WSEAS TRANSACTIONS on POWER SYSTEMS

DOI: 10.37394/232016.2023.18.8 Gil-Vera V. D., Quintero-López C.

Predictive Modeling of Photovoltaic Solar Power Generation

GIL-VERA V. D.
SISCO Research Group,
Luis Amigó Catholic University,
Trans. 51A N° 67 B-90, Medellín,
COLOMBIA

QUINTERO-LÓPEZ C.
NBA Research Group,
Luis Amigó Catholic University,
Trans. 51A N° 67 B-90, Medellín,
COLOMBIA

Abstract: - Photovoltaic solar power referred to as solar power using photovoltaic cells, is a renewable
energy source. The solar cells' electricity may be utilized to power buildings, neighborhoods, and even
entire cities. A stable and low-maintenance technology, photovoltaic solar power is an appealing
alternative for generating energy since it emits no greenhouse gases and has no moving components.
This paper aimed to provide a photovoltaic solar power generation forecasting model developed with
machine learning approaches and historical data. In conclusion, this type of predictive model enables
the evaluation of additional non-traditional sources of renewable energy, in this case, photovoltaic
solar power, which facilitates the planning process for the diversification of the energy matrix.
Random Forests obtain the highest performance, with this knowledge power systems operators may
forecast outcomes more precisely, this is the main contribution of this work.

Key-Words: Forecasting, Generation, Machine Learning, Predictive Modeling, Solar Power.

Received: July 16, 2022. Revised: March 11, 2023. Accepted: April 9, 2023. Published: May 3, 2023.

1 Introduction temperature, and wind speed) and the electrical

The utilization of renewable energy sources instead output of a solar system. Even though these models
of fossil fuels has been emphasized as a way to are initial approximations, [8], [9], they do not
reduce the carbon footprint globally. Global sufficiently take into consideration radiation's
population growth is closely related to rising energy changing and nonlinear character.
consumption and while old energy sources are Because meteorological data variations and
becoming depleted, new energy sources are being intermittencies affect energy output and the
investigated to fill the void, [1]. As a substitute performance index of solar systems, [10], solutions
treatment, more promotion and use of that foresee and assess these changes are necessary.
environmentally friendly energy sources are desired. Predictive models are one of these alternatives since
Solar photovoltaic (PV) energy is presented by they focus on using data processing and analysis to
this group. In 2010, solar energy generated less than find relationships, patterns, and/or trends. Models
1% of the world's electricity, but by 2022, that may be divided into two categories and applied to
percentage is projected to rise to over 28%, [2], the development of prediction models. In the first,
throughout the last three decades. The local climate machine learning (ML) and artificial intelligence
in the area where the system will be deployed has a (AI) are utilized, whereas classical statistics are used
strong correlation with the solar power-producing in the second, [11], [12].
capacity of PV systems, [3], [4]. Among the This work aimed to create a forecasting model
performance models used to estimate generation, the for electricity generation for the next few days. For
models of [5], [6], [7], stand out. 34 days in a row (68,788 entries), a database was
These models illustrate the direct link between utilized to collect data on energy production from a
environmental parameters (solar radiation, air solar photovoltaic power plant every 15 minutes.

E-ISSN: 2224-350X 71 Volume 18, 2023

 WSEAS TRANSACTIONS on POWER SYSTEMS
DOI: 10.37394/232016.2023.18.8 Gil-Vera V. D., Quintero-López C.

The database's variables included the date and time The requirement to satisfy a particular energy
of each observation, the amount of DC power the demand unites the many uses for stand-alone
inverter produced in 15 minutes (in kW), the systems. Because of this, almost all standalone
amount of AC power the inverter (source key) systems have energy storage technology, [20].
produced in 15 minutes (kW), the total amount of According to their related applications, these
energy produced throughout the day, and the total systems may be divided into three groups:
return on investment. professional, rural electrification, and modest
The other sections of this work are: section 2 consumption. Little photovoltaic modules,
provides a broad background on photovoltaic solar frequently built of amorphous silicon, are used in
power, section 3 offers the problem formulation, modest consumption applications to power
section 4 generalization of ML, section 5 problem electronic devices like calculators or watches,
solution, section 6 discussion, and finally mobile phone chargers, tiny power tools, household
conclusions. beacons, etc., [21], [22].
In conclusion, predictive models provide There are many professional applications,
improved network management, detect the need for including radio links, cathodic protection of gas
panel cleaning/maintenance, and locate faulty or pipelines, hotels, traffic signals, air navigation,
inefficient equipment. Predictive modeling may be refrigeration of vaccines, equipment for remote data
used to forecast the long-term power reserve for PV acquisition and transmission, and even power
system design and size as well as minimize supply for satellites and other space equipment,
generation uncertainty. [23].
Due to the extremely high costs associated with
power failures in all of these applications, it is
2 Photovoltaic Solar Power typically chosen to add solar generators and
A set of electrical and electronic components called electrochemical accumulators that are bigger than
a photovoltaic array uses sun radiation to generate technically necessary, [24], [25]. This reduces the
electricity. The photovoltaic module, which is made likelihood of failure.
up of cells capable of converting incident light Often included in development cooperation
energy into direct electrical energy, is the main projects and funded by nonprofit organizations or
component of this system, [13], [14]. institutions like the World Bank or the European
The rest of the equipment in a photovoltaic Union, rural electrification systems provide energy
system is primarily determined by the system's to rural communities that are located distant from
intended use. Grid-connected, off-grid, and pumped traditional power lines, [26], [27]. Solar home
storage systems are the three basic categories under systems (SHS), hybrid power plants, and pumped-
which photovoltaic systems may be categorized. storage systems are the most common types of rural
Grid-connected systems generate electricity that electrification systems. Lighting devices, radio,
is supplied to the traditional grid; they are exempt television, and small power tools may all be
from the requirement for energy storage as they are powered by home systems and hybrid plants,
not directly responsible for meeting customer respectively, [28], [29].
demand or ensuring consumption, [15], [16]. These Domestic systems with 100 W or 200 W power
systems, which may be separated into ground- ratings are often found in a family home, however,
mounted systems and building-mounted systems, occasionally they can also be found in community
comprise inverter equipment that adjusts the centers or medical facilities. A rural village's
electricity supplied by the solar generator to the electrical grid is provided by hybrid power plants
circumstances of the traditional grid to permit the with a solar generator, an electrochemical
proper linkage with the electrical grid, [17]. accumulator, and a generator set or wind turbine.
Above-ground systems often have more power The size of these plants depends on the population
than 100 kW and are entirely intended for energy they serve, with capacity ranging from 10 kW to
generation and related economic efficiency. In- 100 kW, [30], [31]. Pumping systems employ a
building systems perform tasks in addition to motor pump to raise and move water from an
generating energy, such as replacing architectural aquifer to a reservoir or distribution system by using
elements, creating aesthetically pleasing effects, the electrical energy generated by the solar
shading glass, etc. They generally have power generator, [32]. These systems often store energy in
ratings of less than 100 kW and are smaller than the form of potential energy from the water stored in
ground-mounted systems, [18], [19]. the raised reservoir to minimize costs and boost
dependability. Pumping systems can be used to

E-ISSN: 2224-350X 72 Volume 18, 2023

 WSEAS TRANSACTIONS on POWER SYSTEMS
DOI: 10.37394/232016.2023.18.8 Gil-Vera V. D., Quintero-López C.

desalinate water that has been extracted using

reverse osmosis systems, deliver water for human or
animal use, and irrigate private or public
plantations, [33], [34].
An electrical configuration designed to employ
photovoltaics to generate useable solar power is Cell Module Panel
known as a photovoltaic system. A photovoltaic cell
is a type of electrical device that directly transforms
light energy into electricity by harnessing the
physical and chemical phenomena known as the
photovoltaic effect, [35], [36]. Moreover, it is the
fundamental photovoltaic component that serves as
the foundation for solar modules.
When a substance is exposed to light, the Array – PV_System
photovoltaic effect occurs, which produces voltage Electricity Meter AC Isolator Fusebox
Inverter Battery Charge Controller
and electric current, [37]. Several solar cells linked Generation Meter DC Isolator Cabling
in series and/or parallel and enclosed in an Mounting Tracking System
ecologically friendly laminate make up a Fig. 1: From a solar cell to a PV system
photovoltaic module, [38]. A solar array's
fundamental building component is a photovoltaic
panel, a collection of modules, [39], [40]. A 3 Problem Formulation
collection of solar panels that together form the Throughout 34 days, generation data were acquired
entire photovoltaic-producing unit is called a at 15-minute intervals at the inverter level, where
photovoltaic array, [41], [58]. each inverter was connected to several lines of solar
Photovoltaic inverters convert DC electricity panels. Table 1 presents the variables that make up
from batteries or solar arrays to AC power for use the power plant database.
with standard utility-powered appliances. The
inverter acts as the brain of photovoltaic systems Table 1. Database description
since the solar array is a DC source and it takes one Variable Description
to convert DC electricity to the common AC power Each observation's date and time.
used in our homes and offices, [42]. It is crucial to Date_Time Observations were made and recorded every
15 minutes.
understand how weather conditions might affect the Plant_ID Plant identification.
output of the two solar power plants since Source_Key Inverter identification.
photovoltaic systems are heavily impacted by the Amount of DC power produced over 15
DC_Power
weather; in good weather, we receive the most yield, minutes by the inverter (source key) (kW).
while in bad weather, we get the least yield, [43], Amount of AC power produced over 15
AC_Power
minutes by the inverter (source key) (kW).
[44], [45]. The transition from a solar cell to a
The total amount of energy produced in a
photovoltaic system is shown in Figure 1. Daily_Yield
day.
Total_Yield Total investor return.

In the following GitHub link are available the

generation database of the PV plant was analyzed
for 34 consecutive days: https://acortar.link/w8yUqp
Sunlight is the cause of the Plant's Direct Current
(DC) power production between 05:33:20 and
18:00:00, but else there is none. There are 22
inverters in the facility, each linked to several PV
arrays. Each inverter captures its data every 15
minutes. Hence, to determine how much electricity
the plant produced in an hour, we just compute the
contribution of the 22 inverters. There are 22
inverters for data time on May 15, 2020, at 0:00.
Except for the curve of May 20 and 25, which
provides a consistent shape, nearly all the curves are
the same despite some variation between 11 am and

E-ISSN: 2224-350X 73 Volume 18, 2023

 WSEAS TRANSACTIONS on POWER SYSTEMS
DOI: 10.37394/232016.2023.18.8 Gil-Vera V. D., Quintero-López C.

2 pm. DC electricity is only at its peak on 2020-05- second stage is the data input, and the third stage is
25. Data provides us with a logistics-like function, the handling of the discrepancy between the
but after 18:00 the energy progressively declines estimated and measured data, [49]. By taking into
until breaking down completely at 00:00. As you account the fact that the deviation and forecasting
can see, certain daily yield dates (2020-02-06, 2020- abilities of the models depend not only on the
05-19, etc.) have a logistic shape with missing climatic conditions but also on the prediction
values, but not others (2020-02-06, 2020-05-19, horizon, criteria like the treatment of non-linearity,
etc.). Data are logged every 15 minutes, and then we the behavior when using multiple inputs, the
receive a fresh yield. Figure 2, presents the DC prediction horizon, the treatment of the deviation
Power Plot. associated with the prediction, and flexibility,
provide guidelines in model development, [50].
300000 Next, we provide a summary of the ML models
250000 examined in this work:
200000
4.1 Naïve-Bayes
150000 This probabilistic ML method is frequently
employed for classification problems. It is founded
100000
on the Bayes theorem, which calculates an event
50000 probability using information about prior
confounding variables. This model assumes that the
0
00:00 05:33 11:06 16:40 22:13 features used to categorize instances are
Fig. 2: DC Power Vs Time independent of one another in the context of
classification, i.e., one feature’s existence or
Figure 3 presents the Daily DC Power on each absence does not affect the presence or absence of
day of the considered period (34 days). any other feature, [51]. This model is based on the
Bayes Theorem.
𝑃(𝐵|𝐴). 𝑃(𝐴)
P(A|B) = (1)
𝑃(𝐵)

Where P(A) and P(B) are the odds of seeing A

and B in the absence of any provided circumstances.
P(A│B) is the probability that event A occurs given
that B is true, and P(B│A) is the probability of the
contrary case. A and B are events, and P(B) is
different from zero.

4.2 Artificial Neural Network

Fig. 3: Daily DC Power ANN was developed based on how the human brain
functions. Neurons, the linked layers of nodes that
make up ANNs, process and transfer information
4 Machine Learning via mathematical operations. A neuron is the
This computer science area is characterized by an fundamental unit of an ANN. It receives input from
artificial intelligence (AI) method, which is applied other neurons or external sources, computes an
in many different disciplines of study, including output using the weighted sum of the inputs, and
biology, economics, and the energy sector, [46]. ML then applies a nonlinear activation function. Using
enables the creation of models that can make methods like backpropagation, the weights of the
judgments that are challenging for explicit methods, connections between neurons are learned using
such as straightforward numerical and analytical training data, [52]. Equation (2), presents the
techniques, to describe, [47]. In [48], the authors general equation of this model.
𝑛
assert that if representation is feasible, ML models
can identify correlations between predictor variables 𝑜𝑢𝑡𝑝𝑢𝑡 = a(𝑧) = a [(∑ 𝑥𝑖. 𝑤𝑖 ) + 𝑏 ] (2)
and target variables. i=1
Three steps make up ML: the first stage is the
pre-processing and categorization of the data, the

E-ISSN: 2224-350X 74 Volume 18, 2023

 WSEAS TRANSACTIONS on POWER SYSTEMS
DOI: 10.37394/232016.2023.18.8 Gil-Vera V. D., Quintero-López C.

4.3 Support Vector Machine trained using many techniques, including ID3, C4.5,
This model is employed for classification and CART, and Random Forests, [55]. The objective is
regression problems. When dealing with high- to reduce the impurity or entropy of the subsets,
dimensional datasets and non-linear decision which quantifies the level of homogeneity of the
boundaries, SVMs are especially successful. class labels or values, [55]. In this model, entropy is
Finding the hyperplane that optimally separates the a measure of the randomness in the information
various classes in the input data is fundamental in being processed and information gain (IG) is a
this model. When there are more than two classes, decrease in entropy (equation 5).
the hyperplane may be a plane or a higher-
k
dimensional manifold, [53]. In a two-class
classification issue, the hyperplane is a line that IG = Entropy(𝑏𝑒𝑓𝑜𝑟𝑒) − ∑ Entropy(𝑗, 𝑎𝑓𝑡𝑒𝑟)
(5)
j=1
divides the two classes. Equation (3) presents a
feature vector.
4.6 Random Forest
This model is used for supervised learning tasks
𝑔(𝑥) = 𝑤 𝑡 𝑥 + 𝑏 (3) including classification, regression, and others. An
extensive number of decision trees are constructed
For a binary classification problem 𝐶1 , 𝐶2 , if during training using this ensemble learning
𝑔(𝑥) > 0 then x ∈ 𝐶1 , else x ∈ 𝐶2 . approach, which results in a class that reflects the
average of the predictions (regression) or
classifications produced by the individual trees.
4.4 Logistic Regression
This technique works by building a collection of
This model is a typical statistical learning approach
decision trees, each of which is trained using a
for binary classification problems, where the
portion of the input features and training data that is
objective is to predict the likelihood that an event
randomly chosen. The random forest aggregates all
will belong to one of two classes. A probability
of the individual trees’ predictions throughout the
score may be the threshold for a binary logistic
prediction phase to provide a final prediction, [56].
regression prediction result. The logistic function,
In this model, the Gini Index (equation 6) is used to
on which this model is built, converts an output
identify how much impurity has a particular node.
resulting from a linear combination of input data
into a probability score between 0 and 1. 𝑐
Weights are used to model the linear
combination of input characteristics, and they are 𝐺𝑖𝑛𝑖 𝐼𝑛𝑑𝑒𝑥 = 1 − ∑(𝑝𝑗 )2 (6)
learned from training data via maximum likelihood 𝑗=1
estimation, [54]. Equation (4) presents the general
equation of this model, where µ is the midpoint of Where 𝑝𝑗 is the proportion of samples belonging to
the curve and s is a scale parameter. class c for a given node.

1 4.7 K-Nearest Neighbors

p(𝑥) = (4) In supervised learning, this model is used for
1+ 𝑒 −(𝑥−µ )/𝑠
classification and regression applications. It is a
4.5 Decision Tree non-parametric approach because it makes no
assumptions about the distribution of the
This model is employed for classification and
underlying data. Based on a selected distance
regression problems. Each leaf node represents a
class label or a numeric value, and each interior metric, such as Euclidean distance, the K-NN
node represents a decision based on a feature or method finds the K data points in the training set
that is closest to a given data point. The majority
attribute, [55]. Recursively dividing the input space
into subsets based on the values of the input class or mean value of the K nearest neighbors in
the training set is then used to forecast the output of
characteristics is how decision trees are built, the
data is divided into two or more subsets via top- the supplied data point, [57], [58]. The distance
down partitioning, where the most informative functions used in this model can be Euclidean
(equation 7), Manhattan (equation 8), or
feature is chosen at each internal node, [55].
Minkowski (equation 9).
A decision tree may be easily converted into a
collection of rules by mapping from the root node to
the leaf nodes one by one. This model may be

E-ISSN: 2224-350X 75 Volume 18, 2023

 WSEAS TRANSACTIONS on POWER SYSTEMS
DOI: 10.37394/232016.2023.18.8 Gil-Vera V. D., Quintero-López C.

𝑛
radiation that will arrive at a certain location at a
specific time using meteorological and satellite data.
𝑑 = √∑(𝑥𝑖 − 𝑦𝑖 )2 (7) Based on the amount of sunshine and the present
𝑖=1 weather, real-time PV monitors may also be used to
measure current generations and forecast future
𝑛 generations. Another approach is to estimate using
previous PV generation data from the same site and
𝑑 = ∑ | 𝑥𝑖 − 𝑦𝑖 | (8)
season. In this work, the latter is utilized. After
𝑖=1 cross-validation. Table 2 lists the optimal
parameters for each model examined.
𝑛 1/𝑞

𝑑 = (∑(|𝑥𝑖 − 𝑦𝑖 |)𝑞 ) (9) Table 2. Optimal libraries and parameters

𝑖=1 Model Library & Optimal Parameters
Random
Two current strategies utilized in photovoltaic Forest
RandomForestClassifier: {'n_estimators': 49}
solar power generation forecasting models are deep
DecisionTreeClassifier:
learning and statistical approaches. The authors of Decision {'criterion':'gini','class_weight':
[59], propose a hybrid model that blends machine- Tree 'balanced', 'max_depth': 5, 'max_features':
learning approaches with the Theta statistical 'log2, 'splitter': 'best'}
method to more accurately anticipate future solar MLPRegressor: {'activation': 'relu',
power output from renewable energy facilities. ANN- 'hidden_layer_sizes': 4, 'learning_rate':
Long short-term memory (LSTM), gate recurrent MLP 'constant', 'solver': 'adam', 'learning_rate_init':
unit (GRU), AutoEncoder LSTM (Auto-LSTM), 0.5}
and a recently suggested Auto-GRU are among the K-NN KNeighborsClassifier: {'n_neighbors':6}
ML models. Naïve GaussianNB: {'max_features': 'auto',
In [60], the authors suggest a fast-track Bayes 'var_smoothing':1e-8}
methodology to handle two essential issues: long- Logistic LogisticRegression: {'C': 15, 'max_iter':6800,
term solar resource assessment and photovoltaic Regression 'penalty': 'l2', 'tol': 1e-7}
energy forecasting when investigating prospective SVM SVC: {'C': 88, 'kernel': 'RBF', 'tol': 0.001}
locations for PV plant construction. These writers
employed data clustering and probability techniques
while exploring potential sites for PV plant Table 3 displays the findings for each of the
construction. examined accuracy measures.
In [61], the authors used numerous time-series
algorithms to predict PV power generation output to Table 3. Training results
respond swiftly to equipment and panel problems.
Specificity*
Accuracy*

Precision*

F1-Score*
Recall*

The Long Short-Term Memory (LSTM) model Model

exhibited the lowest error rate when compared to
other models for quick PV power generation
estimates, according to the study's findings. Random-Forest .843 .848 .875 .793 .854
Decision Tree .837 .827 .869 .773 .843
ANN-MLP .707 .787 .861 .798 .882
SVM .780 .765 .844 .690 .703
5 Problem Solution K-NN .697 .675 .785 .566 .722
Many variables, including geographic location, Naïve Bayes .588 .537 .622 .476 .672
sunshine intensity, solar panel efficiency, and Logistic Regression .543 .532 .521 .432 .563
meteorological conditions, can have an impact on
the production of photovoltaic energy. However, This finding enables us to determine that
even within the same place, it might change from Random Forest (Accuracy=.843, F1-Score=.854)
day to day. was the model that performed the best. Decision
There are, however, a few approaches to foresee Trees (Accuracy=.837, F1-Score=.843) and ANN-
PV power generation at a certain site and time. One MLP (Accuracy=.707, F1-Score=.882) were two
method is to employ solar radiation prediction more models that did well. The accuracy of the
models, which calculate the quantity of solar positive predictions made by the ANN-MLP is
lower than that of the first two models.

E-ISSN: 2224-350X 76 Volume 18, 2023

 WSEAS TRANSACTIONS on POWER SYSTEMS
DOI: 10.37394/232016.2023.18.8 Gil-Vera V. D., Quintero-López C.

The models with the lowest ability to recognize F1-Score=.854); with this knowledge, power
negative instances (Specificity) were Naive Bayes systems operators may forecast outcomes with more
and Logistic Regression; for both of these models, precision.
this measure was below .50, making them This ML model is a robust algorithm that can
ineffective prediction models. The Logistic handle a large number of features, missing values,
Regression model has the lowest efficiency, with an and noisy data. It is less prone to overfitting than
Accuracy of .543. other algorithms, such as decision trees, thanks to
the use of bagging and feature randomness.
Random Forests is known for its high accuracy in
6 Discussion predicting outcomes, making it a popular choice for
Solar power projections will have a huge impact on many applications.
the future of large-scale renewable energy This ML model is adaptable and may be used for
installations. Predicting solar electricity generation both classification and regression problems. It can
is strongly reliant on changing weather patterns. offer crucial insights into the link between the
Beyond climatic and altitude factors, the total characteristics and the outcome, increasing the
amount of energy produced by a solar station model's interpretability. Overall, ML integration can
depends on its capacity; thus, the total amount of result in improved efficiency, reliability, and cost
energy produced by each solar power plant depends reductions in the photovoltaic power generation
on its capacity. sector.
A solar plant with a higher kilowatt peak (kWp)
capacity will produce more energy than a plant with
a lower kWp capacity. Because the variables that References:
affect PV power generation are unpredictable, it can [1] El-Hadary, M. I., Senthilraja, S., & Zayed, M.
be difficult to predict it with any degree of accuracy. E., “A hybrid system coupling spiral type
With the use of machine learning (ML), it is now solar photovoltaic thermal collector and
feasible to forecast power generation for the electrocatalytic hydrogen production cell:
upcoming few days to manage the grid better, Experimental investigation and numerical
recognize the need for panel cleaning and modeling,” Process Safety and Environmental
maintenance, and spot broken or underperforming Protection, vol. 170, p. 1101–1120, 2023.
machinery. DOI: 10.1016/j.psep.2022.12.079.
Future studies might concentrate on estimating [2] Mamatha, G., & Kulkarni, P. S., “Assessment
the yearly solar energy that a solar power plant is of floating solar photovoltaic potential in
anticipated to yield based on site attributes and local India’s existing hydropower reservoirs,”
meteorological data. Other environmental variables Energy for Sustainable Development, vol. 69,
that may impact daily solar power output, in p. 64–76, 2022. DOI:
addition to the height of a solar power station's 10.1016/j.esd.2022.05.011.
location, include temperature, wind speed, vapor [3] Shi, M., Lu, X., Jiang, H., Mu, Q., Chen, S.,
pressure, solar radiation, day length, precipitation, Fleming, R.M., Zhang, N., Wu, Y., & Foley,
and snowfall. A.M., “Opportunity of rooftop solar
photovoltaic as a cost-effective and
environment-friendly power source in
megacities,” iScience, vol. 25, No. 9, p.
7 Conclusion 104890, 2022. DOI:
Forecasting photovoltaic power output is crucial for
10.1016/j.isci.2022.104890.
grid planning and management because it enables
[4] Yang, H., & Wang, H., “Numerical simulation
operators to better plan and manage energy supply
of the dust particles deposition on solar
and demand, which leads to lower costs and higher
photovoltaic panels and its effect on power
system efficiency. Due to weather fluctuation, the
generation efficiency,” Renew Energy, vol.
difficulty of detecting solar radiation, system
201, p. 1111–1126, 2022. DOI:
capacity uncertainty, and a lack of historical data,
10.1016/j.renene.2022.11.043.
this task may be challenging. It helps to increase
[5] Veeramanikandan, M., Arjunan, T. V., &
prediction accuracy to utilize sensors and modeling.
Jidhesh, P., “Experimental investigation of
The Random Forests method demonstrated the
sandwich glazed solar photovoltaic module,”
highest performance (Accuracy=.843,
Mater Today Proc, vol. 27, p. 136–139, 2020.
Precision=.848, Recall=.875, Specificity=.793, and
DOI: 10.1016/j.matpr.2019.09.069.

E-ISSN: 2224-350X 77 Volume 18, 2023

 WSEAS TRANSACTIONS on POWER SYSTEMS
DOI: 10.37394/232016.2023.18.8 Gil-Vera V. D., Quintero-López C.

[6] Lonergan, K.E., & Sansavini, G., “Business 52, p. 102168, 2022. DOI:
structure of electricity distribution system 10.1016/j.seta.2022.102168.
operator and effect on solar photovoltaic [14] Al-Ghussain, L., Taylan, O., Abujubbeh, M.,
uptake: An empirical case study for & Hassan, M.A., “Optimizing the orientation
Switzerland,” Energy Policy, vol. 160, p. of solar photovoltaic systems considering the
112683, 2022. DOI: effects of irradiation and cell temperature
10.1016/j.enpol.2021.112683. models with dust accumulation,” Solar
[7] Kumar, N., & Pal, N., “Location and Energy, vol. 249, p. 67–80, 2023. DOI:
orientation-based performance analysis of 10.1016/j.solener.2022.11.029.
4.98 kW solar photovoltaic system for [15] Nej, S.K., Sreejith, S., & Chakraborty, I.,
isolated Indian islands,” Sustainable Energy “Dual-Output Multistage Switched-Capacitor
Technologies and Assessments, vol. 52, p. Quadratic Boost (MSC-QBC) DC-DC
102138, 2022. DOI: Converter for Solar Photovoltaic
10.1016/j.seta.2022.102138. Application,” IFAC-PapersOnLine, vol. 55,
[8] Wang, J., Chen, H., Deng, H., & Dong, F., “A No. 1, p. 965–970, 2022. DOI:
combined power and steam system integrated 10.1016/j.ifacol.2022.04.159.
with solar photovoltaic/thermal collector: [16] He, B., Lu, H., Zheng, C., & Wang, Y.,
Thermodynamic characteristics and cost- “Characteristics and cleaning methods of dust
benefit analyses,” Case Studies in Thermal deposition on solar photovoltaic modules-A
Engineering, vol. 39, p. 102477, 2022. DOI: review,” Energy, vol. 263, p. 126083, 2023.
10.1016/j.csite.2022.102477. DOI: 10.1016/j.energy.2022.126083.
[9] Chirwa, D., Goyal, R., & Mulenga, E., [17] Xu, Z., Kong, Q., Qu, H., & Wang, C.,
“Floating solar photovoltaic (FSPV) potential “Cooling characteristics of solar photovoltaic
in Zambia: Case studies on six hydropower panels based on phase change materials,”
power plant reservoirs,” Renewable Energy Case Studies in Thermal Engineering, vol. 41,
Focus, vol. 44, p. 344–356, 2023. DOI: p. 102667, 2023. DOI:
10.1016/j.ref.2023.01.007. 10.1016/j.csite.2022.102667.
[10] Özçelep, Y., Bekdaş, G., & Apak, S., [18] Macías, R.J., Ceballos, C., Ordoñez-Loza, J.,
“Meeting the electricity demand for the & Ortiz, M., “Evaluation of the performance
heating of greenhouses with hydrogen: Solar of a solar photovoltaic - Biomass gasifier
photovoltaic-hydrogen-heat pump system system as an electricity supplier,” Energy, vol.
application in Turkey,” Int J Hydrogen 260, p. 125046, 2022. DOI:
Energy, vol. 48, No. 7, p. 2510–2517, 2023. 10.1016/j.energy.2022.125046.
DOI: 10.1016/j.ijhydene.2022.10.125. [19] Dhass, A.D., Patel, D., & Patel, B.,
[11] Hasheem, M.J., Wang, S., Ye, N., Farooq, “Estimation of power losses in single-junction
M.Z., & Shahid, H. M., “Factors influencing gallium-arsenide solar photovoltaic cells,”
purchase intention of solar photovoltaic International Journal of Thermofluids, vol. 17,
technology: An extended perspective of p. 100303, 2023. DOI:
technology readiness index and theory of 10.1016/j.ijft.2023.100303.
planned behavior,” Cleaner and Responsible [20] Kulikov, A.L., Shepovalova, O.V., Ilyushin,
Consumption, vol. 7, p. 100079, 2022. DOI: P.V., Filippov, S.P., & Chirkov, S.V.,
10.1016/j.clrc.2022.100079. “Control of electric power quality indicators
[12] Pimpalkar, R., Sahu, A., Patil, R. B., & Roy, in distribution networks comprising a high
A., “A comprehensive review on failure share of solar photovoltaic and wind power
modes and effect analysis of the solar stations,” Energy Reports, vol. 8, p. 1501–
photovoltaic system,” Mater Today Proc, 1514, 2022. DOI: 10.1016/j.egyr.2022.08.217.
2022. DOI: 10.1016/j.matpr.2022.11.353. [21] Gómez-Calvet, R., Martínez-Duart, J.M., &
[13] Manoj Kumar, N., Chakraborty, S., Kumar Gómez-Calvet, A.R., “The 2030 power sector
Yadav, S., Singh, J., & Chopra, S.S., transition in Spain: Too little storage for so
“Advancing simulation tools specific to many planned solar photovoltaics?”
floating solar photovoltaic systems – Renewable and Sustainable Energy Reviews,
Comparative analysis of field-measured and vol. 174, p. 113094, 2023. DOI:
simulated energy performance,” Sustainable 10.1016/j.rser.2022.113094.
Energy Technologies and Assessments, vol. [22] Mohammed, C., Mohamed, M., Larbi, E.M.,
Manale, B., Hassan, Z., Jalal, B., & Smail, Z.,

E-ISSN: 2224-350X 78 Volume 18, 2023

 WSEAS TRANSACTIONS on POWER SYSTEMS
DOI: 10.37394/232016.2023.18.8 Gil-Vera V. D., Quintero-López C.

“Extended method for the sizing, energy scarce,” The Electricity Journal, vol. 35, No.
management, and techno-economic 10, p. 107222, 2022. DOI:
optimization of autonomous solar 10.1016/j.tej.2022.107222
Photovoltaic/Battery systems: Experimental [30] Luan, R., & Lin, B., “Positive or negative?
validation and analysis,” Energy Convers Study on the impact of government subsidy
Manag, vol. 270, p. 116267, 2022. DOI: on the business performance of China’s solar
10.1016/j.enconman.2022.116267. photovoltaic industry,” Renew Energy, vol.
[23] Nain, P., & Kumar, A., “A state-of-art review 189, p. 1145–1153, 2022. DOI:
on end-of-life solar photovoltaics,” J Clean 10.1016/j.renene.2022.03.082.
Prod, vol. 343, p. 130978, 2022. DOI: [31] Le, M., Luong, V.S., Nguyen, D.K., Dao,
10.1016/j.jclepro.2022.130978. V.D., Vu, N.H., & Vu, H.H.T., “Remote
[24] Kinsey, G.S., Riedel-Lyngskær, N.C., Miguel, anomaly detection and classification of solar
A.A., Boyd, M., Braga, M., Shou, Ch., photovoltaic modules based on deep neural
Cordero, R.R., Duck, B.C., Fell, Ch. J., Feron, network,” Sustainable Energy Technologies
S., Georghiou, G.E., Habryl, N., John, J.J., and Assessments, vol. 48, p. 101545, 2021.
Ketjoy, N., López, G., Louwen, A., Maweza, DOI: 10.1016/j.seta.2021.101545.
E.L., Minemoto, T., Mittal, A., Molto, C., [32] Oteng, D., Zuo, J., & Sharifi, E., “An expert-
Neves, G., Garrido, G.N., Norton, M., based evaluation on end-of-life solar
Paudyal, B.R., Pereira, E.B., Poissant, Y., photovoltaic management: An application of
Pratt, L., Shen, Q., Reindl, T., Rennhofer, M., Fuzzy Delphi Technique,” Sustainable
Rodríguez-Gallegos, C.D., Rüther, R., van Horizons, vol. 4, p. 100036, 2022. DOI:
Sark, W., Sevillano-Bendezú, M.A., 10.1016/j.horiz.2022.100036.
Seigneur, H., Tejero, J.A., Theristis, M., [33] Singh, S.K., & Chander, N., “Mid-life
Töfflinger, J.A., Ulbrich, C., Vilela, W.A., degradation evaluation of polycrystalline Si
Xia, X., & Yamasoe, M.A., “Impact of solar photovoltaic modules in a 100 kW grid-
measured spectrum variation on solar tied system in east-central India,” Renew
photovoltaic efficiencies worldwide,” Renew Energy, vol. 199, p. 351–367, 2022. DOI:
Energy, vol. 196, p. 995–1016, 2022. DOI: 10.1016/j.renene.2022.09.013.
10.1016/j.renene.2022.07.011. [34] Ilyushin, P.V., Shepovalova, O.V., Filippov,
[25] Liu, D., Qi, S., & Xu, T., “Visual observation S.P., & Nekrasov, A.A., “The effect of
or oral communication? The effect of social complex load on the reliable operation of
learning on solar photovoltaic adoption solar photovoltaic and wind power stations
intention in rural China,” Energy Res Soc Sci, integrated into energy systems and into off-
vol. 97, p. 102950, 2023. DOI: grid energy areas,” Energy Reports, vol. 8, p.
10.1016/j.erss.2023.102950. 1515–1529, 2022. DOI:
[26] Chandel, R., Chandel, S.S., & Malik, P., 10.1016/j.egyr.2022.08.218.
“Perspective of new distributed grid [35] Choi, J., Lee, I.W., & Cha, S.W., “Analysis of
connected rooftop solar photovoltaic power data errors in the solar photovoltaic
generation policy interventions in India,” monitoring system database: An overview of
Energy Policy, vol. 168, p. 113122, 2022. nationwide power plants in Korea,”
DOI: 10.1016/j.enpol.2022.113122. Renewable and Sustainable Energy Reviews,
[27] Prasad, M., & Prasad, R., “Bifacial vs vol. 156, p. 112007, 2022. DOI:
monofacial grid-connected solar photovoltaic 10.1016/j.rser.2021.112007.
for small islands: A case study of Fiji,” [36] Majji, R.K., Mishra, J.P., & Dongre, A.A.,
Renew Energy, vol. 203, p. 686–702, 2023. “Model predictive control based autonomous
DOI: 10.1016/j.renene.2022.12.068. DC microgrid integrated with solar
[28] Ndzibah, E., Pinilla-De La Cruz, G.A., & photovoltaic system and composite energy
Shamsuzzoha, A., “Collaboration towards storage,” Sustainable Energy Technologies
value creation for end-of-life solar and Assessments, vol. 54, p. 102862, 2022.
photovoltaic panel in Ghana,” J Clean Prod, DOI: 10.1016/j.seta.2022.102862.
vol. 333, p. 129969, 2022. DOI: [37] Sadanand, S., & Dwivedi, D.K., “Numerical
10.1016/j.jclepro.2021.129969. modeling for earth-abundant highly efficient
[29] Boretti, A., & Castelletto, S., “Lacking energy solar photovoltaic cell of non-toxic buffer
storage, and nuclear contribution, wind, and layer,” Opt Mater (Amst), vol. 109, p.
solar photovoltaic electricity is expensive and

E-ISSN: 2224-350X 79 Volume 18, 2023

 WSEAS TRANSACTIONS on POWER SYSTEMS
DOI: 10.37394/232016.2023.18.8 Gil-Vera V. D., Quintero-López C.

110409, 2020. DOI: vol. 8, p. 15249–15285, 2022. DOI:

10.1016/j.optmat.2020.110409. 10.1016/j.egyr.2022.11.092.
[38] Gholami, A., Ameri, M., Zandi, M., [46] Lin, B., & Shi, L., “New understanding of
Ghoachani, R.G., Gerashi, S. J., Kazem, H.A., power generation structure transformation,
& Al-Waeli, A. H. A., “Impact of harsh based on a machine learning predictive
weather conditions on solar photovoltaic cell model,” Sustain. Energy Technol.
temperature: Experimental analysis and Assessments, vol. 51, p. 101962, 2022. DOI:
thermal-optical modeling,” Solar Energy, vol. 10.1016/j.seta.2022.101962
252, p. 176–194, 2023. DOI: [47] Zhang, Y., Liu, K., Qin, L., & An, X.,
10.1016/j.solener.2023.01.039. “Deterministic and probabilistic interval
[39] Chindamani, M., & Ravichandran, C.S., “A prediction for short-term wind power
hybrid DDAO-RBFNN strategy for fault generation based on variational mode
tolerant operation in fifteen-level cascaded H- decomposition and machine learning
bridge (15L-CHB) inverter with solar methods,” Energy Convers. Manag., vol. 112,
photovoltaic (SPV) system,” Solar Energy, p. 208–219, 2016. DOI:
vol. 244, p. 1–18, 2022. DOI: 10.1016/j.enconman.2016.01.023
10.1016/j.solener.2022.08.015. [48] Sharma, H., Marinovici, L., Adetola, V., &
[40] Kërçi, T., Tzounas, G., & Milano, F., “A Schaef, H.T., “Data-driven modeling of power
dynamic behavioral model of the long-term generation for a coal power plant under
development of solar photovoltaic generation cycling,” Energy AI, vol. 11, p. 100214, 2023.
driven by feed-in tariffs,” Energy, vol. 256, p. DOI: 10.1016/j.egyai.2022.100214
124506, 2022. DOI: [49] Chen, H., Birkelund, Y., & Yuan, F.,
10.1016/j.energy.2022.124506. “Examination of turbulence impacts on ultra-
[41] Kamil, K., Chong, K. H., & Hashim, H., short-term wind power and speed forecasts
“Excess power rerouting in the grid system with machine learning,” Energy Reports, vol.
during high penetration solar photovoltaic,” 7, p. 332–338, 2021. DOI:
Electric Power Systems Research, vol. 214, p. 10.1016/j.egyr.2021.08.040
108871, 2023. DOI: [50] Chen, Z., Xiao, F., Guo, F., & Yan, J.,
10.1016/j.epsr.2022.108871. “Interpretable machine learning for building
[42] Gassar, A.A.A., & Cha, S.H., “Review of energy management: A state-of-the-art
geographic information systems-based rooftop review,” Adv. Appl. Energy, vol. 9, p.
solar photovoltaic potential estimation 100123, 2023. DOI:
approaches at urban scales,” Appl Energy, 10.1016/j.adapen.2023.100123
vol. 291, p. 116817, 2021. DOI: [51] Li, L., Zhou, Z., Bai, N., Wang, T., Xue,
10.1016/j.apenergy.2021.116817. K.H., Sun, H., He, Q., Cheng, W., & Miao,
[43] Mia, S., Kumer Podder, A., Manoj Kumar, N., X., “Naive Bayes classifier based on
Bhatt, A., & Kumar, K., “Experimental memristor nonlinear conductance,”
verification of a dynamic programming and Microelectronics J., vol. 129, p. 105574,
IoT-based simultaneous load-sharing 2022. DOI: 10.1016/j.mejo.2022.105574
controller for residential homes powered with [52] Monteiro, R.V.A., Guimarães, G.C., Moura,
grid and onsite solar photovoltaic electricity,” F.A.M., Albertini, M.R.M.C., & Albertini,
Sustainable Energy Technologies and M.K., “Estimating photovoltaic power
Assessments, vol. 55, p. 102964, 2023. DOI: generation: Performance analysis of artificial
10.1016/j.seta.2022.102964. neural networks, Support Vector Machine and
[44] Oteng, D., Zuo, J., & Sharifi, E., “A Kalman filter,” Electr. Power Syst. Res., vol.
scientometric review of trends in solar 143, p. 643–656, 2017. DOI:
photovoltaic waste management research,” 10.1016/j.epsr.2016.10.050
Solar Energy, vol. 224, p. 545–562, 2021. [53] Roy, A., & Chakraborty, S., “Support vector
DOI: 10.1016/j.solener.2021.06.036. machine in structural reliability analysis: A
[45] Premkumar, M., Jangir, P., Ramakrishnan, C., review,” Reliab. Eng. Syst. Saf., vol. 233, p.
Kumar, C., Sowmya, R., Deb, S., & Kumar, 109126, 2023. DOI:
N.M., “An enhanced Gradient-based 10.1016/j.ress.2023.109126
Optimizer for parameter estimation of various [54] Maresch, K., Marchesan, G., & Cardoso, G.,
solar photovoltaic models,” Energy Reports, “A logistic regression approach for improved
safety of the under-frequency load shedding

E-ISSN: 2224-350X 80 Volume 18, 2023

 WSEAS TRANSACTIONS on POWER SYSTEMS
DOI: 10.37394/232016.2023.18.8 Gil-Vera V. D., Quintero-López C.

scheme owing to feeder machine inertia,” Contribution of Individual Authors to the

Electr. Power Syst. Res., vol. 218, p. 109189, Creation of a Scientific Article (Ghostwriting
2023. DOI: 10.1016/j.epsr.2023.109189 Policy)
[55] Asvapoositkul, S., & Preece, R., “Decision -Victor Daniel Gil Vera has performed the literature
tree-based prediction model for small signal review, the normalization of the database, training
stability and generation-rescheduling the predictive models in Python, and the statistical
preventive control,” Electr. Power Syst. Res., analysis.
vol. 196, p. 107200, 2021. DOI: -Catalina Quintero López has performed the
10.1016/j.epsr.2021.107200 literature review and the analysis of the results.
[56] Liu, D., & Sun, K., “Random Forest solar
power forecast based on classification Sources of Funding for Research Presented in a
optimization,” Energy, vol. 187, p. 115940, Scientific Article or Scientific Article Itself
2019. DOI: 10.1016/j.energy.2019.115940 This research was funded by the Luis Amigó
[57] Amonkar, Y., Farnham, D.J., & Lall, U., “A Catholic University and was one of the results of the
k-nearest neighbor space-time simulator with research project entitled "Implementation of Smart
applications to large-scale wind and solar Grids in Colombia: a multidimensional analysis" -
power modeling,” Patterns, vol. 3, No. 3, p. Cost Center [0502020950].
100454, 2022. DOI:
10.1016/j.patter.2022.100454 Conflict of Interest
[58] Abushgair, K., "Enhancement of Poly-Crystal The authors have no conflict of interest to declare.
PV Panels Performance by Air-to-Air Heat
Exchanger Cooling System", WSEAS Creative Commons Attribution License 4.0
Transactions on Power Systems, vol. 16, pp. (Attribution 4.0 International, CC BY 4.0)
157-163, 2021. DOI: This article is published under the terms of the
10.37394/232016.2021.16.16 Creative Commons Attribution License 4.0
[59] AlKandari, M., & Ahmad, I., "Solar power https://creativecommons.org/licenses/by/4.0/deed.en
generation forecasting using ensemble _US
approach based on deep learning and
statistical methods", Applied Computing and
Informatics, vol. ahead of print, pp. 1-20,
2020. DOI: 10.1016/j.aci.2019.11.002
[60] Borunda, M., Ramírez, A., Garduno, R., Ruíz,
G., Hernandez, S., Jaramillo, O.A.,
“Photovoltaic Power Generation Forecasting
for Regional Assessment Using Machine
Learning”, Energies, vol. 15, No. 8895, pp. 1-
25, 2022. DOI:10.3390/en15238895
[61] Kim, E., Akhtar, M. S., & Yang, O. B.,
“Designing solar power generation output
forecasting methods using time series
algorithms”, Electric Power Systems
Research, vol. 216, pp. 1-12, 2023. DOI:
10.1016/j.epsr.2022.109073.

E-ISSN: 2224-350X 81 Volume 18, 2023