Information Technology Reference
In-Depth Information
characteristics strongly depend on the underlying algorithm and the problem. There-
fore, Table
10.1
should be considered as a guide for an initial choice of models.
The NB classifier is a simple probabilistic model based on applying Bayes' the-
orem with strong independence assumptions, which has been applied for appliance
and occupancy recognition in various studies [
9
,
12
,
13
,
18
]. Speed and memory
usage of the NB classifier are good for simple distributions, but can be poor for large
datasets [
16
].
The FHMM is a statistical model in which the system under study is assumed to be
a Markov process with unobserved or hidden states. FHMMs have been successfully
applied to the energy disaggregation problem [
8
,
10
,
23
]; however, their complexity
increases with the number of states and the length of the Markov chain [
5
,
8
].
CTs map observations about an item to conclusions about the item's target value,
meaning the predicted outcome is the class to which the data belongs. Decision tree
learning has been proven to be applicable to appliance identification on metering
data in a couple of recent studies [
1
,
18
].
The 1NN classifier is often regarded as the simplest straw man or baseline
approach [
7
], and has been considered for the energy disaggregation task in sev-
eral studies [
12
,
13
,
23
]. 1NN usually has good performance in low dimensions,
but can have poor predictions in high dimensions. For linear search, 1NN does not
perform any fitting [
16
].
10.4.3 Inference of Occupancy
We assume that there exists a direct relationship between appliance usage and occu-
pancy states in residential homes. For instance, if the lighting is turned
on
, we usually
know that the residents are at home, unless someone forgot to turn
the lighting.
Hence, lighting may be a straightforward indicator for occupancy states, enabling
us to verify manually adjusted heating schemes and recommend optimized heating
schedules.
However, heating control is much more complex, because the usage of certain
appliance actually requires to decrease the temperature. For example, when residents
turn
off
the oven or stove, the temperature in the kitchen rises automatically, and
we can reduce heating to save energy, instead of just opening the window. In case
the heating control system would have knowledge about the installation points of
all devices, one could even use the appliance states to control the temperature in
individual rooms.
The knowledge of individual appliance states furthermore allows us to infer
devices that are unrelated to occupancy. For instance, the refrigerator automatically
switches between
on
state every few minutes, no matter if the residents are
at home or not. The same is true for devices in standby mode or appliances such as
the smoke alarm or electronic panels which are constantly drawing power. Therefore,
by just looking at the overall energy consumption of a household it is impossible to
distinguish between occupancy states.
on
and
off
Search WWH ::
Custom Search