A System for Energy Conservation Through Personalized Learning

Mechanism

Aryadevi Remanidevi Devidas, Sweatha Rachel George and Maneesha Vinodini Ramesh

Amrita Center for Wireless Networks and Applications, Amrita Vishwa Vidyapeetham, Amritapuri,

Clappana P. O, Kollam, India

Keywords: Smart Building, Personalized System, Zigbee.

Abstract: Several challenges exist in developing smart buildings such as the development of context aware algorithms

and real-time control systems, the integration of numerous sensors to detect various parameters, integration

changes in the existing electrical infrastructure, and high cost of deployment. Another major challenge is to

optimize the energy usage in smart buildings without compromising the comfort level of individuals.

However, the success of this task requires in depth knowledge of the individual and group behaviour inside

the smart building. To solve the aforementioned challenges, we have designed and developed a Smart

Personalised System for Energy Management (SPSE), a low cost context aware system integrated with

personalized and collaborative learning capabilities to understand the real-time behaviour of individuals in a

building for optimizing the energy usage in the building. The context aware system constitutes a wearable

device and a wireless switchboard that can continuously monitor several functions such as the real-time

monitoring and localization of the presence of the individual, real-time monitoring and detection of the

usage of switch board and equipment, and their time of usage by each individual. Using the continuous data

collected from the context aware system, personalized and group algorithms can be developed for

optimizing the energy usage with minimum sensors. In this work, the context aware system was tested

extensively for module performance and for complete integrated device performance. The study found the

proposed system provides the opportunity to collect data necessary for developing a personalized system for

smart buildings with minimum sensors.

1 INTRODUCTION

Resources for electrical energy are decreasing day

by day, while its use is increasing day by day.

Therefore, we need to reserve electrical energy for

future use. About 50 million tonnes of electrical

energy is wasted globally every year due to

negligence and carelessness (Nationmaster 2014).

Smart buildings not only conserve electrical energy,

but also provide resource sustainability and more

efficient and effective energy monitoring operation.

A building which has context awareness and an

ability to react to it is termed as smart. As a result of

this power conserving nature, many countries have

developed smart buildings (Eun-Kyu Lee and Gadh

2013).

Smart buildings are more effective when the

personal requirement of each person in the building

is studied and reacted accordingly, i.e. a

personalized system in which the electric power is

consumed according to the usage behaviour of the

occupants. The system notes when and where the

occupants enter and exit, what building space the

occupants inhabit, and what time and how long they

occupy that space so that the system can

automatically adjust to the electrical requirements of

each individual (Sinopoli, 2014). Similarly, when a

group of persons enter a room, the system is

designed to have optimal electricity consumption. In

such a case, the group behaviour is studied and

reacted according to the algorithm applied, where

the algorithm supports the majority interest in the

group.

In this work, we designed and developed a

context aware wearable device and a wireless

switchboard to monitor and track the switch on/off

activities performed by each of the individuals in a

building. The system was tested under three

different scenarios namely, the data acquisition

phase, the working phase and when a person is

357

Devidas A., Rachel George S. and Vinodini Ramesh M..

A System for Energy Conservation Through Personalized Learning Mechanism.

DOI: 10.5220/0005453903570363

In Proceedings of the 4th International Conference on Smart Cities and Green ICT Systems (SMARTGREENS-2015), pages 357-363

ISBN: 978-989-758-105-2

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

absent from the room. Whenever an individual

enters a room and performs any switch on/off

activity, his/her presence in the room, along with

their activities, are updated to an aggregator, with

the help of a wearable device and a wireless

switchboard, so that if ever the same individual

enters the room again, the end device can be

automated for the previously used time duration. A

smart phone with Near Field Communication (NFC)

can provide the same functionality of a wearable

device, i.e. to monitor or track the individuals within

a room and to distinguish the switch on/off activity

performed by each individual. However, in order to

learn about the individual’s behaviour, each

individual must carry his smart phone anywhere

within the building, especially when he/she

approaches the switchboard, that in turn can become

a burden to the individual. A better way to solve this

issue is by replacing the smart phone by a wearable

device tied to the wrist so that each time an

individual raises his wrist to switch on any device,

his identity is noted by the wireless switchboard.

The remaining paper is organized as follows. In

section 3, an architecture for the system is explained,

describing the design of each of the devices used in

the system and the working of the whole system. In

section 4, an algorithm for the working of the system

is summarized. Finallyin section 5, the hardware

development of the devices and module-wise and

system-wise testings are discussed.

2 RELATED WORK

Currently, smart buildings feature multi-system

integration with multi-functions that integrate data

from different buildings to r monitor data against

benchmarks or established goals. For effective

management operation based on the human

behavioural study, several innovative ideas are being

incorporated into smart building technology

(Sasidhar and Thomas, 2014). During the early days

of development of smart buildings, developers

aimed to provide fundamental resource services like

water and electricity. Now, developers focus on

providing methods to conserve more energy

resources such as thermal energy.

The authors in a research publication (Sinopoli,

2014) proposed a smart learning based control

system that controls the AC appliance through a

Bluetooth transceiver interfaced to the controller. It

uses light sensors to detect whether any windows are

open before turning on the AC. Through this system,

the researchers saved upto 5% of energy. However,

they did not consider the end appliances other than

AC. In our system, all the electrical end appliances

including TVs, computers, etc are considered.

JinSungByun et. al. proposed an intelligent

system in a building that provides energy saving

services and remote control over consumer devices,

consisting of a set of sensor modules like

temperature sensors, humidity sensors, and light

intensity sensors with an internet interface which

helps to remotely control the end devices at the time

of need (JinSungByun, 2011). Their system saved up

to 16-24% of energy. However, they did not discuss

anything about the topology of sensor networks. The

system that we developed minimizes the use of

sensors, thereby minimizing the cost and

complexity.

Dae-Man Han et. al. proposed a sensor network

based smart, light control system for smart home and

energy control applications. For better device control

and efficient energy management, smart home

networking used IEEE 802.15.4 and Zigbee

networks (Dae-Man Han, 2010). However, they

considered only lighting system applications and did

not take into account other electronic appliances like

TVs. Furthermore, their system did not study the

behaviour of each individual and did not explain the

algorithm for controlling the end devices. Moreover,

they did not find a way for the optimization of

sensor use. Zigbee communication is used in our

system because it provides a low cost and low power

communication. As the use of numerous sensors can

increase the cost and complexity of the system as in

(Jin Sung Byun, 2011

), the sensor usage in our

system is minimized.

Boungju Jeon et. al. proposed a Zigbee based

intelligent self-adjusting sensor (ZiSAS) that can

take into account the limitations of sensor networks

such as battery lifetime, bandwidth, storage

capabilities etc (Boungju Jeon, 2012). It

automatically configures the network topology and

system parameters and detects a node failure or

addition or removal of any node to the network.

Their system reduced energy consumption by 8-

34%. However, the authors did not state the

condition when the residents in the building

exhibited irregular behaviour. This makes it difficult

to generate a common pattern from the same

situation. Furthermore, the researchers did not

explain about the routing protocols and the way

sensors can be optimized.

Yuvraj Agarwal

et. al. presented a ‘presence

sensor platform’ that detects the presence of

occupants in an office building through which the

HVAC equipment can be automatically adjusted

SMARTGREENS2015-4thInternationalConferenceonSmartCitiesandGreenICTSystems

358

according to the context (Agarwal, 2010). The node

is composed of a PIR sensor module that detects

human presence and a magnetic reed switch door

sensor that detects whether the door is opened/

closed. The authors did not consider a method for

battery replenishment and optimization of sensor

usage. Moreover, this system will require many PIR

sensors to be deployed within the room for motion

detection, which is not a cost effective solution. Our

system is more affordable than conventional systems

since it does not require the sensor installation cost

and wiring cost.

rdiczka et. al. proposed a system for detecting

and studying the contextual human behavioural

models in a building by using camera tracking

systems, microphones etc (Brdiczka and Langet,

2009). Through this system, they reduced energy

consumption to 10-15%. The system required full

time working of the sensors for detection, which can

affect the lifetime of nodes. However, the

researchers could not detect the human behavioural

situations other than the stated 6 situations in the

paper. The use of cameras made the system

expensive. Also, they could not provide privacy to

rooms. So in our system, for the purpose of

monitoring the position or activity of the individuals,

cameras and microphones were not used.

A. Fleuryet. al. proposed a smart home that can

measure the activity of a person and help people in

their activities (Fleury, 2008). The environment

within a building or room was monitored using

numerous sensors and microphones to detect any

distress situation or the activities in the room.

However, the researchers could not minimize the

number of sensors used and did not explain the

topology of deployment of the sensor network.

Moreover, the privacy to the rooms is lost. In our

system, microphones are not used for detecting the

presence or activities of a person in a room as it can

break the privacy of occupants.

3 SMART PERSONALIZED

SYSTEM FOR ENERGY

MANAGEMENT

The Smart Personalized System for Energy

Management (SPSE) is a context aware system,

designed to work in office buildings, where people

inhabiting the building exhibit regular behaviour.

The aim of this system is to conserve energy or

optimize energy usage through context aware data

collection and personalized learning mechanisms. A

wearable device has the capability to communicate

with a wireless switchboard to monitor and track the

individuals who have performed the activity of

switch on/off. The information collected by the

wireless switchboard through the wearable device

will be used to develop personalized learning of

each of the individual’s behaviour. This personalized

learning will provide the opportunity to minimize

energy consumption by each individual under

specific time frames. The SPSE architecture is

shown in Figure 1.

3.1 Design of Wearable Device

The wearable device was designed in such a way

that it is compact and handy to use. It should contain

a communication module that can convey the

Figure 1: SPSE architecture.

ASystemforEnergyConservationThroughPersonalizedLearningMechanism

359

presence or absence of person in a room, identify

each individual and be able to distinguish the switch

press of each individual among the many gathered in

the room.

A Zigbee module (XBee Series 1) can help to

convey the presence or absence of individuals in a

room (International, 2014). Since XBee uses 3.3V, a

micro-controller that uses 3.3V power supply -

MSP430G2553 - is best suited to make the device

more compact (Instruments, 2013) (Msp430

launchpad, 2014). An RFID tag (125KHz) is used to

identify and to distinguish the switch press of each

individual among the many gathered in the room.

Two AA batteries of 1.5V each in series are used to

power up the device. In this way, each wearable

device gives a unique identity to each individual in

the building.

3.2 Design of Wireless Switchboard

The wireless switchboard is meant to monitor the

position information of the individual, detect the

switch on/off activity performed by the individual,

transmit the collected information to the aggregator

and perform the controlling operation of end

devices. It uses an active RFID reader that reads

from a passive RFID tag on the wearable device to

indicate that the individual is near to the

switchboard. The device also uses a Zigbee module

(XBee series-1) for communicating the information,

a micro-controller (PIC16F877A) with a voltage

sensing circuit, voltage regulators (5V and 3.3V) and

an electromagnetic relay to relay the current to the

end device.

3.3 Working of SPSE

Each individual within the building carries a

wearable device that provides a unique identity.

Figure 2 shows the sequence diagram for the

working of the system. The Zigbee module on the

wearable device continuously transmits a message to

the wireless switchboard signalling the presence of

the person within the room. The wireless

switchboard continuously monitors the reception of

this message. Whenever the message is not received

beyond a threshold time, the absence of the person

in the room is marked and all the end devices within

that room are turned off. When a person enters the

cabin, turns any switch ON and then turns off the

device after his requirement, his RFID tag identity

along with the information such as what switch he

pressed, the time duration for the device usage is

transmitted to the aggregator. The aggregator is a

Zigbee module connected to a PC that stores the

database of electrical use by each individual and

displays the person’s switch on/off activity

information. This information is displayed on the

serial terminal of X-CTU software. The switch

on/off activity of the individuals is continuously

monitored and updated to the database where a

common pattern is developed for each of them. The

data acquisition phase and working phase occur

simultaneously. The data acquired is analysed using

machine learning algorithm to develop a common

pattern of electrical behaviour for each individual.

The machine learning algorithm can be supervised

or unsupervised learning. For supervised learning,

the database of each individual’s electrical usage is

known a-priori, so that it becomes useful to classify

the individuals from the available training set. For

unsupervised learning, the electrical behaviour of

each individual is necessary to cluster the

individuals and develop a common pattern from

them.

Figure 2: Sequence diagram of the working of SPSE.

4 ALGORITHM

4.1 Wearable Device

First of all, it is important to set the parameters

SMARTGREENS2015-4thInternationalConferenceonSmartCitiesandGreenICTSystems

360

required for the application to run properly, such as,

stop the watchdog timer, set the Digital Crystal

Oscillator (DCO), set the transmit (Tx) and receive

(Rx) pins of UART, set the serial clock and baud

rate to 9600. Then, data must be continuously

transmitted to provide an identity periodically to

each person for 5 seconds interval. The interval is

provided by calling the delay function. The flow

chat for the working of the wearable device is shown

in Figure 3.

4.2 Wireless Switchboard

Whenever a person enters a cabin and performs any

switch/off activity, read the RFID data from the tag

on the wearable device. If the switch is on, turn on

bulb and start a timer. If the switch is off, wait for

the switch on action to be performed. Then, check

whether any XBee data is received from the

wearable device. If no XBee data is received for

more than a threshold time, then turn off the bulb.

Transmit the information such as RFID, switch

number and time duration of use of the device to

aggregator via XBee. In between, detect the switch

off condition. If the switch is turned off, turn off the

bulb and note the timer overflow value. Calculate

the time for which the device was turned on. Then,

transmit the information like RFID, switch number

and time duration of use of the device to aggregator

via XBee. If any XBee data is received again at the

same time after the data has been acquired, then turn

on the bulb for the previously obtained timer value,

indicating the working phase of the system. The data

acquisition phase and working phase occur

simultaneously. The acquired data can be averaged

to obtain the common working pattern for each

individual. Figure 4 shows the flow chart for the

working of a wireless switchboard.

Figure 3: Flow chart of wearable device.

Figure 4: Flow chart of wireless switchboard.

5 HARDWARE

IMPLEMENTATION AND

TESTING

Hardware was developed for both the wearable

device and the wireless switchboard. Wearable

device consists of a RFID tag, Zigbee module and a

micro-controller. Wireless switchboard consists of a

Zigbee module, relay, bulb, switch, voltage

regulators, micro-controller and a RFID reader. The

hardware of both devices is shown in Figure 5 and

Figure 6.

Figure 5: Hardware of wearable device.

ASystemforEnergyConservationThroughPersonalizedLearningMechanism

361

Figure 6: Hardware of wireless switchboard.

As a part of the testing, whenever a person carrying

the wearable device presses the switch, the bulb is

turned on. As the person approaches the

switchboard, the RFID tag transmits its 12 Bytes of

data to the RFID reader in the wireless switchboard,

indicating the identity of the person turning on the

device. The 12 Bytes of RFID data actually contains

a start byte (0x0A) and a stop byte (0x0D), of which

the middle 10 bytes are the original RFID data

(

Rhydolabz, 2014). The person turns the switch off

after 40 seconds of time, which in return turns the

bulb off. As soon as the bulb is turned off, a message

is delivered to the PC via Zigbee (XBee S1),

regarding the switch press information: "Switch-1 is

pressed for 40.00 seconds duration by

720040BF9A." The X-CTU software has the serial

terminal that shows this information. The circled 10

Bytes in Figure 7 are the ID values of RFID.

Figure 7: X-CTU terminal showing test result of the

system.

The three scenarios under test are: data

acquisition phase, working phase and absence of

person in the room. In the data acquisition phase, as

explained above, the details regarding the entry of

the person, his ID, what switch he presses, for what

duration are studied and updated to the aggregator.

In working phase, whenever the same person enters

the cabin again after the data acquisition phase, the

bulb is automatically turned on for the learned

amount of time (40 seconds for the above test). In

case the person was not in the room for more than a

threshold amount of time, the bulb is automatically

turned off. The full hardware set up of the system

under test is shown in Figure 8.

The average walking speed of the individual is

about 5Kmph (Walking, 2015). So the minimum

Figure 8: Hardware of complete system.

time the individual takes to exit the room is,

Time = Distance to the exit door ÷ Average walking

speed of the person (1)

The XBee Series-1 in the wearable device can be

designed to transmit messages to the wireless

switchboard for a room of any dimension. The

individual walks a minimum distance of the room

dimension to exit the room. So, the minimum time

the individual takes to walk out of the room can be

found from equation (1). The received signal

strength of the Zigbee message helps the wireless

switchboard to estimate the distance from the

wireless switchboard to the current location of the

person with wearable device.

The battery replacement of wearable device

depends on the power consumption of the XBee

module and the microcontroller. Power consumption

of wearable device at active mode,

= 3.3 x (45 + 0.23) x 10

-3

= 149.26mW.

Power consumption of wearable device at

standby mode,

= 3.3 x (10 + 0.5) x 10

-6

= 0.03465mW.

Battery capacity = 1500mAh

Lifetime = Battery capacity/ Zigbee current

= 1500/45.23 =33.16 hours

From equation (1), the minimum time the individual

takes to walk out of a 5m x 5m room is 3.6 seconds.

So it is necessary that the XBee message must be

transmitted at least once in every 3.6 seconds in

order to detect the absence of person in room.

The message transmission interval of the XBee

module can be increased to a particular value if the

presence of a person learned from the previously

stored database is found to be within the room for

longer hours. In this way, the battery usage of the

wearable device can be enhanced and stay longer

time.

SMARTGREENS2015-4thInternationalConferenceonSmartCitiesandGreenICTSystems

362

6 CONCLUSIONS AND FUTURE

WORK

A system architecture that conserves electrical

energy by learning the personalized behaviour of

occupants in the building was developed. Based on

the system architecture, we designed a context-

aware wearable device that monitors the proximity

of the individual from the switch board and collect

the switch on/off activities performed by the

individual and a wireless switchboard that

continuously collects personalized data from each

individual. The system hardware has the capability

to learn the individual's behaviour over electrical

appliances through the wearable device, wireless

switchboard and an aggregator. In this work, the

context aware system was tested extensively

multiple times for several days and various time

durations. The system was also tested for module

wise and complete integrated device performance.

The results conclude that the proposed system

provides the opportunity to collect the data

necessary to develop a personalized system for smart

buildings with minimum sensors. As future work, a

definite algorithm to learn the electrical behaviour of

each individual from the data acquired is developed

with the system applied to multiple users by

incorporating Zigbee range limitation to each rooms.

ACKNOWLEDGEMENTS

We would like to express our sincere gratitude to our

beloved Chancellor, Sri. Mata Amritanandamayi

Devi (AMMA), for the immeasurable motivation

and guidance for doing this work.

REFERENCES

Nationmaster, 2014, Webpage.

Eun-Kyu Lee, F., C., Gadh, R., 2013, Fine-grained access

to smart building energy resources. IEEE Internet

Computing.

Sinopoli, J., 2014, Smart buildings now and tomorrow.

BACnet International Journal.

Sasidhar, K., Thomas, N., Subeesh, T., S., 2014, A Smart

Learning based Control System for Reducing Energy

Wastage. In Global Humanitarian Technology

Conference – South Asia Satellite (GHTC-SAS), IEEE

Conference Publications.

JinSungByun, S., P., 2011, Development of a self-

adapting intelligent system for building energy saving

and context-aware smart services. In IEEE

Transactions on Consumer Electronics., vol. 57, no. 1,

pp. 90–98.

Dae-Man Han, J., H., L., 2010, Smart home energy

management system using ieee 802.15.4 and zigbee. In

IEEE Transactions on Consumer Electronics, vol. 56,

pp. 1403–1410.

JinsungByun, S., P., BoungjuJeon, J., N., 2012, An

intelligent self-adjusting sensor for smart home

services based on zigbee communications. In IEEE

Transactions on Consumer Electronics, vol. 58, no. 3,

pp. 794–802.

Agarwal, Y., B., B., 2010, Occupancy-driven energy

management for smart building automation. In

BuildSys 10 Proceedings of the 2nd ACM Workshop

on Embedded Sensing Systems for Energy-Efficiency

in Building, pp. 1–6.

Brdiczka, O., Langet, M., Maisonnasse, J., Crowley, J.,

2009, Detecting human behavior models from

multimodal observation in a smart home. In

Automation Science and Engineering, IEEE

Transactions on, vol. 6, no. 4, pp. 588–597.

Fleury, A., 2008, Sound and Speech Detection and

Classification in a Health Smart Home. In 30

Annual

International Conference of the IEEE, pp. 4644-4647.

Instruments, T., 2013, Mixed signal microcontroller,

Datasheet.

Msp430 launchpad, 2014, Userguide.

International, D., 2014, Xbee multipoint rf modules,

Datasheet.

Rhydolabz, 2014, Rhydolabz-wiki, webpage.

Walking, 2015, Website, Wikipedia.

ASystemforEnergyConservationThroughPersonalizedLearningMechanism

363