Wireless Networks

https://doi.org/10.1007/s11276-018-01918-z (0123456789().,-volV)(0123456789().,-volV)

Using IoT and smart monitoring devices to optimize the efficiency

of large-scale distributed solar farms
Salsabeel Shapsough1 • Mohannad Takrouri1 • Rached Dhaouadi1 • Imran A. Zualkernan1

Ó Springer Science+Business Media, LLC, part of Springer Nature 2018

Abstract
This paper presents a novel IoT-based architecture that utilizes IoT hardware, software, and communication technologies to
enable real-time monitoring and management of solar photovoltaic systems at large scales. The system enables stake-
holders to remotely control and monitor the photovoltaic systems and evaluate the effect of various environmental factors
such as weather, air quality, and soiling. The system was implemented and evaluated in terms of network delay and
resource consumption. Message Queueing Telemetry Transport (MQTT) was used to facilitate wide-scale real-time
communication. The average network delay was found to be less than 1 s, proving the architecture to be ideal for solar and
smart grid monitoring systems. As for resource consumption, the evaluation showed the hardware to consume about 3% of
the panel’s output, while the application also utilized a very small percentage of the CPU. This led to the conclusion that
the proposed architecture is best deployed using low-cost constrained edge devices where a combination of IoT-based
paradigm, efficient MQTT communication, and low resources consumption makes the system cost-effective and scalable.

Keywords IoT
Solar photovoltaic monitoring
Smart renewable energy
Smart grid

1 Introduction categories: man-made influences such as the Photovoltaic

(PV) array size and configuration [6, 7], module configu-
Solar power is increasingly becoming an integral part of ration [8, 9], and panel orientation in relation to the angle
world-wide energy production. This can be attributed to a of sun incidence [10], and natural influences such as
combination of factors including the depletion of conven- atmospheric conditions [11], shading [12–14], and soiling
tional energy sources [1], mainly fossil fuels, and a world- [15]. Many of the described methods for mitigating
wide awakening to the magnitude of the consequences external influences require the ability to collect data from
greenhouse emissions pose on the planet’s vital resources remote solar farms, communicate it in real-time, as well as
[2]. This led to great leaps not only in photovoltaic-related store it for processing and analysis. Remote control and
technology, but also in the scale of utility-grade solar farms dynamic reconfiguration are also essential functionalities in
built every year, with global solar photovoltaic capacity several solutions where a reaction is required. PV moni-
totaling over 385 Gigawatt (GW) in the year 2017 [3]. In toring systems have appeared in literature as early as the
the light of such advances in scale, the study of solar panel late 90 s [16, 17]. Since then, several designs and imple-
performance [4, 5] is more than ever recognized as the key mentations exist for solar monitoring systems [18]. Pro-
to cost-effective implementation and operation. While posed designs utilize hardware and software architectures
increasing efficiency of photovoltaic cells has been an from computer applications, web applications, and industry
ongoing challenge for years, recent work puts equal ecosystems, and adapt them to the specific requirements of
emphasis on the effect that external factors can have on a PV monitoring system. However, the recent rise of
power output. Such factors can be grouped into two Internet of Things (IoT) technologies generated an influx of
software, hardware, and communication technologies
& Salsabeel Shapsough optimized for distributed smart systems such as smart
[email protected] health monitoring, smart cities, and smart education. Such
1
systems share considerable requirements with PV
American University of Sharjah, Sharjah, UAE

123
 Wireless Networks

monitoring systems such as scalability, reliability, and Such discussions served as a motivation for various
physical and computation constraints. In this work, we designs of PV monitoring and management systems for in-
present a novel architecture for real-time monitoring of situ and remote real-time monitoring and output and
large-scale solar power facilities inspired by the Internet of weather forecasting environments.
Things. The work presented here further explores and
evaluates the work first proposed in [19]. 2.2 Smart solar PV monitoring

PV monitoring systems often fall into one of three cate-

2 Background gories. The first type is found in studies that deal with
statistical analysis of data and short-term or long-term
2.1 Challenges in solar farms forecasting. In such systems, data acquired at PV panels are
sent to a proprietary desktop application, or most com-
Soiling is defined as the accumulation of matter such as monly LabVIEW, that performs data analysis and visual-
dust particles, snow, and sand on the surface of PV panels, ization [26–31]. Alternatively, several works opt for simple
effectively reducing the amount of irradiance absorbed by web-based applications for data retrieval and visualization
the PV panels, and thus significantly reducing the power [32]. Data from the multiple acquisition units is aggregated
output. While soiling is widely recognized as a hurdle to through a local gateway to a web server and database,
utility-scale solar power generation, especially for solar making data available online. The third category of sys-
farms located in remote harsh environments such as desert tems adopts industry standards designed for power plants
plains [20], mitigating the issue is not straight forward. The and generation stations [33], or utilize ecosystems like
cost of periodic cleaning can sometimes exceed the cost Supervisory Control and Data Acquisition (SCADA)
ensued by output loss [21]. The issue of soiling then [34–36]. Such systems commonly depend on edge devices
transforms into an optimization problem where the goal is equipped with a variety of sensors and actuators. Envi-
to minimize cleaning costs while maximizing power out- ronmental measurements such as solar irradiance, temper-
put. Collecting information about power generation under ature, humidity, dust level, as well as performance data
various soiling conditions is a vital step to solving that such output voltage and current are recorded at varying
problem. Such information can also be used to find the frequencies and sent to the control center/server. Wireless
ideal tilt angle of a solar panel array, and for sun tracking in protocols are usually preferred over wired protocols for
general. With the addition of remote-control mechanisms, convenience and ease of wiring, although several designs
configuration can be automated to maximize output in each of small-scale PV systems employ Ad-Hoc communication
region. over Universal Serial Bus (USB) or Ethernet.
Furthermore, power output prediction is considered by
many a key for successful integration of solar power into 2.3 Internet of Things in solar PV monitoring
the smart grid. Power intermittency is often the main
hurdle in the face of renewable energy smart grid as it can The recent rise of Internet of Things produced an abun-
cause voltage fluctuations and flicker, increase instability in dance of new software, hardware, and communication
the grid, and lower power quality [22]. Collecting infor- technologies that have been designed to deliver high per-
mation regarding the operation of solar panels over a long formance in large-scale constrained environments. From
period of time can enhance active and reactive control low-cost System-on-Chip (SoC) and embedded computers
methods, thus enhancing renewable energy quality in the such as Arduino and Raspberry Pi to communication pro-
grid. tocols such as Message Queueing Telemetry Transport
Mitigating output intermittency due to other natural (MQTT) protocol, such technologies have been adopted by
influences, on the other hand, is a different problem. While various applications in education, health, transport, and
there is no direct solution that targets the inconvenience environment monitoring. Similarly, a recent trend in the
caused by phenomena such as shading, temperature, or field of PV monitoring has been the adoption of Internet of
humidity [23, 24], long-term remote monitoring of a solar Things (IoT) concepts and technologies [37] for building
farm can provide an insight into the effects of such phe- such systems. An IoT system consists of numerous, geo-
nomena. PV performance as well as ambient environment graphically spread nodes that interact with their surround-
measurements can be used to build models that can be used ings, and send information back to a server periodically
to generate short-term and long-term predictions for and/or based on a certain event. This information is either
expected power output at a certain moment in the future provided to stakeholders in real-time or stored to be
[25]. available on demand. Low-cost, scalability, reliability, and

123
Wireless Networks

security of IoT technology allows one to build very large 3.1 Timeliness
systems.
IoT systems often abide by a generic paradigm which End-to-end latency determines how fast the system can
governs the role of each class of technologies and the types react to events such as sudden changes in power level or
of interactions that take place in the system. This paradigm system faults, within a maximum latency requirement. In
consists of five layers: perception layer, network layer, an isolated solar monitoring system, maximum latency
middleware layer, application layer, and business layer. values can be determined by administrators, and is based
The perception layer is represented by sensing and actu- on the risk factor of such events as well as cost of related
ating elements at the panel level, while the network layer is assets. Once the system is integrated into a smart grid
represented by the wireless intranetworking within a solar ecosystem, latency values should abide by lowest maxi-
farm, and the internetworking between various nodes in the mum latency requirements across the whole ecosystem.
system. The middleware layer hosts technologies respon-
sible for translating raw data into intelligible information, 3.2 Scalability
as well as storage and dissemination of information.
Finally, the application and business layers represent Wide-scale implementation of solar monitoring systems
interested stakeholders. While interfaces used by clients entails that the infrastructure is capable of meeting per-
and researchers fall under the application layer, operations formance requirements regardless of continuous increase in
such as billing fall within the business layer. IoT systems scale. This is achieved by employing components that can
are usually large-scale, distributed, and heterogeneous, scale horizontally and perform load distribution. It also
which is why technologies made for IoT are designed to be applies to software, hardware, and communication tech-
light-weight, scalable, and interoperable. These criteria nologies that are designed to expect and properly manage
apply not only to communication technologies, but also to numerous consequent requests.
hardware, software applications, and storage.
The value that IoT technologies can bring to remote 3.3 Evolvability
monitoring of Solar PV systems has been recognized in
several works which suggested the incorporation of IoT In a large-scale distributed system, change in nodes,
elements in the field. Most such proposals cater to the edge especially the middleware layer, should be invisible to
of the IoT system where low-cost edge sensing and pro- lower and higher layers in the system. As it is the case with
cessing technologies such as Raspberry Pi, Arduino, Bea- most systems, services in a solar monitoring system are
gleBones, and others are used to acquire readings related to subject of change. As the system evolves in scale and
solar panel’s power generation efficiency, as well as the complexity, new application servers, databases, or analyt-
surrounding environment [38–40]. Edge devices often act ics engines may be incorporated. System architecture
as nodes in mesh networks, connected by short-range should be designed so that no downtime occurs due to a
wireless technologies such as ZigBee and Bluetooth. One service modification.
shortcoming of the proposed systems is that beyond the
edge, large-scale operations such as dissemination and 3.4 Heterogeneity
management of data across the whole system are still
performed by classic internet technologies which are not Solar monitoring systems feature an array of sensors and
the most suited for this purpose [41]. The architecture actuators that vary in complexity and data type and format.
presented here caters for scalable, almost-real-time moni- Analog and digital sensors, as well as sensors of varying
toring of Solar PV systems using a full IoT stack. resolutions can exist in a single edge node. Not only should
the edge hardware support such sensors, but device virtu-
alization should be designed so that data representation is
3 System requirements independent of hardware characteristics.

The generic IoT-based solar monitoring architecture pre- 3.5 Durability

sented here aims to satisfy several functional and non-
functional requirements derived from IoT systems as well Solar farms are often located in harsh environments with
as smart grid. Those requirements indicate how well an varying weather conditions such as very high or very low
architecture makes use of functionalities offered by IoT temperatures, humidity, and dust. Hardware, especially
technologies, as well as how seamlessly it can be integrated edge devices, should be selected with an account of such
with existing smart grid ecosystems. adversaries.

123
 Wireless Networks

3.6 Low cost

A key driver behind rapid increase in IoT research is the

challenge to develop new technologies that are signifi-
cantly less-costly than existing technologies, without
compromising on functionality. This includes low-cost
microcontrollers and sensors, and software and communi-
cation technologies that are cheaper to buy and install and
consume significantly less resources, while maintaining a
minimum level of quality of service. The presented archi-
tecture makes use of such technologies as it is considered a
key aspect that allows a solar monitoring system to effi-
ciently and safely increase in scale.

3.7 Security

One of the main challenges facing both solar monitoring

and smart grid systems is cyber threats. Data flowing
through either type of systems is often sensitive informa-
tion that, if compromised, can put the system functionality
as well as its users at risk. Unauthorized access, eaves-
dropping, and hijacking of vital system components are just
few of the many threats facing an online power system
[42]. To prevent attacks, security measures should be
implemented at all layers. This includes authentication to
ensure that only legitimate users can access the system,
end-to-end encryption to provide privacy and confiden-
tiality, and access control to distinguish between various
roles and ensure that only authorized individuals are able to
access information and perform certain actions.

Fig. 1 System architecture

4 Proposed architecture
device is equipped with minimal hardware required to
communicate over a short-range wireless protocol such as
The proposed architecture, shown in Fig. 1, maps the
WiFi or Zigbee, with a network gateway that enables
components of a solar monitoring system to the IoT para-
access to Internet. Alternatively, 3G/4G or LTE modules
digm introduced previously in Sect. 2.3. The perception
can be attached to the edge device to provide direct com-
layer represents edge-level devices responsible for mea-
munication over cellular networks.
suring the output of PV panels as well as environmental
Data acquired at the perception layer is aggregated over
conditions such as ambience, air quality, etc. Such mea-
a wireless network to the middleware layer, where data
surements are taken using a variety of smart sensors that
processing, management, storage, and communication
interface with an edge computing element. Sensors include
technologies convert raw data into intelligible information
temperature, humidity, solar irradiance, and dust sensors,
that can then be consumed by various applications. The
as well as panel current and voltage sensors, among others.
middleware layer houses technologies such as messaging
The choice of edge device usually depends on the com-
brokers for data aggregation, conventional databases
plexity of computation that the device is expected to per-
including SQL and NSQL databases, big data infrastructure
form. While a simpler microcontroller such as Arduino
like Hadoop, analytics engines, and web application ser-
Uno or ESP8266 is enough to acquire sensor readings and
vers. Information leaving the middleware layer is ready to
aggregate them over a wireless network, an edge computer
be consumed through various application layer software
such as Raspberry Pi or BeagleBone offers the possibility
interfaces such as computer application, web browser, and
of performing more sophisticated tasks at edge-level such
mobile applications by users such as system administrators
as device authentication and edge analytics. The edge
and clients. Information can also be fed into a larger

123
Wireless Networks

ecosystem such as smart grid systems, as well as business where any node can initiate one-to-one or one-to-many
and billing services. Throughout the system, information communication at any moment of time [43]. The broker is
flows bottom-up in PV and environment monitoring tasks, responsible for queueing messages from publishers and
and top-down in PV control applications, seamlessly. maintaining a list of subscribers that the messages are
Additionally, machine-to-machine (M2 M) low-level and immediately routed to. Messages are routed based on single
high-level protocols such as ZigBee allow for horizontal or multi-level topics. Data sources can publish to any topic
communication between nodes in a single layer, commonly at any moment of time, while subscribers receive messages
used at the perception layer in edge-computing applica- only from specific topics or levels of topics that they
tions, and among data management elements at the mid- subscribe to. In the context of a solar monitoring system,
dleware layer. this makes it possible to dynamically add and remove
publishing edge devices from the system without the need
to perform any modification at the server side. On the other
5 Communication architecture hand, depending on the number and type of services pro-
vided by the middleware layer, a single server can
Messages originating from edge devices flow through the dynamically change its subscription to start or stop listen-
various layers in the system over secure Message Queuing ing to a certain topic, without affecting any of the pub-
Telemetry Transport (MQTTs), while classic internet pro- lishers or subscribers to that topic. The broker is also able
tocols such as secure HyperText Transfer Protocol to retain the last known message from a device, as well as
(HTTPs) are used exclusively at the application layer be programmed to send a notification in case of a device
between web servers and clients. The dependence on going down -known as the Last Will message.
MQTT to handle most of the communication is driven by
the nature of communication in the system; where the 5.2 Data efficiency
number of clients generating messages can go up to thou-
sands, and the messages are small but high in frequency. In terms of hardware and network resources, MQTT proves
On the receiving end, messages are delivered to one or to be highly efficient [44]. MQTT clients consume less
more servers to be processed, stored, analyzed, or streamed power, active memory, and computational resources com-
in real-time. Real-time steaming of data to end users is pared to HTTP, making the protocol an ideal choice for
done using MQTTs over WebSockets. Unlike native deployment on low-resource edge devices. Furthermore,
MQTTs which is originally designed for M2 M commu- MQTT also consumes significantly less bandwidth. By
nication, MQTT over WebSockets is designed to be com- reducing the header size and overhead, MQTT can send the
patible with web browsers and user applications. The same information as HTTP using fewer bytes by supporting
communication architecture is shown in Fig. 2. connection-oriented communication. The communication
sequence that takes place when an edge device publishes a
5.1 Data flow reading via each protocol is shown in Fig. 3.
Furthermore, in a scenario where an edge device is
Unlike classic HTTP, MQTT is not based on a client-server streaming real-time readings to a user monitor, the size of
topology where exchange of information takes place in the the added header can significantly affect the bandwidth. A
form of a series client requests server responses between scenario with a message sized at 181 Bytes was sent in a
two nodes. Instead, an event-based messaging broker acts simple publisher/subscriber scenario on both protocols.
as the mediator in a network of publishers and subscribers, The resulting number of bytes sent by each protocol during
the process (excluding the SYN/ACK messages in HTTP)
are shown in Fig. 4. Using HTTP, a total of 531 Bytes were
exchanged between the publisher and the server, and 386
Bytes between the server and the subscriber. Using MQTT,
on the other hand, 248 Bytes were exchanged between the
publisher and broker, and between the broker and sub-
scriber. In total, MQTT generated around 46% less over-
head than HTTP. In a real-life implementation, less
overhead is a major advantage in remote settings where
network resources are limited, while data is abundant.

Fig. 2 Communication architecture

123
 Wireless Networks

Fig. 3 Communication sequence performed by a publisher via a HTTP and b MQTT

Fig. 4 Total message size

during a real-time stream on
a HTTP b MQTT

5.3 Distributed processing 5.4 Ease of addressing

A key aspect of an MQTT-Based architecture is that the Every published message in MQTT includes a topic fol-
system supports the option of distributed servers, allowing lowing the format of a URL. For example, a PV panel
for load balancing in large-scale systems as more resources located in ‘‘Region 1’’ of the solar farm and identified by
can easily be integrated into the system. Alternatively, the identifier ‘‘Panel 1’’ tags its messages with ‘‘/re-
different servers can be developed to host different types of gion_1/topic_1’’. Subscribers can choose to subscribe to
applications, and connect to different analytics engines, any topic by sending a one-time subscription request.
making the system not only highly scalable, but also Subscribers can subscribe to one panel (/re-
flexible. On the other end of the architecture, individual gion_1/panel_1), all panels in a region (/region_1/*), or all
solar panels and entire solar farms can easily be added and panels in the system (/*). The ‘‘*’’ character acts as a
removed with minimal disturbance to other part of the wildcard and allows subscribers to subscribe to numerous
system. This is yet another advantage of publish/subscribe publishers with a single request. It is also possible for
architectures, as the subscribers are separated from the multiple panels to publish to one topic, e.g. ‘‘re-
publishers by the broker, which acts as a mediator between gion_1/dust’’, or for subscribers to subscribe to sensor-re-
the two (Fig. 5). lated topics, e.g. ‘‘/*/dust’’. This enables subscribers
interested in certain information, such as dust accumula-
tion, for example, to receive updates any new panels added

123
Wireless Networks

Fig. 5 Communication
architecture

to the system, without the need to pre-register the panel at solar monitoring applications, commands that connect or
the subscribers. disconnect a part of the system in case of an emergency are
sent with QoS 2. QoS 1 is in the middle and is suitable for
5.5 Quality of service real-time applications with tolerance for duplicate packets.
Messages such as real-time power generation and weather
MQTT supports three levels of Quality of Service (QoS) in a solar farm are sent with QoS 1.
[45]. The lowest level is QoS0, which is referred to as
‘‘best effort’’ or ‘‘fire and forget’’. If QoS is set as 0, a
publisher publishes a message once at most. The publisher 6 Proof of concept
does not expect an acknowledgement and discards the
message immediately after the first publishing attempt. As Based on the proposed architecture, a prototype of a solar
for QoS 1, the publisher publishes the message as many monitoring system was designed, implemented, and eval-
times as possible until the subscriber receives it at least uated. The prototype consisted of four unique wireless-
once. The message remains stored in the publisher’s queue enabled edge devices equipped with a variety of sensors
until at least one acknowledgment is received. With QoS 2, and actuators. The edge devices published data over Wi-Fi
a message is delivered exactly once. The publisher sends to a middleware infrastructure consisting of an application
the message once and waits for an acknowledgement. If an server and a database engine. A web application was also
acknowledgement is received, the publisher discards the developed to allow users to monitor and interact with the
message from the queue. If an acknowledgement is not system remotely and in real time.
received within an acceptable time period, however, the
publisher resends the message with a flag indicating that it 6.1 Hardware implementation
is a duplicate. While QoS 0 is suitable for applications with
constrained bandwidth and tolerance for data loss, QoS 2 is Two categories of Wi-Fi-connected Raspberry Pi-based
the highest level of service, and is suitable for applications edge stations were designed, implemented, and evaluated.
where it is critical that only one copy of the message is The first category are normal edge devices, ‘‘PV stations’’,
received, and a duplicate may create an issue. In smart that are responsible for collecting data and relaying

123
 Wireless Networks

commands to control elements like relays dynamically client for the sole purpose of publishing this value. A
controlling the configuration of solar panels. PV stations description of each station’s features is given in Table 1.
are the most common in large-scale solar farms and typi- Figure 7 shows the three versions of the PV Stations. The
cally exist in large numbers. Based on design requirements, PV stations are self-sufficient in terms of power in that the
a single PV station can collect and broadcast data from an batteries that power the stations are charged using the
individual solar panel, or multiple panels that are con- 100-Watt solar panel each station monitors. All commu-
nected inline. In the prototype implementation, three PV nication between edge stations and the rest of the system is
station performed monitoring at individual panel-level, carried out over Wi-Fi.
each equipped with a unique combination of sensors and
actuators to provide different types of monitoring and 6.2 Back-end infrastructure
control. Current and voltage sensors were used to monitor
the performance of panels, while temperature, humidity, At the back-end, an infrastructure that consists of an
and solar irradiance sensors were used to quantify the application server, a database engine, and video streaming
panels’ surrounding environment. Actuators such as digital server work together to provide data management and
switches were used to dynamically change the configura- maintenance services, as well as web application hosting.
tion of the hardware. For example, it is possible to remotely The application server is built using NodeJS?Express and
connect or disconnect panels from charge controls, switch is hosted on Amazon Web Services (AWS), which pro-
edge devices to operate only on battery, or perform more vides a mix of a Platform as a Service (PaaS) and Infras-
complex switching such as routines used in IV curve tructure as a Service (IaaS). The software-level
tracing using a capacitive load [46]. Each PV station communication architecture is shown in Fig. 8. PV read-
publishes periodic sensor readings at a rate set by the ings, system status, as well as user and edge device
system administrator, which can be also changed information are stored in DynamoDB, a document-based
dynamically. NoSQL database hosted on AWS. Unlike SQL databases,
The second category are specialized edge devices DynamoDB stores data entries in JSON format rather than
responsible for surveillance. These edge devices are columns. This allows for seamless storage and manage-
equipped with a video camera for real-time surveillance. ment of heterogeneous information where entries from
Surveillance stations are less common than PV stations, different edges may differ in the names, number, and type
and fewer are required in a solar farm. Surveillance cam- of fields based on sensors and actuators attached to the
eras enable administrators to observe anomalies like edge device. Data is analyzed through R, an analytics
shading, etc. For example, an irregularity was noticed in engine capable of performing a variety of machine learning
solar irradiance data at around 09:00 am where the irradi- and deep learning algorithms. The Surveillance station’s
ance value experienced a sudden jump. Reviewing the video streaming and storage of streamed videos for later
surveillance video (Fig. 6) revealed that the shading was observation is made possible using Youtube’s livestream-
due to the sun rising from behind the neighboring build- ing service. Eclipse’s IoT broker is used to enable secure
ings. Video data that can also be used for image processing MQTT communication between edge devices and the
in research related to weather and soiling effects on solar middleware infrastructure.
panels. Finally, while PV stations send monitored values in The NodeJS server acts as an MQTT subscriber to all
text form, surveillance stations stream video in real-time. edge devices at all levels. This allows it to receive all data
In addition, for troubleshooting and maintenance purposes, in real-time, store it in DynamoDB, process and analyze it,
all stations publish the internal temperature of the micro- and make decisions on when and how information is pre-
controller. Surveillance station runs a separate MQTT sented to users. The main server’s Express interface

Fig. 6 Surveillance cam

screenshots at a 08:45 and
b 09:15am, showing the shading
effect

123
Wireless Networks

Table 1 Edge Nodes

Surveillance station PV station 1 PV station 2 PV station 3

Purpose Surveillance Solar irradiance and PV Solar irradiance and PV output monitoring and IV
output monitoring weather monitoring characteristics measurement
Main RPi Camera Pyranometer Pyranometer 2 isolated voltage ? current sensors
components RPi internal 3 voltage sensors Ambient temperature Temperature sensor
temperature sensor 3 current sensors sensor Relay modules
RPi internal temperature Humidity sensor RPi internal temperature sensor
sensor RPi internal temperature
sensor
Report 1/minute (temp. only) 1/second 1/minute 1/second
frequency
Data type Video stream JSON JSON JSON
Average 508 Bytes 189 Bytes 346 Bytes
packet size

Fig. 7 Edge stations: a PV station 1, b PV station 2, c PV station 3

exposes a variety of functionalities: a real-time map provide data protection and privacy, as well as data
showing locations and statuses of all panels registered in integrity.
the system, a standalone solar irradiance interface (Fig. 9),
a real-time monitoring, control, and surveillance interface
for individual panels (Fig. 10), a history interface that 6.3 Data format and advertisement
allows access to archived data logs from specific dates
(Fig. 11), an export interface that allows users to download Each new PV station added to the system advertises its
full data from any panel for any period of time, and an values by publishing a JSON object as text over MQTT,
analytics interface reserved for future addition of an ana- which the server uses to register the PV station.
lytics engine. Access privilege levels can range from casual Through advertisement, PV stations can be dynamically
users who can only view the real-time solar irradiance, to added, removed, or modified in the system without the
high level administrators who are authorized to send con- need to perform any system modifications. The server
trol commands to the panels, view analytics, and access the stores and updates the PV station’s profile in the data-
surveillance camera. Information traveling through the base and uses the profile to properly display real-time
system is secured at all layers using Transport Layer and log values to the user. An example of an advertised
Security (TLS) [47], which uses public key encryption to PV station profile is shown in Fig. 12. Once the new PV
station has been added to the system, it can start

123
 Wireless Networks

publishing the readings denoted in its profile, accom- 7 Evaluation

panied by a corresponding timestamp. A sample of a PV
station message is shown in Fig. 13. Of all system layers, the edge layer presents the biggest
challenge when it comes to scaling costs. This is mainly
due to the large population of the edge devices in a solar
farm, as well as the geographical location and distribution.
First, in remote solar farms, the most direct approach to
power monitoring edge devices is to allow the devices to
draw current from the solar panels, thus effectively
reducing the total generated output. It is therefore key that
the edge devices consume as little of the generated power
as possible. Another concern is network bandwidth. When
it comes to wireless communication between edge devices
and other nodes in the system, the choice of wireless pro-
tocol may vary based on factors such as available infras-
tructure. While it is possible to provide individual GPRS,
3G, 4G, or Wi-Fi connections for each device, it is some-
times less costly to configure devices to form a mesh where
only one device functions as a gateway. In the latter case,
protocols with lower data rate such as Bluetooth Low
Energy (BLE) or Zigbee may be used within the mesh. In
general, the less bandwidth that a single edge device
Fig. 8 Software-level communication architecture

Fig. 9 Solar irradiance over a full day

Fig. 10 Real-time interface for PV station 2 showing humidity and temperature

123
Wireless Networks

Fig. 11 History interface showing solar irradiance log from PV station 2 on May 24th, 2018

the previous discussion, assessing the performance and

feasibility of operating the edge layer focuses on three
aspects: power consumption, network bandwidth, and
computation power.

7.1 Experimental setup

The experiments were performed using three edge devices:

PV stations 1 and 2, surveillance station, and a bare
Raspberry Pi. The evaluation was carried out in three
Fig. 12 Profile JSON object advertised by PV station 002 stages. The first stage’s aim was to compare the perfor-
mance of stations with different peripherals and modes of
operation. In the first experimental setup, all three edge
stations were ran as they would normally while resource
consumption, internal temperature, and network traffic
were monitored. The second stage explored the resource
consumption when using different wireless protocols at the
edge. For this part, only PV station 3 was used in order to
eliminate the influence of varying hardware components
and frequency. The network profiles for GPRS, 3G, 4G and
Wi-Fi were simulated, while resource consumption and
Fig. 13 Sample of a message published from PV station 002 network performance were monitored. Packet loss per-
centage was also varied to simulate lossy networks. In the
requires for communication, the more options developers third stage, the publishing frequency was varied to examine
have when designing the wireless network infrastructure at its effect on resources consumption. This set of experi-
the farm, and the easier it becomes to choose the most ments was carried out on a separate Raspberry Pi than the
economic edge network topology. Finally, the cost of the ones used in the PV and surveillance stations. This was
edge device hardware is directly related to the complexity mainly to avoid the effect of maximum report frequency
of computation the edge device is required to perform. In for a single sensor on the frequency of publishing the
addition to being able to interface with all the sensors and readings. For example, the wattmeter used in PV station 3
actuators, the edge device is responsible for edge-level requires a total of 1 s to report the readings back to the
processing of data, communicating information to the Raspberry Pi. This made it not possible to experiment with
middleware, as well as receiving and executing commands messaging frequencies higher than 1 message/second.
that are relayed back from the user. The less processing Experimenting with a bare Raspberry Pi, however, made it
power that those tasks require, the less sophisticated the possible to vary the frequency and examine its direct
edge hardware can be; therefore, lowering the cost of influence on resource consumption.
implementing edge monitoring at a large scale. Based on

123
 Wireless Networks

The amount of power consumed by the edge devices station. This is most likely due to the number of sensors
was measured using YoctoWatt [48], which is an isolated and actuators hosted by PV station 3. In terms of compu-
USB wattmeter. Current and voltage values were sampled tational resources, on the other hand, video streaming uti-
at a rate of 1 Hz. Consumption of computational resources lizes almost twice the amount of CPU power compared to
such as CPU utilization and network interface write rate PV stations, and demands a far higher network write rate.
was measure using NMON [49], a Linux-based perfor- The MQTT round trip time, however, remains the same
mance monitoring tool. NMON frequency was also set at regardless of the hardware or the publishing frequency.
1 Hz. The internal temperature of the edge device was also The tradeoff between computational power of edge
measured as a test for device overheating and was mea- devices and the cost ensued by running them is a well-
sured using built-in internal temperature function once known one. However, in this case, the low percentage of
every minute. This is primarily since while power con- CPU utilization combined with a relatively-high power
sumption and CPU utilization can vary over seconds, consumption (ranging between 2 and 4% of the total panel
change in internal temperature is usually slower. Linux’s output) is an indicator that it is possible to reduce the cost
Traffic Control (TC) [50] was used to simulate various while maintaining reliable performance. The low utiliza-
network conditions by applying network filters at the RPi’s tion of Raspberry Pi’s processor means that an edge device
wireless interface in the second set of experiments. As for with lower processing power such as an ESP or an Ardu-
network, Wireshark [51] was used to measure network ino-based microcontroller is more than capable of handling
parameters such as bandwidth utilization, packet loss, the same task. In addition to smaller devices being com-
retransmissions, and total number of bytes exchanged. monly cheaper, they also require less power to operate,
Finally, MQTT round trip latency was calculated by thus saving on the power loss ensued by monitoring
measured the time delay between when an MQTT message devices. As for MQTT end-to-end delay, since the overall
is publisher by edge device, and when it is received by the goal is to allow the system to seemingly integrate with a
subscriber, which is the application server at the middle- Smart Grid, based on literature [52], ideal delay was found
ware layer. to be around 1 s. As shown in Table 2, the requirement was
met as the average end-to-end for all edge devices was
7.2 Results around 0.8 s. This is explored further in Table 3, where the
end-to-end MQTT delay was measured over various
The results of the first and second stages of evaluation are wireless communication protocols. Since communication
shown in Tables 2 and 3. In terms of power consumption, over MQTT is efficient in that it minimizes overhead, the
the station that consumed most power was found to be PV required bandwidth for publishing monitoring data at a rate
station 3, drawing around 70% more power compared to of 1 Hz is around 2Kbps. At the same time, even GPRS,
PV station 2, and 20% more compared to surveillance

Table 2 Summary of resources

Metric Surv. station PV station 2 PV station 3
consumed by edge devices (1 h)
Average power consumption # of samples 3600 3600 3600
Mean 3.142 W 2.237 W 3.737 W
Stdv. 0.165 0.0730 0.0246
Raspberry Pi CPU utilization # of samples 3600 3600 3600
Mean 4.824% 2.931% 2.708%
Min 2.1% 0% 0.3%
Max 16.2% 15.8% 28%
Raspberry Pi internal temp # of samples 1440 1440 1440
Mean 58.475 °C 56.464 °C 51.120 °C
Stdv. 6.277 6.509 2.156
Network write rate # of samples 3600 3600 3600
Mean 245.9 KB/s 2.6 KB/s 2.6 KB/s
Min 209.1 KB/s 0.1 KB/s 0.1 KB/s
Max 622.9 KB/s 29.2 KB/s 26 KB/s
MQTT round-trip latency # of samples 1440 1440 1440
Mean 0.802 s 0.800 s 0.801 s
Stdv. 0.787 0.818 0.806

123
Wireless Networks

Table 3 Network performance

Metric GPRS 3G 4G Wi-Fi
vs. wireless protocol
Typical bandwidth 50 Kbps 750 Kbps 4 Mbps 30 Mbps
# of samples 3600 3600 3600 3600
MQTT round-trip latency
Mean 0.963 s 0.948 s 0.946 s 0.855 s
Stdv. 0.433 0.528 0.491 0.710
Upload bandwidth utilization 2.103 Kbit/s 2.017 Kbit/s 2.015 Kbit/s 2.084 Kbit/s
TCP lost segment 0% 0% 0% 0%
TCP retransmission 1.10% 1% 1% 0%

which the most restrained protocol, supports a rate of at a rate of 0.2 Hz (1 message/5 s) increases average power
50Kbps. consumption by 0.01 W; an increase of 0.1%. The increase
It is therefore safe to assume that the end-to-end delay in power consumption remains small until frequency is
would not be affected by constrained network conditions. increased to 10 messages/second, where the higher fre-
This is further-confirmed when examining the latency quency causes the average to increase up to 1.424 W, or by
values under each wireless network protocol, where it did 5% compared to idle mode. However, taking into account
not exceed 1 s, thus complying with the delay requirement that a frequency of 10 messages/second or 50 messages/
discussed earlier. In addition to simulating the typical second is not neccesary at a wide scale and would only be
profile for each wireless protocol, network performance required in highly crticial components, the increase in
was also monitored for varying degrees of lossy networks. power consumption is considered acceptable. The same
The average end-to-end delay that occurs at each case is trend can be noticed in terms of CPU%, yet the maximum
shown in Fig. 14. As expected, end-to-end delay increases remains less than 7% of the Raspberry Pi’s processing
in lossy networks due to lost segments and re-transmissions power, even at the extreme case.
at the TCP level. However, at 25% packet loss, all net- Next, the amount of network write processes performed
works managed to achieve less than 2 s delay, with 4G and by the devices and the average 2-way bandwidth utilized
Wi-Fi achieving less than 1.5 s latency. were measured. The bandwidth accounts for not only
In the third stage, idle time between consectutive reports published messages, but also keep-alive messages and
was varied from 50 ms to 5 s. The cost in terms of average other connection maintinance packets that may occur in
power consumption and average CPU utiliziation is shown MQTT and TCP, in both directions. The network write
in Fig. 15. As noted earlier, a bare Raspberry Pi was used measurment, on the other hand, accounts for all wireless
in order to eliminate the delay caused by specific hardware communication that a device performs to stay online on a
components. This brought the average power consupmtion WiFi network. The results are shown in Fig. 16. The
to 1.357 W at idle state. Compared to idle state, publishing increase in number of network write commands that the

Fig. 14 End-to-end latency vs.

packet loss

123
 Wireless Networks

Fig. 15 Average power (watt)

and average CPU utilization vs.
idle time between reports

Fig. 16 Network resources vs.

idle time between reports

edge device must perform follows a similar trend to the of publishing does not impose any significant costs, and the
increase in CPU% and consumed power. Assuming 1 Hz to real frequency limitation is set by the hardware, e.g. the
be the ideal frequency, MQTT consumes around 2.61 Kbps maximum sampling frequency of a sensor. Since MQTT
of the network bandwidth. The edge device’s network has low resource requirements, it is possible to choose a
interface should be able to perform network writes at a rate microcontroller with less computational capabilities. Not
of 0.3 Kbps. As the frequency is increased to 1 message/ only would this reduce the cost of hardware, but it would
50 ms, the bandwidth consumed by MQTT increases to 27 also minimize the PV output power lost to edge devices.
Kbps, while the device performs writes at a rate of 5.5 Therefore, the proposed IoT-based architecture makes it
Kbps. possible to implement efficient real-time monitoring of
The evaluation results lead to the following conclusion: large-scale solar farms at a low cost.
an IoT protocol such as MQTT allows for efficient, light-
weight communication for edge devices. Since the protocol
is designed to minimize network overhead, it is possible to 8 Conclusion
achieve reliable, real-time communication over various
wireless protocols. The element that would most influence This paper proposed a novel architecture that utilizes
the cost of edge layer is the hardware. While the publishing Internet of Things (IoT) technologies for real-time moni-
frequency has insignificant influence on amount of power toring of large-scale distributed solar farms. The architec-
consumption, on average, it would understandably increase ture was derived based on a review of the latest trends and
the average CPU utilization and the total number of bytes challenges in solar monitoring, as well as latest break-
sent over the network within a period. Nonetheless, when throughs in IoT technologies. The architecture comprises
examining the weighted average of CPU utilization, i.e. the wireless-enabled edge monitoring devices that measure
CPU% utilized during the short period of time when the output power as well as ambient environment and conveys
edge device is working, there is no significant increase. it back to a data center over MQTT. At the data center,
Therefore, it is safe to assume that increasing the frequency middleware components such as databases, application

123
Wireless Networks

servers, and analytics engines process the data and converts Sustainable Energy Reviews, 49, 672–679. https://doi.org/10.
it into intelligible information that can then be presented to 1016/j.rser.2015.04.165.
7. Khoo, Y. S., Nobre, A., Malhotra, R., et al. (2014). Optimal
stakeholders over applications such as web interfaces and orientation and tilt angle for maximizing in-plane solar irradiation
mobile apps. A solar monitoring system based on the for PV applications in Singapore. IEEE Journal of Photovoltaics,
proposed architecture was implemented and evaluated. The 4, 647–653. https://doi.org/10.1109/JPHOTOV.2013.2292743.
evaluation focused on the cost of running a large-scale 8. Báez-Fernández, H., Ramı́rez-Beltrán, N. D., & Méndez-Piñero,
M. I. (2016). Selection and configuration of inverters and mod-
distributed edge layer, and assessed the consumption of ules for a photovoltaic system to minimize costs. Renewable and
parameters such as power, network bandwidth, and com- Sustainable Energy Reviews, 58, 16–22. https://doi.org/10.1016/j.
putational power required to operate the system at scale. rser.2015.12.067.
The evaluation found that not only was the system able to 9. Reshma Gopi, R., & Sreejith, S. (2018). Converter topologies in
photovoltaic applications—a review. Renewable and Sustainable
provide reliable, secure, and low-latency operation at a low Energy Reviews, 94, 1–14. https://doi.org/10.1016/j.rser.2018.05.
cost, but it is possible to reduce the cost further by sim- 047.
plifying the edge hardware, allowing for large-scale 10. AL-Rousan, N., Isa, N. A. M., & Desa, M. K. M. (2018).
implementation in Megawatt and Gigawatt-scale farms. In Advances in solar photovoltaic tracking systems: A review. Re-
newable and Sustainable Energy Reviews, 82, 2548–2569. https://
addition to allowing stakeholders to monitor the status and doi.org/10.1016/j.rser.2017.09.077.
operation of remote solar farms in real time, the system 11. Said, Z., Arora, S., & Bellos, E. (2018). A review on performance
provides tools to efficiently collect, store, and analyze data and environmental effects of conventional and nanofluid-based
over long periods of time. Such data is key to conducting thermal photovoltaics. Renewable and Sustainable Energy
Reviews, 94, 302–316. https://doi.org/10.1016/j.rser.2018.06.010.
studies related to solar PV panel performance deterioration, 12. Belhachat, F., & Larbes, C. (2018). A review of global maximum
weather forecasting, and the effects of environmental ele- power point tracking techniques of photovoltaic system under
ments such as soiling and shading on PV panels. This is partial shading conditions. Renewable and Sustainable Energy
achieved by utilizing various IoT communication, software Reviews, 92, 513–553. https://doi.org/10.1016/j.rser.2018.04.094.
13. Das, S. K., Verma, D., Nema, S., & Nema, R. K. (2017). Shading
and hardware technologies which have been inherently mitigation techniques: State-of-the-art in photovoltaic applica-
designed for large-scale distributed ecosystems, and tions. Renewable and Sustainable Energy Reviews, 78, 369–390.
therefore fit seamlessly into a large-scale solar power https://doi.org/10.1016/j.rser.2017.04.093.
system. By adopting a modular layer-based paradigm for 14. Ramli, M. A. M., Twaha, S., Ishaque, K., & Al-Turki, Y. A.
(2017). A review on maximum power point tracking for photo-
the system architecture and using IoT technologies, it is voltaic systems with and without shading conditions. Renewable
possible to seamlessly integrate distributed solar farms into and Sustainable Energy Reviews, 67, 144–159. https://doi.org/10.
existing ecosystems such as Smart Grid. 1016/j.rser.2016.09.013.
15. Andenæs, E., Jelle, B. P., Ramlo, K., et al. (2018). The influence
of snow and ice coverage on the energy generation from photo-
voltaic solar cells. Solar Energy, 159, 318–328. https://doi.org/
10.1016/j.solener.2017.10.078.
References 16. Mukaro, R., & Carelse, X. F. (1999). A microcontroller-based
data acquisition system for solar radiation and environmental
1. Islam, M. T., Huda, N., Abdullah, A. B., & Saidur, R. (2018). A monitoring. IEEE Transactions on Instrumentation and Mea-
comprehensive review of state-of-the-art concentrating solar surement, 48, 1232–1238. https://doi.org/10.1109/19.816142.
power (CSP) technologies: Current status and research trends. 17. Wilshaw, A. R., Pearsall, N. M., & Hill, R. (1997). Installation
Renewable and Sustainable Energy Reviews, 91, 987–1018. and operation of the first city centre PV monitoring station in the
https://doi.org/10.1016/j.rser.2018.04.097. United Kingdom. Solar Energy, 59, 19–26. https://doi.org/10.
2. Nuortimo, K., Härkönen, J., & Karvonen, E. (2018). Exploring 1016/S0038-092X(96)00123-5.
the global media image of solar power. Renewable and Sustain- 18. Rahman, M. M., Selvaraj, J., Rahim, N. A., & Hasanuzzaman, M.
able Energy Reviews, 81, 2806–2811. https://doi.org/10.1016/j. (2017). Global modern monitoring systems for PV based power
rser.2017.06.086. generation: A review. Renewable and Sustainable Energy
3. Renewable Capacity Statistics 2018. In: /publications/2018/Mar/ Reviews, 10, 5–20. https://doi.org/10.1016/j.rser.2017.10.111.
Renewable-Capacity-Statistics-2018. /publications/2018/Mar/Re- 19. Shapsough, Salsabeel, Takrouri, Mohannad, Dhaouadi, Rached,
newable-Capacity-Statistics-2018. Accessed 24 May 2018. & Zualkernan, Imran. (2018). An MQTT-based scalable archi-
4. Pandey, A. K., Tyagi, V. V., Selvaraj, J. A., et al. (2016). Recent tecture for remote monitoring and control of large-scale solar
advances in solar photovoltaic systems for emerging trends and photovoltaic systems. Canada: Niagara Falls.
advanced applications. Renewable and Sustainable Energy 20. Burton, P. D., & King, B. H. (2014). Application and charac-
Reviews, 53, 859–884. https://doi.org/10.1016/j.rser.2015.09.043. terization of an artificial grime for photovoltaic soiling studies.
5. Fouad, M. M., Shihata, L. A., & Morgan, E. I. (2017). An inte- IEEE Journal of Photovoltaics, 4, 299–303. https://doi.org/10.
grated review of factors influencing the perfomance of photo- 1109/JPHOTOV.2013.2270343.
voltaic panels. Renewable and Sustainable Energy Reviews, 80, 21. Zapata, J. W., Perez, M. A., Kouro, S., et al. (2015). Design of a
1499–1511. https://doi.org/10.1016/j.rser.2017.05.141. cleaning program for a PV plant based on analysis of energy
6. Malathy, S., & Ramaprabha, R. (2015). Comprehensive analysis losses. IEEE Journal of Photovoltaics, 5, 1748–1756. https://doi.
on the role of array size and configuration on energy yield of org/10.1109/JPHOTOV.2015.2478069.
photovoltaic systems under shaded conditions. Renewable and

123
 Wireless Networks

22. Hu, Z., Li, C., Cao, Y., et al. (2014). How smart grid contributes 36. Pasc, P.-C., & Dumitru, C.-D. (2016). scada system for solar
to energy sustainability. Energy Procedia, 61, 858–861. https:// MPPT controller monitoring. Procedia Technology, 22, 803–807.
doi.org/10.1016/j.egypro.2014.11.982. https://doi.org/10.1016/j.protcy.2016.01.052.
23. Liu, H., Nobre, A. M., Yang, D., et al. (2014). The impact of haze 37. Degener, S. (2016). The Internet of Things: Enabling intelligent
on performance ratio and short-circuit current of PV systems in solar assets. Renewable Energy Focus, 17, 136–137. https://doi.
Singapore. IEEE Journal of Photovoltaics, 4, 1585–1592. https:// org/10.1016/j.ref.2016.06.004.
doi.org/10.1109/JPHOTOV.2014.2346429. 38. Adhya, S., Saha, D., & Das, A. et al (2016). An IoT based smart
24. Huang, P., Zhao, W., & Li, A. (2017). The preliminary investi- solar photovoltaic remote monitoring and control unit. In 2016
gation on the uncertainties associated with surface solar radiation 2nd international conference on control, instrumentation, energy
estimation in mountainous areas. IEEE Geoscience and Remote communication (CIEC) (pp. 432–436).
Sensing Letters, 14, 1071–1075. https://doi.org/10.1109/LGRS. 39. Jihua, Y., & Wang, W. (2014). Research and design of solar
2017.2696973. photovoltaic power generation monitoring system based on
25. Creayla, C. M. C., Garcia, F. C. C., & Macabebe, E. Q. B. (2017). TinyOS. In 2014 9th international conference on computer sci-
Next day power forecast model using smart hybrid energy ence education (pp. 1020–1023).
monitoring system and meteorological data. Procedia Computer 40. Liu, G., Qiu, H., Zhu, L., & Chen, Y. (2017). Architecture and
Science, 105, 256–263. https://doi.org/10.1016/j.procs.2017.01. experiment of remote monitoring and operation management for
219. multiple scales of solar power plants. In 2017 IEEE 2nd advanced
26. Haider-e-Karar, Khuwaja, A. A., & Sattar, A. (2015). Solar information technology, electronic and automation control con-
power remote monitoring and controlling using Arduino, Lab- ference (IAEAC) (pp. 2489–2495).
VIEW and web browser. In 2015 power generation system and 41. Mashal, I., Alsaryrah, O., Chung, T.-Y., et al. (2015). Choices for
renewable energy technologies (PGSRET) (pp. 1–4). interaction with things on Internet and underlying issues. Ad Hoc
27. Wei, Z., Porter, J. R., & Morgan, J. A. (2014). Experiential Networks, 28, 68–90. https://doi.org/10.1016/j.adhoc.2014.12.
learning of digital communication using LabVIEW. IEEE 006.
Transactions on Education, 57, 34–41. https://doi.org/10.1109/ 42. Shapsough, S., Qatan, F., & Aburukba, R. et al. (2015). Smart
TE.2013.2264059. grid cyber security: Challenges and solutions. In 2015 interna-
28. Ya’acob, M. E., Hizam, H., & Radzi, M. A. M. (2014). Real time tional conference on smart grid and clean energy technologies
monitoring and analysis of tropical impact on PV performance (ICSGCE) (pp. 170–175).
based on LabVIEW architecture. Journal of Automation and 43. Yokotani, T., & Sasaki, Y. (2016). Comparison with HTTP and
Control Engineering, 2, 138–142. https://doi.org/10.12720/joace. MQTT on required network resources for IoT. In 2016 interna-
2.2.138-142. tional conference on control, electronics, renewable energy and
29. Ramirez, M., Muñoz, F., Diez, J. V., & Casas, C. (2012). Online communications (ICCEREC) (pp. 1–6).
monitoring system for stand-alone photovoltaic applications- 44. Naik, N. (2017). Choice of effective messaging protocols for IoT
analysis of system performance from monitored data. Journal of systems: MQTT, CoAP, AMQP and HTTP. In 2017 IEEE
Solar Energy Engineering, 134, 034502. https://doi.org/10.1115/ international systems engineering symposium (ISSE) (pp. 1–7).
1.4005448. 45. Banks, A., & Gupta, R. (2014). MQTT Version 3.1. 1. OASIS
30. Anwari, M., Dom, M. M., & Rashid, M. I. M. (2011). Small scale Standard. Accessed 20 Jul 2018.
PV monitoring system software design. Energy Procedia, 12, 46. Shapsough, S., Takrouri, M., Dhaouadi, R., & Zualkernan, I.
586–592. https://doi.org/10.1016/j.egypro.2011.10.079. (2018). Online monitoring and evaluation of PV soiling using a
31. Rezk, H., Tyukhov, I., Al-Dhaifallah, M., & Tikhonov, A. (2017). smart IoT-based IV tracer. In Presented in the international
Performance of data acquisition system for monitoring PV system conference on energy, water & environmental sciences
parameters. Measurement, 104, 204–211. https://doi.org/10.1016/ (ICEWES), Ras Al Khaimah, UAE.
j.measurement.2017.02.050. 47. Dierks, T. (2008). The transport layer security (TLS) Protocol
32. Papageorgas, P., Piromalis, D., Antonakoglou, K., et al. (2013). Version 1.2. https://tools.ietf.org/html/rfc5246. Accessed 19 Jun
Smart solar panels: In-situ monitoring of photovoltaic panels 2018.
based on wired and wireless sensor networks. Energy Procedia, 48. Yocto-Watt—Isolated USB wattmeter (AC/DC). https://www.
36, 535–545. https://doi.org/10.1016/j.egypro.2013.07.062. yoctopuce.com/EN/products/usb-electrical-sensors/yocto-watt.
33. Fuentes, M., Vivar, M., Burgos, J. M., et al. (2014). Design of an Accessed 15 Aug 2018.
accurate, low-cost autonomous data logger for PV system mon- 49. NMON for Linux. http://nmon.sourceforge.net/pmwiki.php.
itoring using ArduinoTM that complies with IEC standards. Solar Accessed 27 Jul 2018.
Energy Materials and Solar Cells, 130, 529–543. https://doi.org/ 50. NetEm. https://wiki.linuxfoundation.org/networking/netem.
10.1016/j.solmat.2014.08.008. Accessed 13 Aug 2018.
34. Govindarajan U, K. P., Ramachandaramurthy, V. K., et al. 51. Wireshark. https://www.wireshark.org/. Accessed 1 Aug 2018.
(2018). Integrating solar photovoltaic energy conversion systems 52. Kansal, P., & Bose, A. (2012). Bandwidth and latency require-
into industrial and commercial electrical energy utilization—a ments for smart transmission grid applications. IEEE Transac-
survey. Journal of Industrial Information Integration, 10, 39–54. tions on Smart Grid, 3, 1344–1352. https://doi.org/10.1109/TSG.
https://doi.org/10.1016/j.jii.2018.01.003. 2012.2197229.
35. Eseye, A. T., Zhang, J., & Zheng, D. (2018). Short-term photo-
voltaic solar power forecasting using a hybrid Wavelet-PSO-
SVM model based on SCADA and Meteorological information.
Renewable Energy, 118, 357–367. https://doi.org/10.1016/j.
renene.2017.11.011.

123
Wireless Networks

Salsabeel Shapsough received also held Visiting Scholar positions at Rice University, Houston,
her B.S. (2014) in Computer Texas, in 1998, 1999 and 2001, and at Tokyo University, Japan, in
Engineering and her M.S. 2013. He is currently a Full Professor of Electrical Engineering and
(2017) in Computer Engineer- the Petrofac Chair in Renewable Energy at the College of Engi-
ing from the American Univer- neering, American University of Sharjah, UAE. His research interests
sity of Sharjah, Sharjah, UAE. are in the areas of energy systems, motor drives, and intelligent
In 2017, she joined the Ameri- motion control systems, with applications to rolling mill drives,
can University of Sharjah as a mobile robots, and solar energy systems. Dr. Dhaouadi is a Senior
Visiting Scholar in the field of Member of the IEEE Industrial Electronics Society.
Internet of Things and smart
solar energy systems where she Imran A. Zualkernan (M’03) re-
is responsible for the design and ceived B.S. (high distinction)
development of solar power and Ph.D. degrees in computer
monitoring systems and tech- science from the University of
nologies. Her research interests Minnesota, Minneapolis, in
include Internet of Things, smart education technologies, and smart 1983 and 1991 respectively.
energy. From 1984 to 1990, he was a
Research Assistant with the
Mohannad Takrouri received his Knowledge Systems Laboratory
B.S. in Mechatronics Engineer- at the University of Minnesota.
ing from Palestine Polytechnic He was an Assistant Professor
University in 2013 and his M.S. with the Computer and Electri-
in Mechatronics Engineering cal Engineering Department,
from the American University Pennsylvania State University,
of Sharjah in 2015. From 2013 College Park from 1992 to
he worked as a research assis- 1995. He was a Principal Design Engineer with AMCS Inc. in
tant at the American University Chanhassen, Minnesota from 1995–1998. He was the Chief Executive
of Sharjah. Since 2016 he is a Officer of Askari Information Systems from 1998 to 2000, and the
robotic researcher at Ibtikar Edu Chief Technology Officer for Knowledge Platform, Inc., Singapore
Tech solutions. His research from 2000 to 2003. In 2003 he joined the American University of
interests are in the areas of Sharjah in the UAE where he is currently an Associate Professor in
motor drives, friction identifi- Computer Science and Engineering. He is the author of more than 170
cation and compensation, and articles. His research interests include consumer systems, sensor-
solar energy systems. based Internet applications, Internet of Things (IoT) and application
of big data to IoT.
Rached Dhaouadi received his
Ph.D. in Electrical Engineering
from the University of Min-
nesota in 1990. From 1990 to
1994 he worked as a Visiting
Researcher with the Hitachi
Research Laboratory, Hitachi,
Ltd., Japan, where he was
engaged in the design and
development of motor drive
systems for rolling mills. In
1994, he was a Visiting
Researcher at Trondheim Insti-
tute of Technology, Trondheim,
Norway. From 1994 to 2000 he
was with the Polytechnic School of Tunisia, University of Tunis. He

123