3 SMART ENVIRONMENT AND AGENT-BASED SIMULATION OF SMART
3.1 Smart Environment
A smart environment is any design that utilizes knowledge of the environment’s context to aid accomplishing various tasks such as navigation, scheduling energy consumption,
notification of activities and evacuation planning. The term “smart” refer to the ability of the
to make decisions or to provide valuable information. The contextual information is defined as
features of information of any entity in the smart environment. For example, the location of
people, the gesture or posture of an individual, the states of certain objects such as doors or
computers, the events ongoing and the activities performed by occupants are all considered to be
contextual information of the smart environment. To collect contextual information, a typical
smart environment is equipped with various sensors, including those detect the density of the
gas, those detect whether there is an occupant inside the detection area, those detect various
states of the objects and those can measure the position of occupants directly. The data obtained
from these sensors are fused in a central control system for handling configurations of energy
supply, air conditioner or providing information related to the current context to the occupants.
Consider a scenario that a smart home application is capable of controlling the cooling system.
We assume the activity of occupants can be labeled as three different categories: sleeping, awake
and absent. From various sensors deployed in the home, the application uses some algorithm to
discover the activity of the occupants. For example, if the occupancy sensor in the bedroom
reports that there is no individual in there, then the application can conclude there is 90%
probability that the occupants are in the state of being awake or absent. By knowing this, the
smart home application turns off the cooling system in the bedroom to save energy consumption.
Moreover, by applying machine learning technology and providing a set of training data to the
smart home application, the application is able to predict when occupant will perform certain
activities. In the bedroom example mentioned above, it is possible for the smart environment
application learn from the data and discover that the occupant normally goes to bed at around
system at around 9:30PM so that when the occupant goes to bed, the air is already cooled and his
comfortableness is maximized.
The information gathered by sensors in a smart environment can be categorized based on
two different resolution criteria: spatial resolution and logical resolution. The spatial resolution
defines the spatial granularity of occupancy problem to be investigated. Under high resolution,
each individual occupant is considered to be the research subject while under low resolution, a
group of individuals in a specific area of the building are considered to be the research subject.
The logic resolution defines the logic granularity of occupancy problem to be investigated. Under high logic resolution, answers to questions such “where is A” are the outcome of the solution to the occupancy problem. Under low logic resolution, answers to question such as “what is happening in the building” are the outcome of the solution to the occupancy problem. The purpose of this dissertation is to build an estimation framework that is capable of estimating
the building occupancy at different spatial resolutions so as to handle different situations. On the
other hand, the possible extensions on integrating questions from different logical resolutions can
also be added. The different resolutions of occupancy problem are illustrated in figure 3.1 and
3.2 where 3.1 presents the spatial resolution of the occupancy problem and 3.2 presents the logic
resolution of the occupancy problem.
(a) Spatial high resolution (b) Spatial low resolution
Figure 3.2 Logic Resolution
Among various information that can be derived from sensors deployed in a smart
environment, the most important one is the location information of occupants, i.e, the occupancy
of the building. This is because occupancy directly related to many aspects of contextual
information of a building such as energy consumption and occupants’ activity. In terms of
energy consumption, occupancy directly informs which parts of the building are in need of
energy and which part of the building do not need energy at all so that the central control system
can effectively handle the energy supply. In terms of activity discovering, occupancy is the key
factors to recognize the ongoing activities of occupants. For instance, when the occupants are
holding a meeting in the building, the detected occupancy in the conference room will be
dramatically increased, in this case, the pattern of occupancy change directly indicates the
activity performed by occupants. Because of the importance of location information, in this
dissertation, we focus on deriving location information of occupants from sensors equipped in
the building. Other research topics related to smart environment such as activity discovery,
activity recognition and learning from data to building occupancy model are out of the scope of
we assume that the smart environment is equipped with binary proximity sensors, which reports
1 when one or more occupants are within its sensing area and reports 0 otherwise. These sensors
have low resolution in detecting the movement of occupants and have the following features in
terms of resolution, accuracy and ambiguity:
1. The sensors provide anonymous position information of occupant. Binary
proximity sensors do not provide measurement of occupant’s identity.
2. Location information obtained is ambiguous when there are multiple occupants in
the sensing area of a sensor. The binary proximity sensor does not have the capability of
identifying the number of occupants in its detection area.
3. Data collected are subject to sensor errors and environmental noise.
4. Sensors are deployed sparsely. The occupants’ locations become latent variables while being outside a sensor’s detection area.
These features make it impossible to directly measure occupants’ positions using binary
proximity sensors. To overcome the difficulties brought by these features, we build an agent-
based model to simulate the occupants’ movements and use particle filter algorithms to assimilate sensor data into the agent-based model dynamically. The data assimilation provides
estimations of occupants’ states in real time, which can be used to supply initial conditions for future simulations to improve simulation results.