Machine Learning in Renewable Energy Application:

Intelligence System for Solar Panel Cleaning

AHMAD AL-DAHOUD1, MOHAMED FEZARI2, ALI ALDAHOUD 3

1Faculty of Architecture and Design, Al-Zaytoonah University of Jordan, Amman, JORDAN

2Electronics and computer architecture at the University of Badji Mokhtar Annaba, ALGERIA

3Faculty of Science and IT University of Jordan, Amman, JORDAN

Abstract: - The objective of this study is to develop an automatic cleaning system for Photovoltaic (PV) solar

panels using machine learning algorithms. The experiment includes two phases. Phase one is to perform testing

and reading of the sensor in 4 different classes which include no-dust, little dust, dusty, and very dusty during

day and night time. The reading was taken using a visual inspection of the solar panel and the sensor reading

using a multimeter. Phase two uses supervised learning to test and calibrate the sensor using the KNN

algorithm. The classification was done using the data gathered from the sensor with one of the main classes

identified. A total of 800 readings were taken. The results show the sensor reading taken during the night was

more stable and accurate due to the sensor's sensitivity to noise which includes: heat and light during the

daytime. Secondly, using machine learning (KNN algorithm) we get a 95% (with K=5) correct classification

for the four main classes which determines the level of cleaning needed for the solar panel.

Key-Words: - Solar panel cleaning using machine learning, Machine learning in renewable energy application,

classification for dust detection, sensor-based dust detection, Machine Learning, Classification

Algorithms

Received: December 8, 2022. Revised: April 6, 2023. Accepted: April 29, 2023. Published: May 16, 2023.

1 Introduction

We anticipate that governments worldwide will

soon be compelled to provide alternative energy

sources to maintain their economies or resort to

programmed blackouts. A proposed solution to

address this pressing energy crisis is the

establishment of national energy centres across the

globe. Jordan has followed suit with the formation

of its own National Energy Research Centre in

Amman. This centre aims to conduct research,

development, and training in the fields of new and

renewable energy while improving energy

efficiency across various economic sectors, [1], [2].

Similarly, Algeria has implemented an ambitious

program to promote the use of renewable natural

resources. It has shifted towards alternative energy

sources, which have become a rational and strategic

choice due to the nature and cost of these resources.

Notably, the research project highlights that Jordan's

progress in this field is ahead of many other

countries, with several universities following the Al-

Zaytoonah University project, which has reduced

electricity costs by 30%.

The implementation of this resource,

particularly in steppe regions beyond the reach of

rural electrification networks, has played a

significant role in settling numerous nomadic

families in proximity to their agricultural lands or

grazing pastures. As a result, the demand for this

type of solar energy equipment continues to surge in

highland and steppe areas, as demonstrated in

Figure 1.

Fig. 1: Irradiation map in Algeria and Jordan

Algeria has an abundant solar field that can help

the country transition from using disproportionate

amounts of fossil fuels to clean energy. Despite

having a significant energy potential exceeding 5

billion Gwh annually, its reliance on non-renewable

energy sources still surpasses its solar resources.

The country experiences an average of 2,250 hours

of sunshine in the north and 3,600 hours (about 5

months) in the south, with respective potentials of

1,700 and 2,650 kWh per year, [10]. However, PV

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.45

Ahmad Al-Dahoud,

Mohamed Fezari, Ali Aldahoud

E-ISSN: 2224-3496

472

Volume 19, 2023

panels encounter various challenges that lead to

energy losses. PV panels which are:

1. Partial shading: the environment of a

photovoltaic module can include trees,

mountains, walls, buildings, etc. It can cause

shadowing on the module which directly affects

the energy collected, [8].

2. Total shading: (dust or dirt) their deposit causes

a reduction in the current and voltage produced

by the photovoltaic generator (3-6%), [14].

3. Nominal power dissipation: the photovoltaic

modules resulting from the industrial

manufacturing process are not all identical.

Manufacturers guarantee lower deviations of

3% to 10% around the nominal power. In

practice, the photovoltaic solar module operates

according to the performance of the worst panel:

the nominal power is therefore generally lower

than that prescribed by the manufacturer, [16].

4. Loss of connections: the connection

between slightly different power modules

causes slightly reduced power operation. They

increase with the number of modules in series

and parallel (3%).

5. Angular or spectral losses: the photovoltaic

modules are spectrally selective; the variation of

the solar spectrum affects the current generated

by them. Angular losses increase with the angle

of incidence, [18].

6. Losses by Ohmic drops: hmic drops are

characterized by voltage drops due to the

passage of current in a conductor of given

material and section. These losses can be

minimized with proper sizing of these

parameters, [6].

7. Losses due to heat: the modules lose on average

0.4% of production per degree higher than the

standard temperature (25 ° C under standard

conditions of STC measurements). The

operating temperature of the modules depends

on the incident solar irradiation, the ambient

temperature, the material colour, and the wind

speed (5% to 14%), [11].

8. Losses due to the DC / AC performance of the

inverter can be characterized by a yield curve as

a function of the operating power (6%), [5].

9. Losses by tracking the maximum power point

the inverter has an electronic device that

calculates in real-time the maximum power

operating point (3%), [3].

10. Losses due to the natural aging of the modules

on average, a module in the open air loses less

than 1% of its capacity per year, [9].

1.1 Problem of Dust on Solar Panels

The sunny deserts are an attractive option for solar

energy, but the high level of dust presents a

significant problem (Figure 2). To maintain optimal

conditions, solar panel owners need a way to clean

the panels regularly. If left uncleaned, the panels can

lose up to 0.4-0.8% efficiency per day and up to

60% after dust storms. However, watering the

panels with water in arid zones can be challenging

and labour-intensive, particularly in remote desert

locations with extreme temperatures that can reach

over 122 degrees Fahrenheit during the day. Despite

this, photovoltaic modules generally do not require

much maintenance.

Fig. 2: Dust of solar panels

The first innovation is the development of automatic

cleaning systems that use specialized robots to clean

solar panels. These systems can be programmed to

clean the panels regularly, reducing the need for

human intervention in remote desert locations. The

second innovation is the use of anti-reflective

coatings on the surface of the solar panels, which

can reduce the build-up of dust and improve their

overall efficiency. These coatings are designed to

repel dust and other particles, making it easier for

wind and rain to remove them. With these

innovations, the maintenance of large solar

installations in desert environments can be made

more efficient, safer, and less costly. Two recent

innovations can contribute to the maintenance of

large installations with greater safety for the

personnel and less risk of damaging the modules:

a. Robot cleaners (remotely controlled by Wi-Fi)

can clean the panels, [13];

b. Anomaly monitoring drones,

Advanced technologies such as drones equipped

with high-resolution gyro-stabilized infrared

cameras are being used in France for centralized

tele-monitoring of solar installations. EDF ENR

Solaire, a subsidiary of EDF Energies Nouvelles,

created a solar roof control center in 2009 that now

monitors 550 installations with a total power of

about 55 MW, including 150 owned by EDF EN.

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.45

Ahmad Al-Dahoud,

Mohamed Fezari, Ali Aldahoud

E-ISSN: 2224-3496

473

Volume 19, 2023

The drones enable remote operators to detect circuit

faults at an earlier stage and intervene quickly and

efficiently. Although drones have limited autonomy

and are expensive to operate, their maneuverability

and speed of intervention make them economically

feasible.

Ensuring proper maintenance of a solar

installation is crucial to maintaining its efficiency

and preventing any loss in production, [17]. Even

minor issues such as bird droppings or a thin layer

of dust can significantly affect the output of the

plant, leading to a decrease in income. In cases

where micro-inverters or optimizers are installed,

only the dirty panels will experience production

loss, whereas a conventional photovoltaic inverter

can impact the entire installation's production, [4].

To address this, various traditional and advanced

cleaning methods are available, as depicted in

Figure 3. Regular cleaning can help maintain

optimal conditions for solar panels, thus ensuring

maximum efficiency and income generation [20],

[21].

Fig. 3: Traditional and advanced methods of

cleaning the solar panels

2 Related Literatures

The significance of solar energy as a renewable

energy source is progressively growing.

However, dust accumulation on solar panels can

significantly reduce their efficiency and output

power. Regular cleaning is essential to maintain the

performance of the solar panels. Various cleaning

methods have been proposed, including manual

cleaning, water sprinkling, and robotic cleaning.

However, determining the frequency and extent of

cleaning required can be a challenging task, [7].

Recently, machine learning techniques have

been used to aid in the detection and classification

of dust on solar panels. The KNN algorithm has

been used to classify data into different categories,

ranging from no dust to very dusty. This method is

effective in determining the level of cleaning

required for solar panels, [12].

Other machine learning algorithms, such as

neural networks and decision trees, have also been

applied to the problem of solar panel cleaning. For

example, a decision tree-based approach was used to

determine the optimal cleaning frequency based on

weather conditions and dust accumulation, [25].

Robotic systems have been developed that use

machine learning algorithms to detect and clean dust

on solar panels automatically. These systems use

sensors and cameras to detect dust on the panels and

apply cleaning solutions using spray nozzles or

brushes. Machine learning algorithms are used

to enhance the cleaning process and ensure that

the panels are cleaned taking into consideration

efficiency and effectiveness, [15].

Overall, the use of machine learning for solar

panel cleaning systems shows promise in improving

the efficiency and reliability of solar energy

production. Further research is needed to develop

more accurate and efficient algorithms for detecting

and cleaning dust on solar panels, [26].

3 Proposed Solution

Solar panels are exposed to various sources of

pollution and fouling in their environment, such as

industrial pollutants, car pollution, rain, acid,

chimneys, pollens, dust, sands, leaves of trees,

moss, mushrooms, salts in a marine environment,

limestone, and residues of cleaning products, [19].

These pollutants not only lower the yield of the

panels but also generate intense heating phenomena

through the "hot spot" effect, leading to premature

wear of the modules. Additionally, the angle of

inclination of PV modules can also lower their

efficiency. To address these issues, an electrical

system consisting of a dust sensor, an Arduino Uno

M-controller, two end races to designate the panel

edges, a circuit L298D acting as a bridge to power

the engine, and a DC motor for the translatory

movement of ballet was simulated under Proteus.

The results were promising, as shown in Figure 4.

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.45

Ahmad Al-Dahoud,

Mohamed Fezari, Ali Aldahoud

E-ISSN: 2224-3496

474

Volume 19, 2023

Fig. 4: Overall system layout.

A graphical user interface (GUI) allows for

interactive control of an application using a mouse,

as opposed to relying solely on keyboard

commands. GUIs typically feature menus, buttons,

scroll bars, checkboxes, lists, and text boxes, as

depicted in Figure 5.

The creation of graphical interfaces in

MATLAB has been made possible since version 5.0

(1997) with the introduction of a dedicated tool

called GUIDE (Graphical User Interface

Development Environment). With GUIDE,

programmers can easily create intuitive graphical

user interfaces using tools such as menus, buttons,

elevators, checkboxes, checklists, and text boxes.

The tool can be launched by clicking on the icon or

typing "guide" in the MATLAB Command

Window.

To connect the M-Controller to the PC, the

"Connect" button is used, which turns green to

indicate a successful connection. Clicking on the

"START" button enables visualization of the

system's state, and the sensor's value and engine

status are displayed. When artificial dust is

deposited on the sensor, the value of the sensor

increases, triggering the engine to start the cleaning

procedure of the PV panel.

The dust sensor comprises a transmitter and an

IR receiver that continuously emit spokes until

obstructed by dust, leading to a decrease in the

sensor value. Figure 6 shows the quantity of dust on

the sensor based on the sensor value. The proposed

dust sensor has four classification categories: no

dust, little dust, dusty, and very dusty.

Using the proposed sensor design, we use

machine learning algorithms to classify the level of

cleanliness based on the reading of the sensor and

categories the level of cleanliness needed into 4

categories (refer to section 4).

Fig. 5: Overall system layout

Fig. 6: Designed Dust Sensor

4 Machine learning for Dust

Classification

Machine learning has emerged as a promising

approach to automate the cleaning of solar panels.

By leveraging sensors, image processing techniques,

and machine learning algorithms, these systems can

detect and classify the level of dirt and debris on

solar panels and automatically clean them. Machine

learning-based cleaning systems have the potential

to significantly reduce the cost and time required for

solar panel maintenance and increase the efficiency

and lifespan of solar energy systems. In this context,

research on the development of machine learning-

based solar panel cleaning systems has gained

significant attention in recent years, [23], [24].

Machine learning has been used in a variety of

applications such as data mining, optimization, and

classification, [22]. In this section, we use machine

learning to classify the level of cleaning needed for

the solar panel. Our approach starts with measuring

sensor sensitivity to multiple factors which include

noise, heat, and light. The experiment includes

taking the sensor reading for (no-dust) surface using

a multimeter, where first the surface will be cleaned

and then the reading is taken. this approach was

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.45

Ahmad Al-Dahoud,

Mohamed Fezari, Ali Aldahoud

E-ISSN: 2224-3496

475

Volume 19, 2023

applied to data gathering during the night and

daytime. A total of 800 reading was taken using a

multimeter and visual inspection of the solar surface

for the dust. The results show that the sensor failed

to detect (no dust) during daylight as the reading

was fluctuating and unstable using a multimeter as

shown in Figure 7 whereas, during the night the

results show that the multimeter reading shows

stability as shown in Table 1, where during the

night-time the average reading was 0.84 volts with

stander deviation 0.3, whereas during the daytime

the average was 0.66 with stander deviation 0.15.

Fig. 7: Sample of Dust level and multimeter reading

for No-dust class

Table 1. Average reading in volts during Daytime

and Night time

Figure 8, shows a small sample of the multimeter

reading for each class during daytime. As shown

noise, heat, and light will affect the reading of the

sensor and the decision-making process for

determining cleaning level. Using the KNN

algorithm we train a classifier to identify the four

different types of classes where we gain a 56% on

average, this indicated that the classifier has above

the chance to correctly classify each class.

Figure 9 shows a small sample for the

multimeter reading for each class during night-time,

where the reading shows stability and consistency

for the voltage reading for the no dust class. As

daytime data was insufficient, we start gathering

sensor readings during the night for identifying four

types of classes which include no-dust, little dust,

dusty, and very dusty. These classes were

introduced to classify the sensor reading and

identify the level of cleaning needed. A total of 400

reading was gathered for the four types of classes

and used the KNN algorithm to classify and cluster

the data gathered. Using the KNN classifier we get

95% accuracy, where we found the best K value is

(5) using Euclidean distance and cross-validation

criteria on the data. We use supervised learning by

labeling the data and training the KNN classifier.

Table 2 shows the KNN classifier and the detection

rate for each class.

Figure 10 shows the cluster for each class after

the classification is done, as shown each class has its

cluster were using the KNN algorithm we can

classify each sensor reading and determine the level

of cleaning needed.

Fig. 8: Sample for the reading for each class during

Daytime

Fig. 9: Sample for the reading for each class during

nighttime

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.45

Ahmad Al-Dahoud,

Mohamed Fezari, Ali Aldahoud

E-ISSN: 2224-3496

476

Volume 19, 2023

Fig. 10: Classification for each dust class

Table 2. KNN classifier accuracy %

5 Conclusions

In this study, the impact of dust on solar panels is

examined by measuring various voltages at different

times of the day. Using machine learning techniques

to utilise and determine the level of cleaning

required for the solar panel a total of 800 readings

were gathered. Using the KNN algorithm to classify

the data into four categories: no dust, little dust,

dusty, and very dusty. The results show that sensor

readings are consistent during night-time but

unstable during daylight hours due to multiple

factors such as heat, noise, and light. By employing

the KNN classifier, the daytime data achieves 56%

correct classification. On the other hand, the KNN

algorithm applied to the night-time data

demonstrates a 95% correct classification with a

value of K=5.

References:

[1] A Al-Salaymeh. Modelling of global daily

solar radiation on horizontal surfaces for

amman city. Emirates Journal for Engineering

Research, 11(1):49–56, 2006.

[2] Anas Al Tarabsheh and Mohammad Ababne.

Analysis of solar radiation in jordan. Jordan J

Mech Ind Eng, 4(6):733–737, 2010.

[3] Bader Naser Alajmi, Khaled Hani Ahmed,

Grain Philip Adam, Stephen Jon Finney, and

Barry Wayne Williams. Modular multilevel

inverter with maximum power point tracking

for grid connected photovoltaic application. In

2011 IEEE International Symposium on

Industrial Electronics, pages 2057–2062.

IEEE, 2011.

[4] Stephen R Allen, GP Hammond, HA Harajli,

CI Jones, MC McManus, and AB Winnett.

Integrated appraisal of micro-generators:

methods and applications. Proceedings of the

Institution of Civil Engineers-Energy,

161(2):73–86, 2008.

[5] Roberto S Faranda, Hossein Hafezi, Sonia

Leva, Marco Mussetta, and Emanuele Ogliari.

The optimum pv plant for a given solar dc/ac

converter. Energies, 8(6):4853–4870, 2015.

[6] MR Hajmohammadi, P Aghajannezhad, SS

Abolhassani, and M Parsaee. An integrated

system of zinc oxide solar panels, fuel cells,

and hydrogen storage for heating and cooling

applications. international journal of hydrogen

energy, 42(31):19683–19694, 2017.

[7] Saeed, S., et al. "Intelligent monitoring and

cleaning of solar panels using Internet of

Things and machine learning: A review."

Renewable and Sustainable Energy Reviews

129, 2020.

[8] Atsushi Kajihara and AT Harakawa. Model of

photovoltaic cell circuits under partial

shading. In 2005 IEEE International

Conference on Industrial Technology, pages

866–870. IEEE, 2005.

[9] Hussein A Kazem, Miqdam T Chaichan, Ali

HA Al-Waeli, and K Sopian. Evaluation of

aging and performance of grid-connected

photovoltaic system northern oman: Seven

years’ experimental study. Solar Energy,

207:1247–1258, 2020.

[10] Antonio Luque and Steven Hegedus.

Handbook of photovoltaic science and

engineering. John Wiley & Sons, 2011.

[11] Mohammad Reza Maghami, Hashim Hizam,

Chandima Gomes,Mohd Amran Radzi,

Mohammad Ismael Rezadad, and Shahrooz

Hajighorbani. Power loss due to soiling on

solar panel: A review. Renewable and

Sustainable Energy Reviews, 59:1307–1316,

2016.

[12] Singaravel, R., et al. "Artificial Intelligence-

Based Robotic System for Automatic Solar

Panel Cleaning: A Review." International

Journal of Precision Engineering and

Manufacturing-Green Technology 7.3, 2020.

[13] Amit Kumar Mondal and Kamal Bansal. A

brief history and future aspectsin automatic

cleaning systems for solar photovoltaic

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.45

Ahmad Al-Dahoud,

Mohamed Fezari, Ali Aldahoud

E-ISSN: 2224-3496

477

Volume 19, 2023

panels. Advanced Robotics, 29(8):515–524,

2015.

[14] Satish Patil and HM Mallaradhya. Design and

implementation of microcontroller based

automatic dust cleaning system for solar

panel. Int. J.Eng. Res. Adv.

Technol.(IJERAT), 2(01):187–190, 2016.

[15] Kumar, V., et al. "Application of Machine

Learning in Solar Panel Cleaning System: A

Review." International Journal of Applied

Engineering Research 13.16, 2018.

[16] Fabio Santoni, Fabrizio Piergentili, Gian

Paolo Candini, Massimo Perelli,Andrea

Negri, and Michele Marino. An orientable

solar panel system for NanoSpace craft. Acta

Astronautica, 101:120–128, 2014.

[17] Arash Sayyah, Mark N Horenstein, and Malay

K Mazumder. Energy yield loss caused by

dust deposition on photovoltaic panels. Solar

Energy, 107:576–604, 2014.

[18] Manoj Kumar Sharma and Jishnu

Bhattacharya. Dependence of spectral factor

on angle of incidence for monocrystalline

silicon based photovoltaic solar panel.

Renewable Energy, 184:820–829, 2022.

[19] Akash Kumar Shukla, K Sudhakar, and

Prashant Baredar. Simulation and

performance analysis of 110 kwp grid-

connected photovoltaic system for residential

building in india: A comparative analysis of

various pv technology. Energy Reports, 2:82–

88, 2016.

[20] Peng Zhang, Lifu Zhang, Taixia Wu,

Hongming Zhang, and Xuejian Sun. Detection

and location of fouling on photovoltaic panels

using a dronemounted infrared thermography

system. Journal of Applied Remote Sensing,

11(1):016026, 2017.

[21] A. Al Dahoud, M. Fezari and A. Al Dahoud,

"Automatic solar panel cleaning system

Design," 2021 29th Telecommunications

Forum (TELFOR), 2021, pp. 1-4, doi:

10.1109/TELFOR52709.2021.9653215.

[22] Aqel, D., Al-Zubi, S., Mughaid, A. et al.

Extreme learning machine for plant diseases

classification: a sustainable approach for

smart agriculture. Cluster Comput 25, 2007–

2020

[23] Singh, K., Singh, M. K., & Srivastava, R. K.

(2021). A comprehensive review on solar

panel cleaning techniques and its automation.

Solar Energy, 225, 177-201.2021

[24] Ail, S. S., & Kapse, S. S. (2019). Automatic

cleaning of solar panels using machine

learning. International Journal of Emerging

Trends in Engineering Research, 7(1), 200-

204.2019

[25] Santhosh Kumar, S., et al. "Dust detection and

cleaning of solar panels using decision tree

algorithm." International Journal of

Engineering and Advanced Technology 8.4S,

: 165-168, 2019.

[26] Huang, Y., et al. "A new machine-learning-

based cleaning algorithm for improving solar

panel efficiency." Applied Energy 246, : 323-

335, 2019.

Contribution of Individual Authors to the

Creation of a Scientific Article (Ghostwriting

Policy)

-Ali Al-Dahoud and Mohamed Fezari collaborated

to design and develop a sensor capable of accurately

measuring the voltage of both clean and dirty solar

panels, as described in section 3.

-Ahmad Al-Dahoud, applied a machine learning

algorithm to analyse the collected data, as outlined

in section 4.

Sources of Funding for Research Presented in a

Scientific Article or Scientific Article Itself

No funding was received for conducting this study.

Conflict of Interest

The authors have no conflict of interest to declare.

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

WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT

DOI: 10.37394/232015.2023.19.45

Ahmad Al-Dahoud,

Mohamed Fezari, Ali Aldahoud

E-ISSN: 2224-3496

478

Volume 19, 2023