Information Technology Reference
In-Depth Information
Fig. 10.1 Framework for heating control and scheduling by means of energy disaggregation tech-
niques
devices. Subsequently, (ii) we use the extracted features as input for the appliance
state classification. For the sake of simplicity, Fig. 10.1 assumes that the individual
appliance models were trained on other households prior to the classification task.
Given the classified
states for each appliance, we can eventually (iii) infer
the occupancy state of the respective household and recommend optimized heating
schedules.
In the following subsections, we describe the (i) feature extraction, (ii) appliance
state classification, and (iii) inference of occupancy in more detail.
on
/
off
10.4.1 Feature Extraction
Given the overall energy consumption of a household and the energy consumption of
the individual appliances in this household, we aim to build a model for each appli-
ance in order to estimate its
states in a previously unknown environment or
household. Since an appliance can be either turned
on
/
off
, the device state iden-
tification can be formalized as a two class problem. For the training of an individual
appliance model, we consider the changes in power consumption that classify the
respective device states. In our approach, the input for the classification model are
two distributions of power changes, which represent the features that characterize
one or the other class.
Figure 10.2 illustrates the feature extraction process on the basis of real-life mea-
surements from the REDDdata set, in particular the energy consumption of (a) House
1 and (b) its refrigerator for a sample time frame of 8h. We can see that (a) the overall
energy consumption is the sum of (b) the Refrigerator's energy consumption and the
energy consumption of other appliances. Given this information, we can derive the
changes in energy consumption by the first-order difference of the power signals.
This step is often referred to as edge detection, since the stable periods in the signal
are filtered out. The edges or changes in power consumption of the overall energy
signal and the Refrigerator signal are shown in Fig. 10.2 c, d, respectively. Knowing
which edges specify (d) the activity of the Refrigerator, we can easily separate the
changes in energy consumption that categorize other devices by considering all the
on
or
off
Search WWH ::




Custom Search