NTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING

DOI: 10.46300/9106.2020.14.88 Volume 14, 2020

An Efficient Intelligent Power Detection

Method for Photovoltaic System
Ayman M Mansour, Jalal Abdallah, and Mohammad A Obeidat
Tafila Technical University,
‘IS, Tafila, 66110,
Jordan

complexity to power systems and makes them more challenging

Abstract— Jordan has experienced a significant increase to the operator. The renewable power generation sources are
in both peak load and annual electricity demand within the divided into two categories: the sources, which have similar
last decade due to the growth of the economy and characteristics to conventional power generation facilities in
population. Photovoltaic (PV) system is one of the most that they are predictable and controllable, such as hydroelectric
popular renewable energy source in Jordan. PV system is generation and biomass; and the variable and intermittent
highly nonlinear with unpredictable behavior since it is sources, such as wind and solar [2].
always subject to many external factors such as severe The effect of wind and solar renewable sources on power
weather conditions, irradiance level, sheds, temperature, system stability is different from that of the conventional power
etc. This makes it difficult to maintain maximum power sources. Wind and solar are depending on variable conditions
production around its operation ranges.
such as weather conditions, as wind speed and solar irradiance,
In this paper, an intelligent technique is used to predict site dependence and types of generators used [3]. Such weather
and identify the working ability of the PV system under
poses many challenges for system operators to ensure grid
different weather factors in Tafila Technical University
stability and reliability.
(TTU) in Jordan. It helps in optimizing power productions
Jordan experienced a significant increase in both peak load
for different operation points. The PV system in Tafila with
size 1 MWp PV generated 5.4 GWh since 2017. It saves and annual electricity demand within the last decade due to a
about € 1.5 million in three years. A real power data from strong growth of economy and population. The peak load of
the PV system and a weather data from world weather Jordan’s electrical system in 2017 was 3282 MW and increased
online site of TTU location are used in this study. Decision to be 3724 MW in 2020. The peak load for the years from 2017
tree technique is employed to identify the relation between to 2020 is shown in Fig. 1
the output power and weather factors. The results show that
the system accuracy is 82.01% during the training phase
and 93.425 % on the validation set.

Keywords—PV system, Decision Tree, weather features,

Power, Real Data.

I. INTRODUCTION

H ighly demand for energy makes the electrical power

systems grow day by day rapidly. Thus, thinking of
establishing new power sources will be the persistent need. A
rapid increase in the cost of fossil fuel, which is used in
conventional energy sources, makes the decision makers Fig. 1 National grid peak load (2017.2020) [1]
employ a new kind of energy sources with lower cost and
environmentally friendly. Alternative Energy (AE) sources Solar and wind energy systems are one of the most prominent
such as wind farms, photovoltaic PV technology, biomass, sources of energy, and their utilization has become increasingly
hydropower and many different new power alternatives are popular due to modular and environmental friendly nature [2-
used. The geometrical region, environmental issues, natural 4]. Jordan has a high solar index as shown in Fig. 2.
sources and weather conditions should be taken into account
before establishing any type of AE source [1].
The use of renewable energy sources adds additional

ISSN: 1998-4464 686

 NTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING
DOI: 10.46300/9106.2020.14.88 Volume 14, 2020

sunlight is vertical, the efficiency of any given solar panel is at

its maximum.
PV is very sensitive to high temperatures and dust, which
reduces its efficiency. Solar PV is more practical for homes and
commercial buildings, while PV technology is very expensive;
its cost has decreased dramatically over time, especially in the
past five years.
Recent researches are still going on to make PV systems
more efficient. Power detection methods are widely used in the
literature, but these methods fail under a rapid change in
weather conditions. Therefore, many improvements have been
made to these algorithms to mitigate inaccurate responses
during abrupt changes in the level of sunshine.
Data mining techniques are used recently in many
applications because of its benefits to develop models and to
Fig. 2 Global solar radiation over Jordan [1]
make decisions [8-15]. Some methods use artificial intelligent
techniques such as artificial neural network as in [16-20]. Other
Fig.3 shows the renewable (PV and Wind) energy projects methods use data mining techniques like support vector
distribution all over Jordan. Most of the RE sources are located machine [21-24] and K-nearest neighbor as in [25-27]. In
in south and far away from the center. The energy is transmitted addition, there are some optimization techniques such as
to the consumer with overhead transmission lines with different genetic algorithm and particle swarm are used to predict and
voltage levels mainly 132 kV and 400 kV [1]. improve the solar system depending on environmental factors
such as the temperature, wind, and cloud [28-31].
Authors in [32] deal with similar situation, but they did their
analysis of power generated from solar system without
considering weather data; instead, they used artificial neural
network to predict the power. They claim that the power can be
forecasted without using weather data. However, the power
generated from PV system is not similar to the conventional
power because it is directly depending on weather factors such
as temperature, wind, and cloud cover. In addition, they claim
that some outliers happened in the system, and they eliminate
them. This is not true in the case of PV system since each day
has its own weather conditions and sometimes the climate
changes suddenly. Such changes affect the power generation,
which cannot be considered as outliers. The developed system
Fig. 3 PV and wind energy project distribution in this paper solve these problems.

Nowadays, producing Energy from solar system is very II. 1 MWP TTU PV POWER PLANT
important as the conventional energy has noticeable bad impact
Jordan government encourages all public universities to
on environment. Many methods are used to analyze, develop, invest in solar PV system to decrease their electricity bills.
design, and control the photovoltaic system. Such methods are Tafila Technical University started with 1 MWp project in 2016
very important especially when integrating renewable sources that cost € 1.4 million. After three years, this successful project
with conventional power [5-7]. has generated 5.4 GWH and save about € 1.5 million. The 1
The efficiency of a solar panel depends on three main factors: MWp TTU PV power plant (Fig.4) consists of 3,876 265-Wp
the efficiency of the model used for a particular panel, the PV modules of type SR-P660 245-265 manufactured by Sunrise
Energy Co. The modules are distributed on the rooftops of
number of photovoltaic model inside each solar cell, and the university buildings and car parks inside the university campus
amount of sunlight that the PV panel received. After installing as shown in Fig. 4. Four transformers are used in TTU electrical
the PV panels, inverters are added to transfer electricity from system to supply the output AC power to the buildings as shown
direct current (DC) to alternating current (AC). The efficiency in Tables 1.
of a solar panel depends on the solar incident angle. When the

ISSN: 1998-4464 687

 NTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING
DOI: 10.46300/9106.2020.14.88 Volume 14, 2020

Table 2. TTU PV system inverters distributed over all campus

Table 1. Transformers of TTU system building and car parks

Transformer Building Description Building Inverter Number Inverter Produced Inverter

number Description Model of Power Power Oversize
Inverters (kwp) (kwp) in (%)
Transformer 1 Car Park 1 2019
Engineering Faculty Business TRIO 2 55.2 60.927 10.375
Girls Hostel College 27.6 TL
Multi-Function T2
Building
Car Park 02 TRIO 2 55.2 59.442 7.685386
Science Building T4 27.6 TL
Storage Hunger Car Park 1 TRIO 4 110.4 112.668 2.054801
Transformer 2 Business College T1 27.6 TL
Labs Hunger Class Room PVI- 1 10 10.825 8.258333
Presidency Building Building 10.0-
T4 OUTD
President Suite TRIO 4 110.4 93.706 -15.1209
Transformer 3 GYM 27.6 TL
Transformer 4 Car Park 02 Engineering PVI- 3 37.5 39.014 4.038
Class Room Building Faculty 12.5-
T1 OUTD
Universal
The system uses 42 inverters distributed in the whole TTU TRIO 2 40.0 41.210 3.026875
campus as shown in Table 2, all are manufactured by ABB. Out 20.0 TL
of the 42 inverters 6 are used for photovoltaic car parks. The TRIO 4 110.4 114.603 3.807292
PV panels are arranged at a tilt angle of 22° and an azimuth 27.6 TL
Girls Hostel PVI- 2 25 28.209 12.836
angle of -16°. T1 12.5-
OUTD
GYM TRIO 8 220.8 206.288 -6.57224
T3 27.6 TL
PVI- 1 10 9.494 -5.05083
10.0-
OUTD
Labs PVI- 1 12.5 12.395 -0.83267
Hunger 12.5-
T2 OUTD
TRIO 1 27.6 22.613 -18.0658
27.6 TL
Multi- TRIO 8.5 2 17 17.294 1.732843
Function TL
Building T1
Presidency RIO 20.0 1 20 20.824 4.123333
Building T2 TL
President PVI- 1 12.5 12.669 1.354667
Suite 12.5-
T2 OUTD
Science PVI- 1 12.5 11.715 -6.27733
Fig. 4 1MWp TTU PV system for rooftop Building 12.5-
T1 OUTD
Storage PVI- 1 12.5 12.1683 -2.65333
From Table 2, it’s cleared that all inverters are working Hunger 12.5-
within the acceptable range of oversize percentage specified by T1 OUTD
ABB string configuration report. Such oversizing will not affect TRIO 1 27.6 26.620 -3.54801
the inverter expected lifetime and the inverters operate 27.6 TL
normally. Total 871.9 912.693 ---

The working principle of a PV plant is simple; photovoltaic

cells are connected in series or parallel in order to obtain the
desired current and voltage value for the PV module.

ISSN: 1998-4464 688

 NTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING
DOI: 10.46300/9106.2020.14.88 Volume 14, 2020

Modules are mounted on structures (typically aluminum

anodized) that fix the modules either to the roof or to the
ground. Ground mounted systems usually tilt the inclination
angle of the modules in order to optimize the radiation input
during the year. Roof mounted systems usually fix the modules
parallel to the roof if the slope is sufficient for good radiation
levels. On the other hand, tilted systems are also available for
flat roofs. The modules are interconnected using standard
electric copper cables.
Some ground-mounted systems use trackers (one or two
axes) in order to follow the sun path and increase the amount of
solar radiation received at the surface of the modules. The PV
array is then connected to the inverter, which converts direct
current (DC) into alternating current (AC). The inverter
Fig. 6 Energy produced by Tafila Technical University PV
incorporates Maximum Power Point Trackers (MPPT) in order
power plant in 2019
to follow the constantly changing current and voltage output of
the array. The current and voltage output are changing mainly
because of changing conditions like irradiance and temperature.
The inverter output is then connected to a meter, which registers III. SYSTEM DESIGN AND METHODOLOGY
the amount of energy that has been fed into the grid. In this paper, output power forecasting system involves five
Fig. 5 shows the output AC power generated from one phases: Data gathering and extraction phase, data preprocessing
inverter type TRIO 27.6, with generated energy 48.41 MWh phase, feature selection phase, the learning phase, and the
from GYM building modules over one-year 2019. The energy classification phase. In data gathering and extraction, the
produced by TTU PV power plant in 2019 is shown in Fig. 6. training and test set are obtained from both Tafila PV system
and weather station databases. The second phase is to pre-
process the extracted data, including cleaning unnecessarily
information, normalizing and labeling. In selected features,
dimensionality reduction of the features is being conducted. In
the learning phase, the target is to build a model using a part of
the data. The last step is using the remaining of the pre-
processed data to test the model. A test set is used to determine
the accuracy of the model. Fig. 7 shows the block diagram of
the developed system phases.
Fig. 5 Output AC power generated from one inverter of GYM
building modules, 2019

Fig. 7 Block diagram of the developed system phases

ISSN: 1998-4464 689

 NTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING
DOI: 10.46300/9106.2020.14.88 Volume 14, 2020

Fig. 8 Training model

The learning model (Fig. 8) is based on Decision Tree. predicted based on the training instances. The Decision maker
Decision Tree classifier uses the training data to build a tree output will be the amount of output power. It has four labels;
model that can be used later for classification purposes. Very Low, Medium, High, Very High. Fig. 9 shows the testing
Currently, many decision tree algorithms exist including model block diagram.
Random Forest, Random Tree, J48, and CART. Based on the
output accuracy, the features set is selected. The best set is A decision tree partition the input space of the dataset into
associated with the higher output power accuracy. There are mutually exclusive regions by giving each region a label. The
many different reduction techniques available including decision tree that consists of a root and internal nodes grows
Principal Component Analysis (PCA), and Chi-Squared for from a root node, by determining the best split that partition the
feature selection and reduction. In this paper, PCA is used as a region at internal nodes in to disjoint smaller subset and proceed
reduction technique [13],[15]. down to the leaf node (terminal nodes) labeled as – Very Low,
Medium, High, and Very High. In order to do the split an error
function that quantifies the performance of a node t in separating
The extracted data from the databases includes Average data from different classes. The used error function is impurity
Temperature, Wind speed, Humidity, Visibility, Heat index, function. The best-known impurity functions for splitting is
Pressure, Weather description, cloud cover and generated entropy function and Gini index. By using the impurity
power. After preprocessing phase and feature selection, six Function φ, the impurity measure of a given node is calculated
features are used. Five features are used as input features and to do the splitting. Fig 10 shows the steps of decision tree
one feature is used as output class. The six features are Average methodology.
Temperature, Humidity, Pressure, cloud cover, Heat index and
generated output (output class). The Decision maker is based
on decision tree classifiers. When a new data values, previously
unseen, is presented to the decision maker, the class is

Fig. 9 Testing model

ISSN: 1998-4464 690

 NTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING
DOI: 10.46300/9106.2020.14.88 Volume 14, 2020

Fig. 11 Generated output power from PV system in Tafila

Technical University

Table 3. Tafila PV station Database sample

Entity Id -> 10807492
Entity Name -> Tafila Technical University
Field -> Generated Energy
Timestamp kilowatt-hours
2/1/2020 0:00 1707.1
2/2/2020 0:00 3255.1
2/3/2020 0:00 4732.1
2/4/2020 0:00 3962.7
2/5/2020 0:00 4807.5
2/6/2020 0:00 4915.1
2/7/2020 0:00 1398.1
2/8/2020 0:00 1532.2
Fig. 10 Decision tree methodology 2/9/2020 0:00 339.4
2/10/2020 0:00 641.1
2/11/2020 0:00 3607.5
IV. EXPERIMENTS AND RESULTS 2/12/2020 0:00 4892.3
The data used in this paper is extracted from two databases. 2/13/2020 0:00 5045.7
The first database is Tafila PV database. This database is
2/14/2020 0:00 4063.4
provided from inverter manufacture (ABB) as cloud database.
The extracted data includes the power produced from the PV 2/15/2020 0:00 5096.5
system from December, 23th , 2016 to February, 29st, 2020. 2/16/2020 0:00 4038.3
This data includes 1164 records as shown in Fig. 11. Part of the
data is shown in Table 3. 2/17/2020 0:00 5005.4
2/18/2020 0:00 4270.8

ISSN: 1998-4464 691

 NTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING
DOI: 10.46300/9106.2020.14.88 Volume 14, 2020

Table 4. Weather station database sample

The second database is the weather station database, World

Weather Online database. World Weather Online is dedicated
to provide global weather forecast and weather content for
websites, businesses and the travel industry. This database
covers approximately 3 million cities/towns worldwide and its
weather forecast is trusted and used by a wide variety of
companies and organizations from SME's to large corporate Fig. 12 Pressure data
clients. The weather station currently operates two high-tech
weather data centers; one situated in Denmark and the other in
Germany. The station delivers reliable and accurate weather
information for any geo-point in the world. The used weather
model is run along with other metrological models like
European Centre for Medium-Range Weather Forecasts, World
Meteorological Organization, NASA weather satellite imagery,
NOAA GFS2 model and JMA model for research and training
purposes. It deliver the most accurate weather forecast possible. Fig. 13 Temperature data
The extracted data from weather station includes many features
such as maxtempC, mintempC, avgtempC, totalprecipMM,
windspeedKmph, sunhour winddirdegree, weatherDesc,
humidity, visibilityKm pressureMB, cloudcover, heatIndexC,
dewPointC, windChillC, windGustKmph, feelsLikeC, sunrise,
sunset, moonrise, and moonset. A sample from the weather
database is shown in Table 4.
In this paper, just eight weather features are used. These
features avgtempC, windspeedKmph, weatherDesc, humidity,
visibilityKm pressureMB, cloudcover, and HeatIndexC. The
other features has been eliminated. Fig. 14 UV Index data
Fig. 12, Fig. 13, Fig. 14 and Fig. 15 show examples of the
extracted features. It shows the weather data from July 1st ,
2008 to Februry, 12, 2020 (4232 records) for Tafila Technical
University location where the PV system is located.

Fig. 15 Wind data

ISSN: 1998-4464 692

 NTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING
DOI: 10.46300/9106.2020.14.88 Volume 14, 2020

Table 5. Bended database (Weather database and PV system database)

A normalization is applied to the output power values in  Level 4: value from 0.80 to 1.0 represents Very
order to label them. The power values are normalized to values High Similarity generated power amount.
between zero and one. Each generated output power is assigned
as a level that belongs to one of the four levels, i.e., The levels The blended database has been constructed from the two
are shown below: databases in order to train and test the decision maker model
 Level 1: value from 0.00 to 0.39 represents Very after assigning the power output level for each record. Table 5
Low generated power amount. shows some records from the bended database.
 Level 2: value from 0.40 to 0.59 represents
Medium Similarity generated power amount. Fig. 16 shows the histogram of five feature with a projection
 Level 3: value from 0.60 to 0.79 represents High of output power label presented as colors.
Similarity generated power amount.

Fig. 16 Input data histogram

ISSN: 1998-4464 693

 NTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING
DOI: 10.46300/9106.2020.14.88 Volume 14, 2020

Recall (R) is the ratio of the relevant data among the retrieved.
In this study, WEKA (Waikato Environment for Knowledge Precision (P) is the ratio of the accurate data among the
Analysis) is used to construct decision trees according to the retrieved data. Their formulas are given as follow:
training set with the standard algorithm J48. Weka is a popular
suite of machine learning software written in Java, developed at 𝑹𝒆𝒄𝒂𝒍𝒍(𝑹) =
𝑻𝑷
if TP+FN > 0, otherwise undefined.
the University of Waikato, New Zealand. Weka is a free 𝑻𝑷 𝑭𝑵

software tool available under the GNU General Public License. The recall of a class “Very Low” is defined as:
It contains a collection of visualization tools and algorithms for 𝑎
data analysis and predictive modelling that support data pre- 𝑅 =
processing, clustering, classification, regression, visualization, 𝑆𝑢𝑚_𝐶1
and feature selection. Weka has a powerful Graphical User 𝑻𝑷
Interface that supports its functionality. 𝑷𝒓𝒆𝒄𝒊𝒔𝒊𝒐𝒏(𝑷) = if TP+FN > 0, otherwise undefined.
𝑻𝑷 𝑭𝑷
Once the features are extracted and grouped into a feature The precision of a class “Very Low” is defined as:
vector, classification takes place, where the output power
values are classified in one of the following classes: Very Low, 𝑎
𝑃 =
Medium, High, Very High. J48 (C4.5) is an algorithm used to 𝑆𝑢𝑚_𝑟1
generate a decision tree. It has been developed by John Ross
Quinlan [33]. C4.5 is an extension of Quinlan's earlier ID3 Classifier F1 rating is the harmonic mean of the classifier recall
algorithm. The decision trees generated by C4.5 is normally and the precision. It is given as
used for classification. It has been ranked first in the Top 10
Algorithms in Data Mining algorithm [34]. 𝟐∗𝑷∗𝑹
𝑭𝟏 =
Java was selected as the development language, and J2SDK 𝑷+𝑹
version 1.6.0_22 was used for the Java running environment. where R represents the recall, , and P represents the precision
Access database 2013 is adopted for the development of
database. Java database connectivity (JDBC), an application- Accuracy, which indicates the fraction of correctly classified
programming interface for the Java programming language, is samples among all the samples, obtained by:
used to access the database. JDBC could wrap a structured
query language (SQL) statement, send it to the database, and 𝑻𝑷 + 𝑻𝑵
𝑨𝒄𝒄𝒖𝒓𝒂𝒄𝒚 =
retrieve the desire. 𝑻 𝑷 + 𝑻 𝑵 + 𝑭𝑷 + 𝑭𝑵
Part of extracted data from January, 1st, 2017 to February 12,
2020 (774 records excluding 2018 in order to be used later for 𝑎+𝑓+𝑘+𝑃
validation) has been used for training and testing the decision =
𝑇𝑜𝑡𝑎𝑙
maker module. The confusion matrix for each class (Very Low,
Medium, High Very High) is constructed. The confusion matrix Kappa statistic is used to give the agreement between
has the form shown in Table 6. developed model and the actual PV system. A Kappa score
ranges between 1 which shows full agreement and 0 which
Table 6. Confusion matrix shows no agreement. Table 7 shows the resulted power output
Very Medium High Very evaluation results.
Low High
Very Low a b c D Sum_r1 Table 7. Overall performance results (training and testing set).
Medium e f g H Sum_r2 Decision Tree
High i j k L Sum_r3 Correctly Classified Instances 634
Very m n o P Sum_r4 Correctly Classified 82.0181 %
High Percentage
Sum Sum_c1 Sum_c2 Sum_c3 Sum_c4 Total Incorrectly Classified 139
Instances
Incorrectly Classified 17.9819 %
10-fold cross validation is used in this paper. In K-fold the Percentage
training set will be randomly splitted into K sets that have Kappa statistic 0.7547
approximately the same size. Then the Decision Tree will be Mean absolute error 0.1356
trained using (K-2) subset. One of the two remaining subsets Root mean squared error 0.2604
will be used for validation and the last for testing. This process Relative absolute error 37.1043 %
will be repeated K times, while a different subset is used for Root relative squared error 60.9173 %
testing and validation. Total Number of Instances 773

The performance measurements used for this paper are recall, Another performance indicated by confusion matrix is shown
precision, classifier F1 rating and accuracy. They are defined as in Table 8. This confusion matrix is built based on data testing.
follows:

ISSN: 1998-4464 694

 NTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING
DOI: 10.46300/9106.2020.14.88 Volume 14, 2020

We constructed the confusion matrix for each class (Very Low, than 0.4, and fair to good agreement for kappa coefficient
Medium, High, Very High). between 0.40 and 0.75. In this paper, Kappa coefficient is 0.75,
which shows excellent agreement.
Table 8. Confusion matrix (training and testing set).
Output Real System As the number of features in database was eight features, we
Power used PCA dimensionality reduction method, and we selected
Very Medium High Very around 63% of the features (5 features) with the highest
Low High importance. The Accuracy percentage for each number of
Very 128 3 4 0 features is shown in Table 10.
Low
Medium 19 112 11 3 Table 10. Feature selection
High 6 9 182 50 Number of Features Accuracy (%)
Model

Very 5 8 21 212 2 64.32%

High 3 71.53%
4 73.75%
The performance measurements result is shown in Table 9. 5 82.01%
6 82.01%
Table 9. Performance measurements 7 82.01%
TP FP Precision Recall F- Class 8 82.01%
Rate Rate Measure
0.948 0.04 0.810 0.948 0.874 very Using another 365 cases from January,1st, 2018 to December
low 31st 2018 from the original extracted databases not previously
0.772 0.03 0.848 0.772 0.809 Medium used in the training or used in cross validation. Table 11 shows
0.737 0.06 0.835 0.737 0.783 High 44 records of these cases as an example with both real system
0.862 0.10 0.800 0.862 0.830 very output level value and the model output. Fig. 17 shows a part of
high the learned decision tree associated with these records which is
triggered to make power output decision.
There is excellent agreement if the Kappa coefficient is
greater than 0.75, poor agreement for Kappa coefficient less

Fig. 17 Part of decision tree used in validation step

ISSN: 1998-4464 695

NTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING
DOI: 10.46300/9106.2020.14.88 Volume 14, 2020

Table 11. Sample of validation data set

Temperature

(real System)
Cloud Cover

Heat Index

Power out

Power out
Humidity

Pressure
Average

(Model)
ID
1 36 30 1006 5 36 medium medium
2 35 26 1010 5 35 medium medium
3 32 45 1012 5 33 medium medium
4 26 44 1014 8 27 medium high
5 27 24 1010 19 26 medium medium
6 25 24 1017 25 24 medium medium
7 25 27 1011 35 24 medium medium
8 30 24 1008 6 29 medium medium
9 28 41 1006 7 28 medium medium
10 27 17 1011 15 26 medium medium
11 34 29 1011 9 33 high high
12 32 37 1016 5 32 high high
13 30 48 1016 5 31 high high
14 32 34 1015 13 31 high high
15 31 21 1006 11 30 high high
16 28 45 1015 7 29 high high
17 29 43 1014 7 29 high high
18 31 25 1008 12 29 high high
19 28 47 1013 8 28 high high
20 29 15 1008 42 28 high high
21 29 25 1008 35 28 high high
22 29 29 1008 19 28 high high
23 29 28 1007 10 28 high high
24 27 48 1005 6 28 high high
25 28 29 1015 8 27 high high
26 26 52 1014 9 27 high high
27 26 54 1013 19 27 high very low
28 28 22 1010 27 27 high high
29 25 51 1015 10 26 high high
30 27 22 1014 24 26 high high
31 27 25 1010 33 26 high high
32 26 18 1010 16 25 high high
33 25 33 1009 16 25 high high
34 25 21 1011 67 24 high high
35 34 15 1014 6 32 high high
36 33 26 1013 16 32 high high
37 30 23 1012 31 29 high high
38 26 29 1012 5 25 very high Very high
39 25 32 1012 5 25 very high very high
40 25 17 1012 10 24 very high very high
41 25 31 1012 6 24 very high very high
42 30 47 1015 14 32 very high very high
43 28 34 1008 7 28 very high very high
44 27 42 1008 20 27 very high very high

ISSN: 1998-4464 696

 NTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING
DOI: 10.46300/9106.2020.14.88 Volume 14, 2020

The prediction result by the system is highly matched by the

actual PV system generation. The achieved results are shown References
in Table 12 and Table 13. These coefficients suggest excellent [1] A. Al-Azzam and Amani, ‘The Grid (Opportunity vs.
agreement between the real system and the developed model. Challenges)’, Jordanian Forum Renew. Energy, 2014.
[2] V. Khare, S. Nema, and P. Baredar, ‘Solar-wind hybrid
Table 12. Confusion matrix (validation set) renewable energy system: A review’, Renew. Sustain.
Output Real System Energy Rev., vol. 58, pp. 23–33, 2016.
Power [3] V. Vittal and T. G. Heydt, ‘Impact of large scale PV
Very Medium High Very generation’, pp. 1–1, 2014.
Low High [4] R. Shah, N. Mithulananthan, A. Sode-Yome, and K. Y.
Very 88 2 0 1 Lee, ‘Impact of large-scale PV penetration on power
Low system oscillatory stability’, IEEE PES Gen. Meet. PES
Medium 4 77 1 2 2010, pp. 1–7, 2010.
High 1 3 68 5 [5] P. S. Panel and P. Affairs, ‘Integrating Renewable
Model

Very 2 2 1 108 Electricity on the Grid A Report by the APS Panel on

High Public Affairs’, Leadership, pp. 1–38, 2010.
[6] G. Balaban, G. C. Lazaroiu, V. Dumbrava, and C. A. Sima,
‘Analysing Renewable Energy Source Impacts on Power
System National Network Code / Romania’, 2017.
Table 13. Overall performance results [7] Obeidat, M.A. and Al-Sarayrah, R.M., “The Impact of
(validation set) Renewable Energy Integration on Stability of the Jordanian
Precision Recall F- Class National Grid,” International Journal of Computer
Measure Applications, 975, p.8887, 2019.
0.967 0.926 0.946 very low [8] Ayman M. Mansour, Murad M. Alaqtash, Mohammad
0.916 0.916 0.916 Medium Obeidat “Intelligent Classifiers of EEG Signals for
0.971 high Epilepsy Detection,” WSEAS Transactions on Signal
0.883 0.924 Processing, Volume 15, 2019.
0.955 0.931 0.942 very high [9] Jafar.Abu Khait, Ayman M Mansour and Mohammad
Correctly Classified Percentage 93.425 % Obeidat, “Classification based on Gaussian-kernel Support
Kappa statistic 0.912 Vector Machine with Adaptive Fuzzy Inference System,”
Total Number of Instances 365 PRZEGLĄD ELEKTROTECHNICZNY. Vol 5, pp 16-24,
2018.
[10] B. Hawashin et al., "Efficient Texture Classification Using
Independent Component Analysis," 2019 IEEE Jordan
V. ACKNOWLEDGMENT International Joint Conference on Electrical Engineering
The authors wish to express their thanks for the support of and Information Technology (JEEIT), Amman, Jordan, pp.
Tafila Technical University. The authors very much appreciate 544-547, 2019.
[11] Ayman M. Mansour, “Decision Tree-Based Expert System
the support by Kawar Energy Company and Electricity
for Adverse Drug Reaction Detection using Fuzzy Logic
Distribution Company (EDCO).
and Genetic Algorithm,” International Journal of
Advanced Computer Research (IJACR), Vol 8(36), 2018.
[12] Mohammad A. Obeidat and Ayman M. Mansour, “EEG
VI. CONCLUSION Based Epilepsy Diagnosis System using Reconstruction
In this paper, a new model of 1 MWp TTU PV system is Phase Space and Naïve Bayes Classifier,” WSEAS
analyzed using intelligent decision classifier. The developed Transactions on Circuits and Systems, Volume 17, 2018.
system gives the relation between the generated output power [13] Ayman M. Mansour, “Texture Classification using Naïve
and different weather conditions. The model of the system is Bayes Classifier,” International Journal of Computer
built based on real output power and weather databases. The Science and Network Security (IJCSNS), vol. 18, no. 1,
results show high accuracy and precision. In the future work January 2018
different intelligent techniques will be used to increase system [14] Bilal Hawashin, Ayman M. Mansour and Shadi
accuracy and precision. Aljawarneh, “An Efficient Feature Selection Method for
Arabic Text Classification,” International Journal of
Computer Applications (IJCA), vol. 83, no. 17, pp. 1-6,
December 2013.

ISSN: 1998-4464 697

 NTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING
DOI: 10.46300/9106.2020.14.88 Volume 14, 2020

[15] D.A. Al Nadi and Ayman Mansour, "Independent International Conference on Renewable Energy Research
Component Analysis (ICA) for texture classification", 5th and Applications (ICRERA). IEEE, 2016.
International Multi-Conference on Signals and Devices, [29] Ebrahimi, S. Mohammadreza, et al. "Parameters
IEEE SSD, 2008. identification of PV solar cells and modules using flexible
[16] Almonacid, F. J. M. F., Rus, C., Hontoria, L., & Munoz, F. particle swarm optimization algorithm." Energy 179: 358-
J. “Characterisation of PV CIS module by artificial neural 372, 2019.
networks. A comparative study with other methods,” [30] Rajan, Naik Aashay, et al. "Solar PV array reconfiguration
Renewable Energy, 35(5), 973-980,2010. using the concept of standard deviation and genetic
[17] Garud, Kunal Sandip, Simon Jayaraj, and Moo‐Yeon Lee. algorithm." Energy Procedia 117: 1062-1069, 2017.
"A review on modeling of solar photovoltaic systems using [31] Kumari, P. Ashwini, and P. Geethanjali. "Adaptive genetic
artificial neural networks, fuzzy logic, genetic algorithm algorithm based multi-objective optimization for
and hybrid models." International Journal of Energy photovoltaic cell design parameter extraction." Energy
Research, 2020. Procedia 117: 432-441, 2017.
[18] Almonacid, Florencia, et al. "Review of techniques based [32] Ordiano, Jorge Ángel González, et al. "Photovoltaic power
on artificial neural networks for the electrical forecasting using simple data-driven models without
characterization of concentrator photovoltaic technology." weather data." Computer Science-Research and
Renewable and Sustainable Energy Reviews 75 : 938-953, Development 32.1-2: 237-246, 2017.
2017. [33] Quinlan JR. “C4. 5: programs for machine learning,”
[19] Rodríguez, Fermín, et al. "Predicting solar energy Elsevier; 2014.
generation through artificial neural networks using weather [34] Wu X, Kumar V, Quinlan JR, Ghosh J, Yang Q, Motoda
forecasts for microgrid control." Renewable energy 126: H, et al. “Top 10 algorithms in data mining,” Knowledge
855-864,2018. and Information Systems volume 14, pp 1–37, 2008.
[20] Monteiro, Raul VA, et al. "Estimating photovoltaic power
generation: performance analysis of artificial neural
networks, Support Vector Machine and Kalman filter." Dr. Ayman M Mansour received his
Electric Power Systems Research 143:643-656, 2017. Ph.D. degree in Electrical
[21] Kazem, Hussein A., Jabar H. Yousif, and Miqdam T. Engineering from Wayne State
Chaichan. "Modeling of daily solar energy system University in 2012. Dr. Mansour
prediction using support vector machine for Oman." received his M.Sc degree in Electrical
International Journal of Applied Engineering Research Engineering from University of
11.20: 10166-10172, 2016. Jordan, Jordan, in 2006 and his B.Sc
[22] Zendehboudi, Alireza, M. A. Baseer, and R. Saidur. degree in Electrical and Electronics
"Application of support vector machine models for Engineering from University of
forecasting solar and wind energy resources: A review." Sharjah, UAE, in 2004. He graduated top of his class in both
Journal of cleaner production 199: 272-285, 2018. Bachelor and Master. Currently, Dr. Mansour is an Associate
[23] Bae, Kuk Yeol, Han Seung Jang, and Dan Keun Sung. Professor in the Department of Communication and Computer
"Hourly solar irradiance prediction based on support vector Engineering, Tafila Technical University, Jordan. He is also the
machine and its error analysis." IEEE Transactions on director of the Energy Research Center at Tafila Technical
Power Systems 32.2: 935-945, 2016. University. His areas of research include Communication
[24] Mojumder, Juwel Chandra, et al. "Application of support Systems, Multi-agent Systems, Fuzzy Systems, Data Mining
vector machine for prediction of electrical and thermal and Intelligent Systems. He conducted several researches in his
performance in PV/T system." Energy and Buildings 111: area of interest. Dr. Mansour is a member of IEEE, Michigan
267-277,2016. Society of Professional Engineers, IEEE Honor Society (HKN),
[25] Liu, Zhao, and Ziang Zhang. "Solar forecasting by K- Society of Automotive Engineers (SAE), Tau Beta Pi Honor
Nearest Neighbors method with weather classification and Society, Sigma Xi and Golden Key Honor Society.
physical model." 2016 North American Power Symposium .
(NAPS). IEEE, 2016.
[26] Kartini, Unit Three, and Chao Rong Chen. "Short term Prof, Jalal M. Abdallah was born in
forecasting of global solar irradiance by k-nearest neighbor Amman, Jordan. He received his
multilayer backpropagation learning neural network M.S.1994 and Ph.D.1998 in electrical
algorithm." Proceedings of the International Conference on engineering from Vinitsa Stat Technical
Graphics and Signal Processing. 2017. University, Ukraine. Since 2005, he has
[27] Valgaev, Oleg, Friederich Kupzog, and Hartmut Schmeck. joined Tafila Technical University where
"Building power demand forecasting using K-nearest he is currently a Professor at the electrical
neighbours model–practical application in Smart City Engineering Department and the vice president of academic
Demo Aspern project." CIRED-Open Access Proceedings affairs. Dr. Jalal Abdallah main research interests are power
Journal 2017.1: 1601-1604, 2017. system analysis, monitoring, and diagnostic of high-voltage
[28] Mercado, Katheryn Donado, Jamer Jiménez, and M. power transformers.
Christian G. Quintero. "Hybrid renewable energy system
based on intelligent optimization techniques." 2016 IEEE

ISSN: 1998-4464 698

 NTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING
DOI: 10.46300/9106.2020.14.88 Volume 14, 2020

Dr. Mohammed Obeidat Received his He is a member of IEEE, Tau Beta PI Honor Society and
PhD in Electrical Engineering from Golden Key Honor Society. He was given the honor to be a
Wayne State University in 2013, M.Sc Sigma Xi member from the Board of Governor, in 2012. He
degree in Electrical Engineering was demonstrated excellent research and academic abilities as well
from Yarmouk University , Jordan in as professional potentials.
2006, and B.Sc degree in Electrical Dr. Obeidat has published several journal papers. Dr. Obeidat
Engineering from Jordan University of research of interest in the field of intelligent control systems,
Science & Technology, Jordan in 1999. renewable energy, intelligent systems, and mechatronics
He is an Associate professor in power and mechatronics
department at Tafila Technical University. Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the Creative
Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en_US

ISSN: 1998-4464 699