden layer 1 and 2 respectively, whilef (3) (⋅) is Finally, since the parameters in the PID controller cannot be nega
the activation function of the output layer. tive, the neuron activation function of the output layer selects the non-
The input vector I of the neural network corresponds to the input r, negative Sigmoid function, and the expression is as follows.
output y, deviation e, and system bias at different times of the system
(the value is set to 1). The output vector Y corresponds to the three
parameters kp, ki, and kd of the PID controller respectively.
1915
L. Tian et al. Energy Reports 11 (2024) 1910–1924
()
ex By using the fastest descent method to explore the error surface, the
f (3) ⋅ = (10)
ex + e− x weight matrix at the minimum value of E(k) can be obtained. When the
gradient is zero, the optimal value can be identified. By modifying the
The input of the output layer is:
weight along the negative gradient direction of the surface, the bottom
∑
5 of the error surface can be reached. The gradient formula and specific
(11) calculation formula of the output layer are:
(3) (3) (3)
vh (k) = whj2 xout,j2 = W (3) X (3)
j2
(3) ( )[
∂E(k) ∂E(k) ∂y(k) ∂u(k) ∂xout,h ∂v(3)h (k) y(k) − y(k − 1)
= ⋅ ⋅ ⋅ (3) ⋅ ⋅sgn ⋅ e(k)
Where, h= 1, 2, 3; vh (k)is the input of the h-th neuron in the current
(3)
∂w(3)
hj2
∂y(k) ∂u(k) ∂x(3)
out,h ∂vh ∂w(3) u(k) − u(k − 1)
( ) ( ) hj2 ( )] ( )
output layer. whj2 is the weight coefficient matrix from the j2-th neuron to
(3)
the h-th neuron in the hidden layer 2 to the output layer. − e k − 1 , e(k).e(k) − 2e k − 1 + e k − 2 ⋅f ′(3) v(3) (3)
h (k) ⋅xout,j2
1916
L. Tian et al. Energy Reports 11 (2024) 1910–1924
Table 1
Simulation parameters of PV Array.
(1) Temperature: 25 ◦ C Irradiance: 1000 W/m2
Project Parameter
Open-circuit voltage VOC /V 309.00
Short-circuit current ISC /A 13.63
Maximum power point power /W 3041.43
Maximum power point voltage/V 258.00
The DPGS satisfies the following algorithm steps for the first circuit:
1917
L. Tian et al. Energy Reports 11 (2024) 1910–1924
1918
L. Tian et al. Energy Reports 11 (2024) 1910–1924
Fig. 14. PV Array output power control by traditional PID and DPGS. (Temperature: 25 ◦ C, Irradiance: 1000 W/m2).
Fig. 15. The curve of SignalBuilder and PV output power (case 1).
Table 3
Comparison between DPGS and P&O MPPT (case1).
Maximum power Minimum power
Response time r%
(Temperature: 25 ◦ C Irradiance: 600 W/m2) (Temperature: 25 ◦ C Irradiance: 600 W/m2)
1919
L. Tian et al. Energy Reports 11 (2024) 1910–1924
Fig. 16. The curve of SignalBuilder and PV output power (case 2).
Table 4
Comparison between DPGS and P&O MPPT (case 2).
Maximum power Minimum power
Response time r%
(Temperature: 45 ◦ C Irradiance: 1000 W/m2) (Temperature: 45 ◦ C Irradiance: 1000 W/m2)
Table 5
Comparison between DPGS and P&O MPPT (case3).
Maximum power Minimum power
Response time r%
(Temperature: 45 ◦ C Irradiance: 1000 W/m2) (Temperature: 45 ◦ C Irradiance: 1000 W/m2)
Fig. 17. The curve of SignalBuilder and PV output power (case 3).
1920
L. Tian et al. Energy Reports 11 (2024) 1910–1924
Fig. 18. Grid connection voltage, current, and DC bus voltage of DPGS (case 3).
The preceding boost simulation circuit is shown in Fig. 12: environmental temperature is Celsius (◦ C).
The solar output power can be managed using DL-PID in conjunction
with the P&O algorithm, allowing it to remain stable at the MPP when
the ambient conditions vary. The Fig. 13. represents the internal struc 5.2. Results and discussion
ture of DL-PID.
sfun_BP is an S function written by .m file. In the program set 5.2.1. PV output power at constant temperature and irradiance
learning rate μ = 0.3, forgetting factor α = 0.05. After building the Setting the environmental temperature at 25 ◦ C and the irradiance at
model, run the simulation. The performance of DPGS was tested under 1000 W/m2, the comparison between the DL-PID controller and the
three different photovoltaic operating scenarios: environmental tem traditional PID controller used in the MPPT of PV shows the following
perature changes, irradiance changes, and both changes. The unit of results. Fig. 14 shows the PV Array output power control by traditional
PID and DPGS when the environmental temperature is set to 25 ◦ C and
1921
L. Tian et al. Energy Reports 11 (2024) 1910–1924
Fig. 20. Comparison of PV output power corresponding to three types of networks in case 1, case 2, and case 3.
1922
L. Tian et al. Energy Reports 11 (2024) 1910–1924
Table 6 at around 400 V, and the high-order harmonics in the inverter current
Comparison between DPGS, single hidden layer, three hidden layers, and P&O are removed through filtering to obtain the grid-connected current. The
MPPT. total output power is 2746.0 W, and the system efficiency is about 97.1
Three %.
Single Traditional
DPGS
hidden layer
hidden
P&O
Fig. 19 depicts the modification of the three DL-PID’s control pa
layers rameters kp, ki, and kd, during case 3. The research shows that it can
Response case1 0.43 0.30 0.44 0.62 successfully follow the PV array’s maximum power point in DPGS with
time case2 0.40 0.27 0.40 0.60 little power fluctuation, ensuring that the PV power output is constantly
(t/s) case3 0.42 0.28 0.42 /
maintained near the maximum value. DPGS has excellent environmental
Average response
time (t/s)
0.417 0.283 0.420 0.610 adaptability as well as strong self-adaptability.
Average power ripple 0.022
0.104 % 0.024 % 5.312 %
rate % 5.2.5. Comparison of DPGS with single hidden layer and three hidden
layers in stable PV output power
We compared the effectiveness of DPGS with single hidden layer
stabilization.
network and three hidden layers network in stabilizing PV output
It is revealed in Table 2. that the stability power ripple rate is about
power. The corresponding optical output power of these three networks
2.37 %, the highest power value is 3041.22 W, and the minimum power
under the three environmental changes of case 1 to case 3 is shown in
value is 2969.02 W. After the simulation began, the DL-PID controller
Fig. 20.
kept the PV Array’s output power steady at around 0.4 s. After stabili
At the same time, we calculated the power ripple of the photovoltaic
zation, the power reaches a maximum of 3041.34 W, a minimum of
panels of three types of networks after stabilization under three envi
3040.74 W, and a fluctuation rate of around 0.020 %.
ronmental conditions: 25 ◦ C 600 W/m2, 25 ◦ C 1000 W/m2, and 45 ◦ C
The output power of a PV Array controlled by a DL-PID is more
1000 W/m2. The response times of the three networks under case 1 to
stable, the response time is quicker, and the curve is smoother when the
case 3 operating conditions were averaged, which are shown in Table 6.
perturbation and observation method is also chosen as an MPPT algo
Fig. 20 and Table 6. show that all three networks can stabilize the PV
rithm. When compared to utilizing the conventional PID controller, the
output power under frequently changing environmental conditions, and
PV Array’s output power is constantly maintained at its maximum point,
the power ripple is greatly reduced when compared to traditional P&O.
and power fluctuation is much reduced.
However, the response time increases as the number of network layers
increases, which may be a result of the increase in data after the number
5.2.2. PV output power with constant temperature and varying irradiance
of network layers increases, leading to an increase in computational
The simulation time is set to 3 s. The temperature and irradiance
complexity.
change linearly and the change curve is generated by SignalBuilder. The
DPGS is superior to a single hidden layer network based on the data
performance of the DPGS remains accurate in three different situations.
results, which has a lower output power ripple after stabilization. The
The curve of SignalBuilder is set as depicted in Fig. 15 in the PV Array
tracking effect of the three hidden layers network is nearly identical to
temperature is constant and the irradiance varies in case 1. The tem
that of DPGS for the MPP of PV, increasing the system’s computation
perature value is always held at 25 ◦ C, and from 1 to 2 s, the irradiance
requirements. Additionally, the photovoltaic grid-connected system’s
increases from 600 to 1000 W/m2.
MPP only requires to collect data regarding voltage and current.
Table 3. shows the results of the PV output power in case 1. The
Continued growth in the number of hidden layers is unnecessary, and
average photovoltaic output power, whether utilizing conventional P&O
there might even be overfitting. Considering comprehensively, DPGS
or DPGS, is 1814 W, which is in line with the desired value at a tem
with double hidden layers has the best effect on MPPT under dramatic
perature of 25 ◦ C and an irradiation of 600 W/m2. Table 3 shows that
environmental changes.
while there is no significant difference in maximum power between the
P&O algorithm and the DPGS method, there is a substantial difference in
6. Conclusion
minimum power, which causes considerable power variations in con
ventional P&O algorithms.
A data-driven photovoltaic grid-connected system (DPGS) is
designed in this paper. The single-phase two-stage grid-connected sys
5.2.3. PV output power with varying temperature and constant irradiance
tem is selected as the research object, and the PID controller improved
In case2, when the irradiance is unchanged and the temperature
by the deep neural network can adjust the values of kp, ki, and kd
changes, the curve of SignalBuilder is set as shown in Fig. 16. The
adaptively according to the changes of environmental factors, which
irradiance remains at 1000 W/m2, and the temperature changes from 1 s
ensures the maximum power output of the photovoltaic grid-connected
to 2 s, increasing from 25 ◦ C to 45 ◦ C..
system is stable at the maximum value. DPGS with a double-hidden layer
The outcomes of case 2 are similar to those of case 1, with the
structure outperformed single-layer neural networks and three-layer
photovoltaic output power matching the predetermined value (in
networks. The stable power ripple of DPGS is 0.022 %, and the
Table 4., from 2 s to 3 s, temperature: 45 ◦ C, irradiance: 1000 W/m2).
response time is around 0.4 s. The results of the simulations show that
When employing the DPGS algorithm, the output power fluctuation of
DPGS works effectively with various environmental changes. Hence,
solar systems is greatly decreased, and the response time is improved by
DPGS provides an effective solution for industrial applications.
about 33 %. .
CRediT authorship contribution statement
5.2.4. PV output power with both varying temperature and irradiance
In case 3, the SignalBuilder curve is set as Fig. 17 when the irradiance
Chaoyu Dong: Conceptualization, Supervision, Writing – review &
varies concurrently with the temperature. From 0.5 to 1 s, the temper
editing, Project administration. Luyu Tian: Conceptualization, Soft
ature and irradiance rise from 600 W/m2 at 25 ◦ C to 1000 W/m2 at
ware, Methodology, Writing – original draft. Yunfei Mu: Supervision,
45 ◦ C. Waveform descends from 2 s to 3 s before returning to its initial
Writing – review & editing. Hongjie Jia: Supervision, Writing – review
value. Table 5 shows the comparison between DPGS and P&O MPPT in
& editing.
case 3.
Fig. 18 shows the grid-connected voltage, grid-connected current,
and the DC bus voltage of DPGS in case 3. The bus voltage is maintained
1923
L. Tian et al. Energy Reports 11 (2024) 1910–1924
Declaration of Competing Interest Lakshmi, G.V., Reddy, K.H., 2022. Improved tunicate swarm search-based MPPT for
photovoltaic on a "grid-connected" inverter system. Environ. Sci. Pollut. Res. 29,
78650–78665. https://doi.org/10.1007/s11356-022-21157-2.
The authors declare that they have no known competing financial Li, MR, Dong, CY, Xiong, BY, Mu, YF, Yu, XD, Xiao, Q, et al., STTEWS, 2022.
interests or personal relationships that could have appeared to influence A sequential-transformer thermal early warning system for lithium-ion battery
the work reported in this paper. safety. Applied Energy 328. https://doi.org/10.1016/j.apenergy.2022.119965.
Lian, K.L., Jhang, J.H., Tian, I.S., 2014. A maximum power point tracking method based
on perturb-and-observe combined with particle swarm optimization. IEEE J.
Data availability Photovolt. 4, 626–633. https://doi.org/10.1109/jphotov.2013.2297513.
Lim, S.-H., Yoon, S.-G., 2022. Dynamic DNR and solar PV smart inverter control scheme
using heterogeneous multi-agent deep reinforcement learning. Energies 15. https://
No data was used for the research described in the article. doi.org/10.3390/en15239220.
Liu, W.J., Niazi, K.A.K., Kerekes, T., Yang, Y.H., 2019. A review on transformerless step-
Acknowledgements up single-phase inverters with different DC-link voltage for photovoltaic
applications. Energies 12. https://doi.org/10.3390/en12193626.
Mazzeo, D., Matera, N., De Luca, P., Baglivo, C., Congedo, P.M., Oliveti, G., 2021.
This paper is supported by the National Natural Science Foundation A literature review and statistical analysis of photovoltaic-wind hybrid renewable
of China (U23B6006 and 52277116). system research by considering the most relevant 550 articles: an upgradable matrix
literature database. J. Clean. Prod. 295 https://doi.org/10.1016/j.
jclepro.2021.126070.
References Montans, F.J., Chinesta, F., Gomez-Bombarelli, R., Kutz, J.N., 2019. Data-driven
modeling and learning in science and engineering. Comptes Rendus Mec. 347,
Aldair, A.A., Obed, A.A., Halihal, A.F., 2018. Design and implementation of ANFIS- 845–855. https://doi.org/10.1016/j.crme.2019.11.009.
reference model controller based MPPT using FPGA for photovoltaic system. Renew. Nosratabadi, S.M., Hooshmand, R.A., Gholipour, E., 2017. A comprehensive review on
Sustain. Energy Rev. 82, 2202–2217. https://doi.org/10.1016/j.rser.2017.08.071. microgrid and virtual power plant concepts employed for distributed energy
Arcos-Pardo F.J., Monroy-Morales J.L., Hernández-Ángeles M., Campos-Gaona D. resources scheduling in power systems. Renew. Sustain. Energy Rev. 67, 341–363.
Modeling of photovoltaic grid connected generation system based on three level NPC https://doi.org/10.1016/j.rser.2016.09.025.
converter. In: Proceedings of the 2017 IEEE International Autumn Meeting on Razykov, TM, Ferekides, CS, Morel, D, Stefanakos, E, Ullal, HS, Upadhyaya, HM, 2011.
Power, Electronics and Computing (ROPEC) 2017. p. 1–6.http://doi.org/10.1109/ Solar photovoltaic electricity: Current status and future prospects. Sol. Energy 85,
ROPEC.2017.8261642. 1580–1608. https://doi.org/10.1016/j.solener.2010.12.002.
Aref, M., Abdelaziz, A.Y., Geem, Z.W., Hong, J., Abo-Elyousr, F.K., 2023. Oscillation Saravanan, S., Babu, R.N., 2016. Maximum power point tracking algorithms for
damping neuro-based controllers augmented solar energy penetration management photovoltaic system - a review. Renew. Sustain. Energy Rev. 57, 192–204. https://
of power system stability. Energies 16. https://doi.org/10.3390/en16052391. doi.org/10.1016/j.rser.2015.12.105.
Bajwa, A.A., Mokhlis, H., Mekhilef, S., Mubin, M., 2019. Enhancing power system Shrestha, A., Mustafa, A.A., Htike, M.M., You, V., Kakinaka, M., 2022. Evolution of
resilience leveraging microgrids: a review. J. Renew. Sustain. Energy 11. https://doi. energy mix in emerging countries: Modern renewable energy, traditional renewable
org/10.1063/1.5066264. energy, and non-renewable energy. Renew. Energy 199, 419–432. https://doi.org/
Bendib, B., Belmili, H., Krim, F., 2015. A survey of the most used MPPT methods: 10.1016/j.renene.2022.09.018.
Conventional and advanced algorithms applied for photovoltaic systems. Renew. Sibtain, D., Murtaza, A.F., Ahmed, N., Sher, H.A., Gulzar, M.M., 2021. Multi control
Sustain. Energy Rev. 45, 637–648. https://doi.org/10.1016/j.rser.2015.02.009. adaptive fractional order PID control approach for PV/wind connected grid system.
Bertsimas, D., Gupta, V., Kallus, N., 2018. Data-driven robust optimization. Math. Int. Trans. Electr. Energy Syst. 31 https://doi.org/10.1002/2050-7038.12809.
Program. 167, 235–292. https://doi.org/10.1007/s10107-017-1125-8. Silva, R.A.E., Baptista, J.M., Brito, M.C., 2018. Data-driven estimation of expected
Chen B., Sun P., Liu C., Chen C.-L., Lai J.-S., Yu W., et al. High efficiency transformerless photovoltaic generation. Sol. Energy 166, 116–122. https://doi.org/10.1016/j.
photovoltaic inverter with wide-range power factor capability. In: Proceedings of the solener.2018.03.039.
27th Annual IEEE Applied Power Electronics Conference and Exposition (APEC). Qu H. , 2022. Research on control strategy of photovolatic grid connected inverter based
Orlando, FL 2012. p. 575–582. http://doi.org/10.1109/APEC.2012.6165877. on improved BP neural network and PID. Hunan Univ. of Tech. http://doi.org/10
Chou K.-Y., Yang C.-S., Chen Y.-P., 2020. Ieee. Deep Q-network based global maximum .27730/d.cnki.ghngy.2022.000446.
power point tracking for partially shaded PV system. 7th IEEE International Lee H.H., Phan Quoc D., Le Dinh K., Le Minh P., Nguyen Truong Dan V. The New
Conference on Consumer Electronics - Taiwan (ICCE-Taiwan). Taoyuan, TAIWAN. Maximum Power Point Tracking Algorithm using ANN-Based Solar PV Systems. IEEE
http://doi.org/10.1109/ICCE-Taiwan49838.2020.9258116. Region 10 Conference on TENCON 2010. Fukuoka, JAPAN2010. p. 2179–2184. htt
Cabrera-Tobar, A, Bullich-Massagué, E, Aragüés-Peñalba, M, Gomis-Bellmunt, O, 2016. p://doi.org/10.1109/TENCON.2010.5686721.
Review of advanced grid requirements for the integration of large scale photovoltaic Thangam, T., Muthuvel, K., 2022. Passive fractional-order proportional-integral-
power plants in the transmission system. Renew. Sust. Energ. Rev. 62, 971–987. derivative control design of a grid-connected photovoltaic inverter for maximum
https://doi.org/10.1016/j.rser.2016.05.044. power point tracking. Comput. Electr. Eng. 97 https://doi.org/10.1016/j.
Chouksey, A., Awasthi, S., Singh, S.K., 2020. Fuzzy cognitive network-based maximum compeleceng.2021.107657.
power point tracking using a self-tuned adaptive gain scheduled fuzzy proportional Wang, H.M., Wang, G.Q., Qi, J.C., Schandl, H., Li, Y.M., Feng, C.Y., et al., 2020. Scarcity-
integral derivative controller and improved artificial neural network-based particle weighted fossil fuel footprint of China at the provincial level. Appl. Energy 258.
swarm optimization. Fuzzy Sets Syst. 381, 26–50. https://doi.org/10.1016/j. https://doi.org/10.1016/j.apenergy.2019.114081.
fss.2019.02.007. Williams, N.J., Jaramillo, P., Taneja, J., Ustun, T.S., 2015. Enabling private sector
Dong, C., Chu, R., Morstyn, T., McCulloch, M.D., Jia, H., 2021. Online rolling investment in microgrid-based rural electrification in developing countries: a review.
evolutionary decoder-dispatch framework for the secondary frequency regulation of Renew. Sustain. Energy Rev. 52, 1268–1281. https://doi.org/10.1016/j.
time-varying electrical-grid-electric-vehicle system. IEEE Trans. Smart Grid 12, rser.2015.07.153.
871–884. https://doi.org/10.1109/TSG.2020.3020983. Yang, B., Yu, T., Shu, H., Zhu, D., An, N., Sang, Y., et al., 2018a. Energy reshaping based
Dong, C., Wang, J., Li, X., Jin, C., Jiang, W., Zhang, Z., et al., 2023. Synthetic resilience passive fractional-order PID control design and implementation of a grid-connected
exploration and economic defense strategy for microgrid-level AC/DC hybrid energy PV inverter for MPPT using grouped grey wolf optimizer. Sol. Energy 170, 31–46.
system. IEEE Trans. Power Electron. 38, 2565–2576. https://doi.org/10.1109/ https://doi.org/10.1016/j.solener.2018.05.034.
tpel.2022.3207767. Yang, B., Yu, T., Shu, H., Zhu, D., Zeng, F., Sang, Y., et al., 2018b. Perturbation observer
Ferrero Bermejo, J., Gomez Fernandez, J.F., Olivencia Polo, F., Crespo Marquez, A., based fractional-order PID control of photovoltaics inverters for solar energy
2019. A review of the use of artificial neural network models for energy and harvesting via Yin-Yang-Pair optimization. Energy Convers. Manag. 171, 170–187.
reliability prediction. A study of the solar PV, hydraulic and wind energy sources. https://doi.org/10.1016/j.enconman.2018.05.097.
Appl. Sci. -Basel 9. https://doi.org/10.3390/app9091844. Yang, B., Yu, T., Shu, H., Zhu, D., An, N., Sang, Y., et al., 2018c. Perturbation observer
Gutierrez-Oliva, D., Colmenar-Santos, A., Rosales-Asensio, E., 2022. A review of the state based fractional-order sliding-mode controller for MPPT of grid-connected PV
of the art of industrial microgrids based on renewable energy. Electronics 11. inverters: design and real-time implementation. Control Eng. Pract. 79, 105–125.
https://doi.org/10.3390/electronics11071002. https://doi.org/10.1016/j.conengprac.2018.07.007.
Harrag, A., Messalti, S., 2015. Variable step size modified P&O MPPT algorithm using Yap, K.Y., Sarimuthu, C.R., Lim, J.M.Y., 2020. Artificial intelligence based MPPT
GA-based hybrid offline/online PID controller. Renew. Sustain. Energy Rev. 49, techniques for solar power system: a review. J. Mod. Power Syst. Clean. Energy 8,
1247–1260. https://doi.org/10.1016/j.rser.2015.05.003. 1043–1059. https://doi.org/10.35833/mpce.2020.000159.
Hirsch, A., Parag, Y., Guerrero, J., 2018. Microgrids: a review of technologies, key Yuan, Z., Xiong, G., Fu, X., 2022. Artificial neural network for fault diagnosis of solar
drivers, and outstanding issues. Renew. Sustain. Energy Rev. 90, 402–411. https:// photovoltaic systems: a survey. Energies 15. https://doi.org/10.3390/en15228693.
doi.org/10.1016/j.rser.2018.03.040. Yun, X., 2017. Photovolatic power generation system efficiency evaluation. N. China
Inthamoussou, F.A., De Battista, H., Mantz, R.J., 2012. New concept in maximum power Electr. Power Univ. https://doi.org/10.7666/d.Y3263630.
tracking for the control of a photovoltaic/hydrogen system. Int. J. Hydrogen Energy Zhou, N., Price, L., Dai, Y.D., Creyts, J., Khanna, N., Fridley, D., et al., 2019. A roadmap
37, 14951–14958. https://doi.org/10.1016/j.ijhydene.2012.01.176. for China to peak carbon dioxide emissions and achieve a 20 % share of non-fossil
Kermadi, M., Berkouk, E.M., 2017. Artificial intelligence-based maximum power point fuels in primary energy by 2030. Appl. Energy 239, 793–819. https://doi.org/
tracking controllers for Photovoltaic systems: comparative study. Renew. Sustain. 10.1016/j.apenergy.2019.01.154.
Energy Rev. 69, 369–386. https://doi.org/10.1016/j.rser.2016.11.125.
1924