PERSONALISED AMBIENT INTELLIGENCE IN BUILDINGS

VIA CONTEXT-AWARE AGENTS

Soi Luong

Technologies for Sustainable Built Environments

University of Reading, Department of Physics, Reading, RG6 6AF, United Kingdom

Samuel Chong

Capgemini U.K.

77-79 Cross Street, Sale, M33 7HG, United Kingdom

Keywords: Context-Aware Agents, Intelligent Buildings, Hawthorne Effect, Intelligent Agents, Ambient Intelligence.

Abstract: One of the concepts of intelligent buildings is to maximise occupant comfort. Optimising the environment

for a single occupant is a simple procedure but for a shared environment presents a difficult challenge. This

paper proposes the use of context-aware agents that utilises the Hawthorne Effect theory for providing and

resolving the conflicting preferences in a multi-tenant building environment. This is an ongoing research

and the solution consists of a multi-agent framework for interacting with the environment and learning the

occupant’s preferences through a user-interface. The agents will learn the occupant’s preference using a

simple heuristic question and answer method and perception through networked sensors. The Hawthorne

Effect solution attempts to resolve the conflicting requirements by fluctuating the heating and lighting

during the course of the day to meet all occupants preferences.

1 INTRODUCTION

A typical building has a set of common services

such as heating, ventilation and air conditioning

(HVAC) systems, lighting, security systems and

electrical technologies. All or most of these services

are essential to provide a habitable environment that

is both comfortable and satisfies occupant needs.

The underlying components of these services are

sophisticated but can be simply operated by a switch,

automated or programmed to run at a certain time of

the day.

Large organisations have many offices to

accommodate their employees. They are required to

provide an ergonomic environment that is compliant

with legislation and regulations. It is also of business

interest that they configure the internals of their

buildings to maximise employee productivity. For

example, an office is populated by occupants and

computer systems but they must also have bright

reflective lighting, central heating systems to keep

the occupants warm during the winter periods, air

conditioning systems to keep the temperature and

humidity levels low during summer periods and

ventilation to extract CO

from the environment.

These services must be fully functional when an

occupant is present in the office. However, in a

multi-tenant office the services need to be shared

and the environment is most likely unable to meet

everyone’s preferences.

The concept of intelligent buildings was

introduced over 20 years ago and its common

purpose is to ‘maximise occupant comfort’ and

‘minimise life-cycle cost’. The purpose suggests that

we should think about what the occupants want and

how intelligent buildings can help an organisation.

In the following, we examine the use of context-

aware intelligent agents to provide personalised

heating and lighting services for human

surroundings in a shared office environment.

This is an ongoing piece of research which

focuses on examining the agent methodologies

which are applicable to maximising occupant

comfort. A solution has been proposed and a

prototype has been built based on the study of

Luong S. and Chong S.

PERSONALISED AMBIENT INTELLIGENCE IN BUILDINGS VIA CONTEXT-AWARE AGENTS.

DOI: 10.5220/0003258900170023

In Proceedings of the Twelfth International Conference on Informatics and Semiotics in Organisations (ICISO 2010), page

ISBN: 978-989-8425-26-3

related work and current research of intelligent

agents and multi agent systems. This paper discusses

our contributions to Personalised Ambient

Intelligence by presenting a heuristic question &

answer interface and the inclusion of the Hawthorne

Effect theory.

The paper is presented as follows. In section 2,

we discuss the problems associated with this

research and construct scenarios for the purpose of

validation. Section 3 contains an analysis of the

related work by academic research and other

pioneering projects. We look at our proposed

solution in Section 4. Finally, we critically evaluate

and conclude the paper in Section 5.

2 AMBIENT INTELLIGENCE

IN BUILDINGS

Raising the topic of ambient intelligence one would

certainly envision an array of intelligent devices

seamlessly integrated with the environment. The

devices know who you are, react to your actions and

the environment, and anticipate your desires before

you even show it. In reality, ambient intelligence

involves ubiquitous computing, user profiling and

intelligent systems (Shadbolt, 2003); this denotes the

core concept of utilising inexpensive, networked

context-aware devices simultaneously with human-

computer interaction and behaviour perception for

pattern recognition to construct human requirements.

Our approach comprises of three research

objectives for designing a system that is ‘seamlessly

integrated’ with the devices and the environment.

2.1 Ambient Intelligence

The system should govern the electronic network

that is sensitive to occupant’s needs. The problem

with existing building services is that when it is

required the occupant has to operate it themselves

and they may forget to turn if off when they leave

the building. Personalisation is required as each

tenant has different preferences in which they may

want to configure their desired environment. Rather

than having the occupant repeatedly informing the

system what they want, it should be able to

anticipate and change the environment accordingly

by continuously recalibrating itself (Shadbolt, 2003).

2.2 Context Awareness

The building services must have the ability to sense

and react to the environment through devices, such

as, RFID readers for sensing human presence, a

thermometer to read the room temperature, etc. But

such devices must have a level of sophistication to

perceive a situation or human factors: social

environment, i.e. co-location of others and group

dynamics (Schilit, 2004).

2.3 Intelligent Agent

Intelligent agents are the main focus of the

simulation which should be responsible for

observing and reacting to provide a dynamic

environment through a context-aware interface.

Intelligent agents in buildings act autonomously and

discreetly from occupants requiring initial training

and calibration to accumulate its knowledge base of

user profiling to provide personalised services in an

office environment.

Figure 1: How agent interacts with environment through

sensors and effectors.

Intelligent agents are categorised into four classes

based on the degree of perceived intelligence and

capability:

1. Simple reflex agent

2. Goal-based agent

3. Utility driven agent

4. Learning agent

Simple reflex agent methodology is based on a

condition-action rule: if certain criteria meet the

condition then it will perform an action. Other

complex agents, such as, goal-based and utility

driven agents behave in a similar fashion to find

possible solutions to achieve its goal. The actions

performed are repeated until it produces the desired

result or alternatively attempts other methods. Utility

driven agents are more selective about which

method to undertake and choose the best possible

method to achieve a goal (Kozma, 1998). A learning

agent has the ability to act independently and learn

ICISO 2010 - International Conference on Informatics and Semiotics in Organisations

from past experience. It needs to be able to “adapt its

behaviour according to user needs” (Pissinou, 1997).

Each agent is able to assess the current state of the

environment and perform appropriate actions but a

learning agent has the ability to learn from a

dynamic context, for example, perceiving the social

environment in a multi-tenant office, maintaining a

history record of human behaviour and adapting to it

to provide a personalised environment.

For this research, we evaluate three scenarios for

the purpose of constructing a simulation:

1. User-interaction and personalisation

Occupants require a method of communicating

with the system that manages the building

services. Normally, the occupant uses the light

switch on the wall to activate the lights, a

thermostat to control the heating or an air-

condition control panel to provide cool air both

of which requires the occupant to enter a value in

degree Celsius or Fahrenheit. In order to train a

computer, a user-interface is essential to capture

the occupant’s preferences.

2. Shared services

In an office environment, HVAC and lighting are

shared amongst the occupants. The HVAC

operates as a single unit to regulate the room

temperature. Lighting is slightly different as

there could be several switches for operating

lights on separate networks. It would be

beneficial if these services could operate

independently to provide a personalised service

and minimise energy expenditure. For example,

the computer system can set the four air

conditioning systems in the office to operate at

different temperature levels.

3. Social environment

There are scenarios where all occupants may

gather in a room to hold meetings. Prior to

entering an assistant to the organiser prepares the

room five minutes before the meeting begins to

ensure the room is at an acceptable temperature

and the lighting is turned on for the attendees.

The co-location of the occupants and group

dynamics in this situation suggests that the

computer must be able to handle a variable

context and prepare a personalised service in

anticipation.

The scope of this project is to implement a software

agent that intelligently manages the heating and

lighting of a building office environment. Simulated

devices include sensors, radiators, air conditioning

and lights.

3 RELATED WORK

Intelligent Buildings is a very broad area of research

that comprises of many components which are

grouped under four categories: “facilities

management, information management, connectivity

and overall control” (Flax, 1991). However, this

paper focuses on facilities management of heating

and lighting in the building.

3.1 Intelligent Agents/Multi-Agent

Systems

There are four classes of intelligent agents as

discussed in Section 2 and they each act and perform

differently from one another. To summarise, the four

classes of agents are capable of simply operating by

condition-rules, reiterate the best results, be selective

about the best results and learn from their

environment. The learning agent appears to be

suitable for our simulation as one of the main

requirements is personalisation.

Figure 2: Intelligent Agent Framework.

Typically, this type of agent has a data repository for

recording knowledge. But, the definition of learning

in this scenario would be a domain-specific

knowledge base and not an agent that utilises

learning techniques, such as, neural networks and

genetic algorithms (Chen, 1995). For this project,

the intelligent agent keeps a repository of the heating

and lighting preferences of each user.

There are many limitations with using a single

agent to be the user-interface, negotiator, processor

and performer. With the introduction of multi-agent

PERSONALISED AMBIENT INTELLIGENCE IN BUILDINGS VIA CONTEXT-AWARE AGENTS

systems, the agents interact and negotiate with each

other to solve complex problems which a monolithic

system, such as, a single agent cannot do. Multi-

agent systems is beneficial in the scenario of not

only managing heating and lighting service but user-

interaction, sensors, effectors from having their own

agent and a negotiation mechanism can be applied

between the agents to solve disputes (Mo, 2002).

In a multi-agent system one agent should manage

the sensors: constantly observe for the co-location of

occupants, anticipate user behaviour, analyse social

activity, logging user check-in and check-out time;

another agent can focus on energy savings and

building controls; and one more agent can manage

the interaction between the user to ensure the user

preferences are recorded and sent to an agent that is

responsible for controlling the services.

3.2 Occupants Comfort

The human comfort zone can be used to describe the

occupants’ comfort. The optimal body temperature

is at 36.8

C (Elert, 2005) and the “occupant’s

thermal comfort is dependent on the temperature,

humidity and air of the room” (Meier, 1994). Some

researchers suggest that heating affects the comfort

level and productivity of an occupant that is subject

to heat stress if comfort zones are not met. Lighting

also has an effect on working productivity (Beld,

2001), as the age of the occupant increase their

requirement for lighting also increases. The light

intensity varies amongst people and occupants

accept that having strong light intensity in an office

is very important because a lot of creative work is

performed and lighting should “guarantee sufficient

visual performance for tasks concerned” (Beld,

2001).

Some of the more recent work like MASBO

(Qiao, 2006) uses an approach that balances energy

efficiency, occupants’ preferences and priority to

achieve occupant comfort; Conflict Resolution

Architecture is presented by (Lee, 2008) where they

try to solve the conflicting requirements of

occupants using a “priority” and “privilege-average”

approach. ZhengChun Mo (2002) formulated an

equation for maximising occupant comfort and

minimising electricity costs when one of the agents

receives a conflicting request. This sort of conflict

resolution is only made possible in a multi-agent

system because a single agent would not have

enough parameters for negotiating with itself: a

single agent will serially perform energy

consumption calculations and send commands to the

effectors, but it would not raise the issue of whether

that command is cost effective.

4 CONTEXT-AWARE AGENTS

FOR PERSONALISATION IN

INTELLIGENT BUILDINGS

The metrics for identifying temperature is generally

Celsius (°C), a common measurement in the United

Kingdom, and Fahrenheit (°F). The illuminance of a

light bulb is dependent on the energy used and the

brightness can be measured by the amount of

wattage it consumes. Another metric used to

measure the light emittance is LUX. But, in an office

environment, do the occupants know how bright

their lighting is in LUX? Obviously, they know what

temperature the room is at because of a thermostat or

a thermometer but does it display the actual

temperature: different room dimensions have

different temperature readings which also depend on

where the thermometer or thermostat is placed.

Overall, the occupants are concerned about whether

all the lights are working and are very bright, one

might find the lights too bright but will have to

compromise because they cannot change it, and that

if the temperature is not comfortable then they will

either continuously adjust the thermostat until they

get the desired result or use the air-conditioning.

4.1 User-Interface

The user-interface is designed with a user-centred

approach to minimise user intervention and also

taking into consideration the usability of the system

that the occupants are going to ‘train’ and calibrate

via heuristic question and answer methodology. An

expert system called MYCIN utilised this

methodology to solve medical problems: a series of

closed-questions were used to identify illnesses

(Buchanan, 1984). Heuristic question and answer

can be used as the training simulation to question the

user if they are satisfied with the temperature or

lighting and to fine tune the services until the user

gets their desired results.

ICISO 2010 - International Conference on Informatics and Semiotics in Organisations

1: Is occupant satisfied with temperature/lighting?

2: Increase or decrease the temperature/lighting?

Figure 3: Training Simulation Data Flow Diagram.

The intelligent agent can then learn the occupant’s

preferences by recording the results of the training.

This eliminates the purpose of displaying the

Celsius/Fahrenheit and LUX values because the user

can assess and feel the environment changes in real-

time therefore keeping the user-interface as simple

as possible.

4.2 Multi-Agent System & Hawthorne

Effect

An office environment has many heaters, air-

conditioning and lights. These services can be

segregated into zones where each zone will have a

certain number of occupants. Each zone has a set of

agents for handling the interaction and

communication with its occupants via a user-

interface, sensors to perceive human actions and co-

locations, and managing services within that zone.

The multi-agent system makes up a network of

intelligent agents that is used for the intelligent

building architecture.

UI: User-

Interface

1: User Agent

2: Effector

3: Local

Agent

4: Sensor Agent

5: Environment

Figure 4: Multi-Agent Framework.

The diagram above shows our multi-agent

framework design for the simulation. As the sensors

and effectors are simulated these has been included

in the User Agent and Effector component, as shown

in the diagram in which they communicate with the

User-Interface.

To tackle the problems of sharing services

between occupants as well as trying to maximise

occupant worker productivity the Hawthorne Effect

suggests varying the environment so that it meets all

the occupants’ needs (Draper, 2008). The intelligent

agents will not only calculate the occupant’s desired

temperature and lighting but can also adjust them

both accordingly at consistent intervals. For example,

if one occupant’s preferred temperature is at 24°C

and another occupant has a preferred temperature of

24.5°C then the intelligent agent adjusts the

temperature to fluctuate between 24°C and 24.5°C

so that both occupants are satisfied at several

durations within each hour.

The Hawthorne Effect has been criticised by the

Maslow’s Hierarchy of Needs which states that

increase productivity is dependent on the

individual’s temperament and their need for self-

actualisation as well as job satisfaction rather than

the environment they are in (Maslow, 1943). Other

criticisms have been referenced by Olson (2004) as

well as the research outcome of the Hawthorne

Effect concluded that worker productivity was

temporarily higher because the workers were

motivated by the attention and being observed by the

research team. However, the intelligent agent in this

simulation that utilises the Hawthorne Effect theory

is trying to provide the best possible work

environment for its occupants and to satisfy their

requirements. The agent must also understand that

there could be a possibility of having a large gap

between the two occupants’ preferences which

might result in dramatic changes in their

environment and that should be avoided by

calculating an intermediate result that will benefit

both occupants. For example, if the preferences were

24°C and 26°C then the agent should use

intermediate values of 24.5°C and 25.5°C to reduce

the dramatic effects of having significantly different

temperatures. Although, this method may increase

energy usage as to compare with using the average

value of constant 25°C the intelligent agent’s

priority is for human comfort.

4.3 Simulation Design

The main focus of the simulation is how the agents

respond to the environment but to do this another

PERSONALISED AMBIENT INTELLIGENCE IN BUILDINGS VIA CONTEXT-AWARE AGENTS

part of the system should simulate the environment

itself. Figure 5 illustrates the environment and

external variables as required to help make the

simulation as realistic as possible.

Figure 5: Simulation Design.

The variables external to the scope are simulated

values to stimulate the intelligent agents. These will

be implemented as part of the system.

5 CRITICAL DISCUSSIONS

& FUTURE WORK

A simulation was developed to test the three

scenarios stated in Section 1. As this is an on-going

research project we can make initial evaluations

using test cases of the expected outcome.

5.1 User-Interface

The main objective of the user-interface is simplicity

in communicating with the agent which supports the

concept of “seamless integration” of intelligent

devices. The user-interface is designed as simple as

possible but transmits enough information to the

agent containing adequate intelligence for perception

and anticipation. The user-interface has been

designed to show the temperature and lighting

measurements to the occupants but this could cause

some implications because some users may know

what temperature settings they prefer and not relying

on the agent to perceive the occupants’ feelings for a

better judgement.

5.2 Training the Agent

The system allows the user to ‘train’ the agent and

make it understand the occupant’s preference. The

training uses a set of closed question, for example,

“Are you satisfied with the temperature/lighting?

Yes/No”. If the user selects “No” then it proceeds to

ask “What would you like to do with the

temperature/lighting? Increase/Decrease”. If the user

is satisfied then the system records the user’s

preferences. However, realistically the environment

is unable to adjust immediately within seconds and

therefore this training process could take a very long

time causing a placebo effect with the occupant

thinking that the environment has adjusted to their

preferences, temporarily. Another problem raised

here is the decision making logic of both

temperature and lighting is very simple. Probable

solution could be to use a binary algorithm to

produce better results for the learning agent.

5.3 Hawthorne Effect

In this simulation, the office environment is

segregated into zones. For example, Zone A has two

occupants with different temperature and lighting

preferences. Occupant A prefers the room

temperature to be at 23°C, 500LUX and Occupant B

would like the room temperature to be 25°C,

525LUX. The agent should fluctuate the temperature

and lighting between the two occupants preferences

at a very slow rate, i.e. cycle between these

preferences twice per hour. This method may

maximise user satisfaction but there are two

potential issues: changing the temperature is likely

to result in an increase in energy consumption and

altering the levels of light throughout the day may

cause more issues for occupants. But, as with other

methodologies (Chen, 1995; Meier, 1994; Buchanan,

1984) these resolves conflicts through compromise

but at least the solution here ensures that each

occupant is satisfied for various moments

throughout the day.

A fourth test case scenario was prepared but the

simulation did not support it. The fourth scenario

was to simulate how the agent interacts with the

environment when it senses that all the occupants

have entered the meeting room. From its historical

data the agent knows a meeting is scheduled ten

minutes prior to the meeting that was held regularly

on the same day and time. However, due to the

complexity and lack of development time this part of

the system was not fully implemented. It should

show how the agent tries to cater for all occupants

by calculating the average preference and fluctuating

from that measurement. It should also continue to

monitor who is actually present in the meeting room

and adapt the temperature and lighting to users’

preferences.

In conclusion, the results of this research project

could be criticised for a number of issues. In

hindsight, it could be argued that the use of a

qualitative research methodology would have been

useful in gaining a better understanding of the users’

ICISO 2010 - International Conference on Informatics and Semiotics in Organisations

preferences in terms of the management of heating

and lighting. The project had assumed that the

Hawthorne Effect is preferable with little backing

from building occupants.

The heuristic training method could also be

criticised for its accuracy and its assessment of the

environment. For example, if two users are in the

same room and require a very similar temperature,

they may not be able to tell the difference between

the system’s behaviour in fluctuating between the

two selected heat settings. As a result, it would not

be possible to verify whether the Hawthorne Effect

theory is preferred by the occupants to an alternative

e.g. keeping the temperature to an average of the two

users’ preferences. External factors, such as the

user’s activity within the environment could also be

taken into account i.e. depending on a user’s level of

physical activity; the judgment of the environment’s

heat level could be misjudged and therefore have a

negative impact on the user’s perception of the

system and the heuristic training method.

There are several improvements or further

research that could be made, as listed below:

 The user agent could benefit from a new

learning algorithm. Possibilities of using a

binary search algorithm to locate a particular

value when training the heating or lighting.

 A dedicated agent that is solely responsible for

maintaining performance and efficiency of

energy usage.

 Improving energy efficiency as this was not

included as one of the objectives in this paper.

The agents should take into account the internal

and external ambient environment and adjust

lighting and heating usage efficiently whilst

maintaining a comfortable environment. It is

expected that the proposed solution could

benefit greatly from this, for example, if the

outdoor air temperature is cooler than inside the

office then we can use natural ventilation to

dilute the outdoor cold air with the indoor warm

air. Going green and being sustainable is

currently a major topic. Research into agent

technology could be of great benefit to

implementing sustainable buildings for the

future.

REFERENCES

Shadbolt, N., July 2003, ‘Ambient Intelligence,’ IEEE

Intelligent Systems.

Schilit, B. N., et al., December 2004, ‘Context-Aware

Computing Applications’, IEEE Workshop on Mobile

Computing Systems and Applications.

Kozma, J., April/May 1998, ‘Intelligent agent: Getting

others to do your work for you,’ IEEE Potentials.

Pissinou, N., et al., 1997, ‘A Roadmap to the Utilization of

Intelligent Information Agents: Are Intelligent Agents

the link between the Database and Artificial

Intelligence Communities?’ IEEE.

Flax, B. M., April 1991, ‘Intelligent Buildings,’ IEEE

Communications Magazine.

Chen, H., April 1995, ‘Machine Learning for Information

Retrieval: Neural Networks, Symbolic Learning, and

Genetic Algorithms,’ Journal of the American Society

for Information Science.

Mo, Z., August 2002, ‘Intelligent Buildings and Intelligent

Agents – A Human-Centered Framework for Building

Controls’.

Elert, G., 2005, ‘Temperature of a Healthy Human (Body

Temperature),’ The Physics Factbook. [Online].

Available: http://hypertextbook.com/.

Meier, A., March/April 1994, ‘Some Like It Hot,’ Home

Energy Magazine Online, Home Energy, [Online].

Available: http://www.homeenergy.org/.

van den Beld, G., et al., July 2001, ‘Industrial Lighting

Productivity,’ Ingineria Iluminatului.

Qiao, B., et al., December 2006, ‘A Multi-Agent System

for Building Control,’ Intelligent Agent Technology.

Lee, H., et al., July 2008, ‘A Conflict Resolution

Architecture for Comfort of Occupants in Intelligent

Office’.

Buchanan, B. G., et al., 1984, ‘The Mycin Experiments of

the Stanford Heuristic Programming Project,’ Rule

Based Expert Systems. [Online] Available:

http://www.aaaipress.org/.

Draper, S. W., 2008, ‘The Hawthorne, Pygmalion, placebo

and other effects of expectation,’ University of

Glasgow, Department of Psychology [Online].

Available: http://www.psy.gla.ac.uk/.

Maslow, A. H., 1943, ‘A Theory of Human Motivation’.

Olson, R., et al., January 2004, ‘What We Teach Students

About the Hawthorne Studies: A Review of Content

Within a Sample of Introductory I-O and OB

Textbooks’, The Industrial-Organization Psychologist.

PERSONALISED AMBIENT INTELLIGENCE IN BUILDINGS VIA CONTEXT-AWARE AGENTS