You are on page 1of 9

1

An IoT Based Smart Energy Management of HVAC


System
Mohammed Rashid, Mazin AliKarar, Ragini Gupta, Dr. Imran Zualkernen
American University of Sharjah, UAE

advanced technological concepts that can be used in this field


Abstract—Today, the increasing cost of energy has let many are Big Data and Business Analytics. These two concepts
organizations to come forward with smart ways for monitoring, coupled with Internet of Things technology will be playing an
controlling and saving energy. A smart Energy Management important role for the benefit of big enterprises and small
System (EMS) can contribute in cutting down the cost through
suitable power saving and data analytics techniques which can also
growing business. The IoT technology can provide a ubiquitous
help in reducing carbon emission, thereby conserving the computing platform to monitor and track the activities of
environment. With the rampant rise of Internet of Things different households on a massive scale. Big data provides the
technology, the IoT platform can help revolutionize energy system with the ability to collect huge amounts and wide variety
management for home/office utilities, energy service providers, of data in a very high frequency from the surroundings. This
and end-customers. The IoT platform can address a vast range of data is collected using a large number of different wireless
energy management services perpetuated by millions and trillions
sensors distributed in the environment. The data will be
of connected sensors and other devices. The mesh network of
connected devices, network, and wireless transmission will lead to captured and stored in a central scalable repository unit for
a large accumulation of data in the energy sector. Thus, in order analysis. This is where the analysis part comes using business
to accomplish a holistic domain of energy data, we intend to analytics engines. Data Analytics plays an integral role in
develop a big data driven smart energy management system. The optimizing energy management decisions by aiding different
system prototype can be applied to different energy sectors; institutions and authorities to gain sensible insights from the
residential, industrial and commercial for a variety of white goods.
large volumes of collected data. The data will be analyzed and
To scope the analysis domain, we will imitate the system for HVAC
(Heating, Ventilation and Air Conditioning) units in UAE. Since monitored using predictive analysis and advanced methods to
air conditioning contributes to 60% of Dubai’s electricity generate valuable and useful information It can help the
consumption, the team intends to apply the system design on a stakeholders to study data usage patterns of different devices in
country level accommodating different communities in UAE. buildings, and generate scorecards for each device to make
Thus, in this paper, we present a comprehensive work on the intelligent decisions. The authorities can use this information to
design and implementation of an assistive platform for smart
energy management of HVAC units using Internet of Things
identify where the power is consumed inefficiently and
technology and Data analytics. implement an energy saving process towards it. This results in
greater operational efficiency and advanced failure detection. It
Index Terms— Internet of Things, Data Analytics, Energy also optimizes the system and reduces the energy waste which
Management System, HVAC leads to decrease in bill costs.
Such an arrangement of energy management systems could
bring benefits for both, costumers and energy companies. The
I. INTRODUCTION energy companies can use the power consumption data
available with analytics engines to provide flexible and on-
R Educing energy and power consumption is an issue on
which big researches are being conducted for a long time.
Using the energy efficiently saves money, increases energy
demand supply and beneficial marketing strategies. The power
customers, being aware of their consumption behavior and
security in the future and reduces carbon footprint at large. having a close interaction with the electricity company, they
Thus, the demand for smart energy management is rapidly can be able to adjust and optimize their power consumption and
reduce their electricity bills.
increasing in the domain of smart homes and smart cities.
Smart energy management systems can incorporate a vast
However, the lack of low cost, easy to deploy, and low
number of power consuming machines and devices. However,
maintenance technology has restricted the large scale considering the highest power consumption due to centralized
deployment of such systems. According to the 2011 census, HVAC systems in buildings, the team decided to focus on
around 250,000 buildings exist in different emirates of UAE. energy management of HVAC units. In order to have an
This number is expected to increase by 20% by the year effective-cost saving energy management system, it is
2030[1]. Accommodating the increasing number of buildings important to control the operation of these HVAC systems
without compromising on energy efficiency is a critical issue. depending on the aggregate power consumption over the years,
The sheer quantity of data collected throughout different cities the peak power consumption, the weather/atmospheric
of a country is envisaged to present multiple challenges in data conditions and the consumption slab rates. This is where the
storage, organization and analysis. Having these issues, many combination of Internet Of Things technology and Big Data
researches are being carried out trying to use the advanced analytics comes into play for implementing energy
technologies for achieving the optimum use of power. Two management solutions on a local scale to national level at large.
2

An example of this setting can be United Arab Emirates. The occupants by using RFID and IR sensors. In [10] Akkaya et al.
homeowners developed an occupancy monitoring mechanism using RFID,
sensors and camera. The system collect large amount of data on
Thus, the novelty of our system lies in its state of art localization of people. This data is used to improve energy
architecture to provide an integrated platform for smart energy efficiency of buildings. The HVAC control is performed by
management of different stakeholders ranging from the home applying 4 stages: occupancy detection, occupancy counting,
owner customers to utility providers. Moreover, the system can occupancy tracking, and behavior recognition.
extend its application to monitor power consumption of any After reviewing the literature, the team found that
machine or device besides the HVAC units. there many issue not solved yet. The first one is the Lack of
standard which is an important issue that needs to be addressed
II. PREVIOUS WORK immediately to enable interoperability and integration of IoT
In the research done by Kim el al. [2], they developed a systems. Consumption of resources is another area ignored by
Business Platform for demand side management for energy. It most developers, the impact of which will be visible as the
is basically a cluster of distributed energy resources which are amount of instances increase. Security wise, most IoT
controlled centrally. The system is constructed of 2 big data implementations lack proper security measures. Security must
clusters: one is Hadoop and the other one is Spark installed. The be implemented starting from the physical layer to the
energy usage data is collected every 15 minutes and then saved application layer. Thus, IoT systems being developed should
into a relational database. The queries can be executed in follow a standard, be efficient in consumption of resources and
parallel to reduce time. In [3], Lee et al. built a system for be resilient against cyber-attacks.
storing, indexing, and searching for huge-scale energy usage
data. Their implementation is a proof-of-concept that big data III. PROPOSED Solution
technologies can be effectively applied to the large-scale energy An efficient energy management system should ensure
domain. Their system consists of: a cluster infrastructure, monitoring of building atmospheric conditions, equipment
automatic data-collection, data loading to a distributed file status, bill tracking utility, and record generation [11]. To
system, a distributed database for managing large-scale power render these services, four divisions of business processes are
data, and data access APIs. The distributed database is used for considered:
storing the big data of power measurements. Bergelt, Vodel and  Consumption Analysis for Monitoring: Analytical
Hardt [4] introduced a big data implementation for wireless techniques can help identify different areas of power
sensor networks. They relied on an assumption that the network consumption discrepancies across different buildings over
have a huge number of sensors, consequently, the collected data a span of time.
will be enormous. They used a combination of database-  Asset Efficiency Analysis: The efficiency of an HVAC
orientation and wake-up technology to rise their system unit in a building can largely depend upon the age of the
efficiency in terms of energy and power consumption. Zhou el building or environmental conditions. Using different
al. in [5] presented a digitized system for improving energy benchmarking techniques, ideal efficiency levels for a
consumption efficiency. They used a big data paradigm where device can be computed and compared with specific cases
huge amounts of power consumption data are generated, which involve excessive energy consumption.
collected and stored. They used 3 dimensions to slicing and  Root Cause Analysis: This type of analytics can be
dicing users’ energy consumption data. These dimensions are performed for scenarios where energy consumption is
time, type of the households and geographical area. Grolinger, more than the expected measure. The analysis can help
Katarina et al. in [6] introduced sensor-based forecasting justify or invalidate an estimated conjecture on average
system for event-organizing venues halls. They used smart power consumption.
meters to measure energy consumption daily, hourly and store  Predictive Analysis: A predictive analysis can be
these in big data centers. Machine-learning alongside with performed by identifying the pattern of device
neural networks (NN) approaches are used to analyze and performance and its power consumption through trend
predict power consumption. Corotinschi and Gaitan in [7] lines.
introduced a solution to manage of electricity and heating
 Access Control of Devices: The end user client
systems to achieve a desired performance. Certain scenarios are
application should provide access control to switch
defined to reduce the power consumption of the building. The
ON/OFF the device remotely.
system has sensory capabilities to interact with the environment
and collect information and it is also able to learn and make  Bill Tracking Utility: Billing is one of the business
decision automatically using machine learning approaches. In process offered by our system. The end devices report
[8], Moreno et al. introduced an automation platform that daily consumption once every day to the middleware via
deploys a network of Home Automation Modules (HAM) MQTT message. The server upon receiving the message
connected to all the appliances and control panels to updates the power consumption charge in the database.
automatically control the devices. The collected information Moreover, the payment history is also preserved for
from HAM is provided to the end users through a web interface. analytics purpose to track bill defaulters.
Moreno et al. in [9] presented a holistic IoT based platform to
collect information in a smart environment. The architecture These business processes are accommodated on a national
implements a mechanism to provide indoor localization of the level as the data sent to the EMS is analyzed on four different
3

levels of stakeholders: Home Owner, Community Owner, City


Owner, and Country Owner. A brief description of these
stakeholders authorized for different services is as follows: The
Homeowners are entitled to Consumption Analysis, Bill
tracking Utility and device access control process. For every
household, the total power consumption of the ACs will be
evaluated on a monthly and yearly basis. The total price
charged for every utility will also be computed to help generate
power consumption bills. For the next level, the Community
Owners are eligible to consumption, asset efficiency and
predictive analysis for monitoring total power consumption.
The energy usage of every household in a community will be
aggregated for evaluating which buildings are using the most
energy. The analysis will help the community owners to
identify a pattern of energy consumption performance with
respect to temperature, humidity, and building age. This energy
usage will be depicted through different types of graphs and
timeline. Based on the data analyzed, peak time analysis as well
as load scheduling events can be planned and executed by the
community owners. On the next higher level, that is, City
Owner, cumulative power consumption from different
communities of the city will be computed. The analyzed data
will be used to create benchmarks based on past record for
assessing root cause analysis. These benchmarks will help
generate monthly/yearly prediction for power demand in Figure 1: System Architecture
advance. Moreover, for the highest level authority of Country
Owner, the data reported from different cities will be examined B. System Hardware
and compared to observe the impact of total energy To collect raw data on the edge device, a Photon
consumption within a state boundary. The aggregated energy development kit is used with its interfacing with DHT11
consumption can be compared to the overall energy generation temperature/humidity sensor and noninvasive CT current
from the Central power station. This will assist the state sensor. It is easy to use, powerful with embedded STM32 ARM
jurisdiction to set climate goals, prioritize or implement energy Cortex M3 microcontroller [12], and also connected to the
strategies, and make data driven energy decisions. cloud which is not available with any other micro-processor
Additionally, the Country owners can have a comprehensive platform like Raspberry Pi or microcontroller like Arduino.
view of which communities and houses fall under different slab Photon module can be programmed in a web IDE available
categories every year. through any web browser. It can store the code online and flash
it on-the-fly to any device across the world. Thus, in this way it
A. System Overview:
provides a lightweight OS on the edge in contrast to Raspberry
The Figure 1 below represents proposed system architecture Pi which runs a fully integrated Linux OS locally. Additionally,
of the work. The architecture comprises of the Photon using a photon eliminates the overhead of using the extra
development kit installed within different buildings to measure hardware of a wireless Wi-Fi shield with Arduino and
temperature, humidity and power consumption. The recorded Raspberry pi to connect all the IoT devices.
data is published to the MQTT broker as different photon kits
are subscribed to the broker as clients. These client messages C. Communication Protocol
are sent from the broker to the Node.js web framework, The proposed system involves end-devices acting as
Express. Using the Node.js language, the sensor data is publishers and middleware acting as subscriber. MQTT was
retrieved and stored in a database for data warehousing. On the used as the communication protocol using the MQTT provides
analytics server side, the BI engine fetches the data from data a non-polling method of communication. Moreover, it being
warehouse for advanced data processing and evaluation results. lightweight provides the ability to support a large number of
Finally, a cross-platform mobile application is developed for connections. Access control was enforced to prevent
the end users to render different business processes as discussed unauthorized access to certain topics. Some topics such
earlier. consumption reporting topic and status change reporting topic
is configured as write only. So only those with required
privilege can read what is being published. While some topics
like control command passing is configured as read only, so that
unauthorized controlling of device is prevented.
D. Software Architecture of the Middleware
Different software tools and servers are used in the
development of the middleware as shown in Fig 1 above:
4

1. MQTT broker viewing/paying and viewing graphs appropriate to the level of


MQTT broker provides a medium for the communication the user. Web service is based on Express framework and uses
between the edge devices and the middleware. Mosquitto server body-parser module to extract data from the request.
is used as MQTT broker. Authentication service returns an api key, which changes per
session, to carry out other operations such as
2. Storage viewing/controlling devices and viewing properties. To an
Hadoop and HBase database are used for storing the edge administrator webserver offers services to add/remove users,
devices’ sensor data and user’s information. The proposed register/unregister properties under a user, grant/deny
system should be capable of handling and processing large privileges to a user. To carry out administrative task a special
volume of data. Scalability is a feature that cannot be key is required, which will only be known to administrators.
compromised in such a system. Hadoop was chosen to store This secret key changes every month. Additionally, to ensure a
real-time consumption data. This data will be used by ETL tools secured communication the services are invoked using https.
like Pentaho Data Integration to load data into Online
Analytical Processing (OLAP) servers like Pentaho BI server 5. Analytics Engine
or IcCube. Data analytics is required to make smart decisions from large
Operational database should also be capable of handling large volumes of data gathered from a mesh network of connected
volume of data with lowest delay possible. HBase satisfies both devices and sensors. For this reason, the team experimented
the requirement. Thus, HBase was chosen as the operational with Pentaho Business Intelligence server (Enterprise Edition)
database to store user details, property details, device details, for analyzing and processing the raw data into meaningful
and other administrative details. Moreover, HBase runs on top results. By default it runs on Apache Tomcat server. Pentaho BI
of Hadoop. The proposed system has three hBase tables, one relies on Mondrian schema, which is a relational based OLAP
table for storing user details, one for storing house and device engine (ROLAP) to provide reports to the user. Pentaho BI
information and the last one is designed to hold a secret key server incorporates execution of an ETL (Extract-Transform
which will be used to do administrative tasks such as registering and Load ) script, where it extracts the data files from Hadoop,
new user, granting privileges to users and registering house transforms it into a star schema or Mondrian schema which is
under a house. The secret key will be updated every month. an aggregation of all the tables as facts and dimensions, and
finally loads this schema onto Pentaho Data source. This is
3. MQTT Interface in Node.js followed by slicing and dicing of schema cubes for analysis and
In the current implementation, MQTT interface is used to rendering reports. A cube is an aggregated collection of
communicate with the edge devices using MQTT protocol. It is measures and dimensions data that allows queries to extract
used to have access to the JSON Messages passed by the end data very quickly. These Cubes are used to create reports for the
devices. Node.js is used as the programming language since the users using the Pentaho Analyzer plug in. The reports/charts are
process involves a large number of I/O operations. A module pulled into the Dashboard designer that offers advanced
named WebHDFS was used to interface Node.js with Hadoop. parameter and prompt options. Finally, the dashboard URL is
MQTT module was used to receive messages from the end- integrated into the web UI for end user access.
devices. The messages will be exchanged in JSON format. For
ease of analytics software this will be converted to CSV format IV. IMPLEMENTATION
using JSON2CSV module and dumped to Hadoop. The real-
time consumption details will be reported every 30 seconds if In this section the hardware and software components that
the device is operating. This information is published with a are used in our solution implementation are described in details.
QoS level of 0, as packet loss is tolerable for this topic. The Package diagram in Fig. 2 represents the integrated
Additionally, every end device is required to send power software modules and hardware parts in a modular fashion.
consumption details for the day to the middleware via MQTT
once every day. This detail is published with QoS level of 2, as
the message must not have any duplicate copies and the
retrieval of this message is crucial. This data will be used to
calculate the consumption charge and will be stored in HBase
database using HBase module in Node.js. Moreover, when the
device changes its state, it has to report the state change via
Mosquitto server to the middleware. The state will be written to
the HBase table.
4. Webserver
The client application accesses the storage database through
different web services implemented using Node.js. These
services are used to read data from the HBase database and send
it back to the requester. It also involves a large number of I/O
operations. Web service is used by the application designed for
the system to authenticate, control devices, view registered
properties, and view registered devices, monthly bill Figure 2: Package Diagram Modules
5

1. Hardware: 2. Software:

The hardware part consists of a Photon development kit i. Benchmarking and Data Analysis:
which is the main processor on the edge. This kit has an
embedded Wi-Fi chip which gives it the ability to access the One of the primary analysis technique in data mining is
internet without any extra modules. An RFID reader (with tags) Benchmarking. Benchmarking the data sets can help identify
is used for the local control of the AC appliances. Switching which devices/buildings should be focused on for setting
on/off the AC devices is physically achieved by implementing optimal energy management goals and policies. The Pentaho
Photon relays bank. Temperature, humidity and transformer Business Analytics serves as an optimum platform for
current sensors are set for measuring these substantial benchmarking real time data and generating user-interactive
quantities. A 220 volts AC fan is used as a proof of concept. reports. For the system design different benchmarking
Depending on the different business processes, the photon scenarios are deemed for four different stakeholder levels. For
kit is programmed to function differently for each scenario. For homeowners the analysis setups comprise of benchmarking
the consumption analysis business process, the Photon is power consumption of different devices in a house,
encoded to collect the temperature, humidity and power benchmarking weather variations with respect to power
consumption data from the sensors every 2 seconds. This data, consumption annually. A screenshot of the Home Owner Level
attached with the user, house and device details, is sent to the analysis graphs created on Pentaho is as shown in Figure 4
MQTT server in JSON format to a specific channel. To below.
implement access control of devices through client application,
the Photon also reports the status of the AC device whenever
the status of device changes by the end user. To enable the
billing utility process, the photon transmits the daily power
consumption for the specific device by using the current
consumption after calculating how much time the device status
is on. Figure 3 below gives a complete view of the hardware
implementation of the project.

Figure 4 (a): Annual Power consumption of each device


(b): Total Annual Power Consumption (c): Annual Power
Consumption relationship with weather variants

In the case of community owners, there are three types of


settings involved, benchmarking power consumption of all
houses in a community for different years, benchmarking for
Figure 3: Hardware components power consumption per square feet of house, and benchmarking
with building age categories. A screenshot of the benchmark
charts is shown in Figure 5 below

Formula used for Power Calculation is:

Power = ((Irms/1.414)*220.0)

Here, Irms is the root mean square AC current measured by


the Non-invasive CT current sensor.
6

Figure 5 (a): Annual Power Consumption of each house


(b): Annual power consumption of each house per unit
area (c):Community Annual power consumption
relationship with average annual temperature (d): Heat
Grid-House Annual Power Consumption with building age
-wise

The City Owners are authorized to assess data set distribution


across various communities within the city. The City Owners
can examine the power consumption spread across different
communities on a yearly and monthly basis. The trend line
graphs can help predict the nature of power consumption of
different communities in the forthcoming years. A screenshot
of the charts created is shown in Figure 6 and Figure 7 below.

Figure 7 (a):Slab Report (b): Energy Map with cities and their power
distribution annually (Red denotes the highest power consumption,
Yellow denotes moderate power consumption, Green denotes the
lowest power consumption annually) (c) Weather Map across UAE
cities depicting temperature-humidity variation

ii. Client Application:

A client application has been developed to facilitate the use


of IoT services available in our implementation. The cross
platform tool, Cordova framework, has been used to develop
the end user application. Cordova is an open-source mobile
development framework. The advantage of using it is that it
allows us to use the standard web development languages like
HTML, Javascript and CSS, instead of using complicated plat-
Figure 6 (a): Annual average Power Consumption of each form-specific APIs. It also insures the cross-platform support
community (b): Annual Power Consumption relationship which means that only one application has to be developed for
of city season-wise (c): Annual Power Consumption Trend different operating systems (androind, ios, etc.) without the
line graph need for reimplementation. However, like any other
technology, Cordova has also its disadvantages.

The UI part of the application is developed using Onsen UI,


The Country Owners are the highest level authority in the since Cordova is just a wrapper for a web application, it doesn’t
stakeholder hierarchy. These owners are entitles to benchmark come with the user interface components, animations, and other
different cities in the country on an annual basis. They can interactive UI elements found in native applications. Onsen UI
inspect power consumption of different communities and is an HTML5 framework used to develop reusable user
houses based on consumption slab range. Additionally, the city interfaces. For the functionality part of the application, we used
owners can observe the temperature, humidity and power AngularJS. AngularJS is a JavaScript framework used to
consumption variation across cities using the Geo-Map facility. simplify both the development and testing of mobile
A screenshot of these benchmark results is shown in Figure 8 applications.The application is developed in a way that gives
below: every stakeholder a different view to the data analytics on
Pentaho according to his privileges and authorities. Once a user
logs in, a service will be run to get the user privileges and the
UI components that he/she will be able to see consequently. For
example, for the home owner, there are 2 services available for
him/her.
7

The remote control services for his/her own house


appliances and power consumption data. The analyzed data on
Pentaho server are provided in the application HTML codes in
iframe tags. The remote control of devices page and the annual
power consumption analysis are shown in the Figure 9 and 10
respectively. For other stakeholders, different views of data
analytics are provided in the application, each according to
his/her privileges and privileges domains (if there are any).

Figure 10 (a): Pending Bill amount displayed with Pay Bill Button
view (b): User notified with the paid bill status

V. EVALUATION AND TESTING


In order to evaluate the overall performance of our system,
the team developed a set of criteria. This criteria in order of
decreasing priority is, Scalability to incorporate different
stakeholders on a national level, throughput and latency for
communication between edge devices and MQTT broker,
scalability test for web server, and speed performance using
Figure 8: Remote control of the devices Hadoop storage. Considering the above criteria, an evaluation
scheme for our project prototype has been modeled. This
scheme examines the latency and throughput of MQTT
communication with respect to varying client threshold for QoS
0 and QoS1 reliability of the broker. Additionally, a Speed
Performance test for Hadoop and Scalability test for Webserver
is also conducted. The tests were performed using the Ubuntu
Server setup for running the MQTT, Hadoop and the Web
server.

i. MQTT Broker Testing


Figure 9: House annual power consumption Wireshark network analysis tool is used to measure the
following metrics: Throughput, Latency, and Packets dropped
in transmission. For an experimental comparison, a logarithmic
For the provision of bill tracking service, the user can view scale of clients was chosen to send 1, 10, 100, and 1000
his/her current bill details for the registered house and pay the consecutive messages with a reliability of QoS 0 and QoS 1 in
pending amount. The user will be notified if the bill amount is each case. The following Figure 11 represents the different test
paid. A screenshot of the bill utility page is shown in Figure case scenarios and their results:
11 below.

Figure 11: MQTT Test Cases and Results


8

Through the evaluation results on MQTT server, we clients. The MQTT communication performance can greatly
deduced that the maximum threshold for the broker is only improve using a more powerful machine for MQTT broker
1000 concurrent client messages and the connection is setup. Scalability can be improved with the bridge capability
broken beyond this range. Benchmark results for offered by the MQTT broker. Also, with proper load balancing
Throughput Vs. Number of clients is shown in Fig 12 techniques the scalability of MQTT can be improved.
below. Moreover, using distributed MQTT brokers like Apache Kafka
can help alleviate the issue of scalability of MQTT.

Throughput Vs. Number of i. Scalability Test for Webserver

Clients 4000 users were simulated using a webserver Stress Tool.


150000 Each user will be making post request to authentication server
Throughput (Bps)

and will be responded with 1 or -1 depending on the correctness


100000
of passed parameters as stored in HBase. CPU and Network
50000 QoS 0 utilization was high when all requests started coming as shown
0 QoS 1 in Figure 14. Almost all requests were answered immediately.
1 10 100 1000 Memory remained almost 85% throughout the test. This is due
Number of Clients
to HDFS write activity which runs asynchronously in the
background. All requests were responded by the server.

Figure 12: Throughput Vs. Number of Clients Graph

It can be inferred from the above result that since QoS 1


ensures a better reliability than QoS 0, therefore the measured
throughput for QoS 1 is higher than QoS 0 until 100 clients.
However, the throughput for QoS 1 decreases and is lower than
QoS 0 for more than 100 client messages. This discrepancy can
be accounted for the fact that as the number of clients reaches
to 1000, there is a significant packet loss during message
transmission. Benchmark result for Latency Vs. Number of
client messages is shown in Figure 13 below.

Latency(s) Vs. Number of Clients


200000
Latency (ms)

150000
100000
QoS 1 Figure 14: Server Performance during the request phase
50000
0 QoS 0
ii. Speed Test for Hadoop Database
1 10 100 1000
Number of Clients Time required to write data to Hadoop using WebHDFS is
crucial to the project. The time required should be as low as
Figure 13: Latency (s) Vs. Number of Clients possible for a large system. The time difference between
submission of job and completion of job was measured for 4000
From the above results it can be deduced that the concurrent requests. The average time it took for WebHDFS to
overall latency for QoS 0 is always lower than that of QoS 1 for write was found to be around 66 sec. A summary of the
all categories of clients. This difference is due the fact that in response time to Complete a write by WebHDFS is shown in
QoS 0 message delivery is not acknowledged whereas in QoS Figure 15 below.
1 an acknowledgment for confirmed message reception is sent,
which adds to the latency of message transmission. Moreover, Average response time (ms) 65,925.30
it can be deduced from Figure 8 that for 1000 clients’ message Maximum response time (ms) 122,382.00
transmission there is a packet drop of 2.9% in QoS 0 and 2.6%
Minimum response time (ms) 329.00
in QoS 1. However, since QoS 1 is more reliable than QoS 0,
therefore the loss percentage of messages for QoS 0 is higher Median response time (ms) 67,378.00
than that of QoS 1. From the above experimented data, we can Figure 15: Response Time by WebHDFS
infer that these results simulate a small community of 1000
9

Similarly, time required to fetch 4000 files from Hadoop by IEEE, Queenstown, 2014, pp. 53-58.
executing an ETL script was measured. Two attempts were doi: 10.1109/SAS.2014.6798916
made to measure the time required to read from Hadoop. Each [5] Zhou, Kaile, and Shanlin Yang. "Understanding household
file is ~62 Bytes in memory. In the first attempt, the time taken energy consumption behavior: The contribution of energy big
was 3481ms. In the second attempt, the ETL took 4062ms to data analytics." Renewable and Sustainable Energy
complete the task. These measures are for 4000 concurrent Reviews 56 (2016): 810-819.
communication. Thus, this simulates a small city and ensures a [6] Grolinger, Katarina, et al. "Energy Forecasting for Event
decent performance level. Moreover, results can be enhanced Venues: Big Data and Prediction Accuracy." Energy and
by running the setup in a distributed environment. Buildings 112 (2016): 222-233.
[7] G. Corotinschi and V. Gaitan, "Smart cities become possible
VI. FUTURE PROSPECT thanks to the Internet of Things", 2015 19th International
The current system design can enhance its scalability domain Conference on System Theory, Control and Computing
by making it compatible with existing HVAC compliant (ICSTCC), 2015.
communication protocols such as BACNet or LonWorks. [8] M. V. Moreno, B. Úbeda, A. F. Skarmeta and M. A.
BACNet is an ISO 16484-5, and ANSI/ASHRAE 135-1995 Zamora, "How can We Tackle Energy Efficiency in IoT
standard [13]. Having multiple operating interfaces for Based," MDPI , no. 14, p. 33, 2014
controlling HVAC systems of the building is difficult. Thus, [9] M. Moreno, J. Santa, M. Zamora and A. Skarmeta, "A
integrating the system design with BACNet compliant devices holistic IoT-based management platform for smart
will provide a single platform solution with interoperability environments", 2014 IEEE International Conference on
among different HVAC systems of the building. Another Communications (ICC), 2014.
addition to the current system design can be machine learning [10] K. Akkaya, I. Guvenc, R. Aygun, N. Pala and A. Kadri,
technologies. Applying machine learning technologies would "IoT-based occupancy monitoring techniques for energy-
enable the system to analyze the data more efficiently and it also efficient smart buildings", 2015 IEEE Wireless
makes the systems able to predict data for the future. The Communications and Networking Conference Workshops
system can be largely scaled with the implementation of wake- (WCNCW), 2015.
up technology for the used sensors like temperature, humidity, [11] A. Bhardwaj, "TATA Consultancy Services-White Paper,"
and current sensor. 2015.[Online].Available:
http://www.tcs.com/SiteCollectionDocuments/White-
Papers/BPS-Internet-of-Things-Smart-Energy-Management-
VII. CONCLUSION
1015-1.pdf.
Our proposed project is set to open new avenues for smart [12] "Prototyping tools for the Internet of Things", Particle,
energy management on IoT Platform. Hadoop and Data 2016. [Online]. Available: https://www.particle.io/prototype.
analytics is used extensively for the process of data mining in [Accessed: 05- Apr- 2016].
order to develop smart algorithms on large data for decision [13] Downloads.siemens.com, 2016. [Online]. Available:
making. The system design of our project provides a complete https://www.downloads.siemens.com/download-
environment for smart energy management such as remote center/Download.aspx?pos=download&fct=getasset&id1=A6
control of devices, automatic generation of bills, and record V10209534. [Accessed: 29- Feb- 2016].
generation to view the energy impact of appliances and the
related price charged through an end user mobile application.

REFERENCES
[1]S. Writer, "Greening the existing building stock of the UAE
– Middle East Construction News", Meconstructionnews.com,
2011.[Online].Available:
http://meconstructionnews.com/557/greening-the-existing-
building-stock-of-the-uae. [Accessed: 16- May- 2016].

[2] M. Kim, J. Choi and J. Yoon, "Development of the Big Data


Management System on National Virtual Power Plant," 2015
10th International Conference on P2P, Parallel, Grid, Cloud and
Internet Computing (3PGCIC), Krakow, 2015, pp. 100-107.
[3] W. Lee, B. W. On, I. Lee and J. Choi, "A big data
management system for energy consumption prediction
models," Digital Information Management (ICDIM), 2014
Ninth International Conference on, Phitsanulok, 2014, pp.
156-161.
[4] R. Bergelt, M. Vodel and W. Hardt, "Energy efficient
handling of big data in embedded, wireless sensor
networks," Sensors Applications Symposium (SAS), 2014

You might also like