Information Technology Reference
In-Depth Information
2000
(a)
1500
1000
500
08:00
10:00
12:00
14:00
16:00
2000
(b)
1500
1000
500
08:00
10:00
12:00
14:00
16:00
(c)
1500
1000
500
08:00
10:00
12:00
14:00
16:00
1500
(d)
1000
500
08:00
10:00
12:00
14:00
16:00
Time of Day
10
4
10
4
(f)
(e)
10
2
10
2
10
0
10
0
0
200
400
600
800
1000
1200
1400
1600
0
200
400
600
800
1000
1200
1400
1600
Watt
Watt
Fig. 10.2
Energy consumption of
a
House 1 and
b
its Refrigerator over an interval of 8h. Plot
c
and
d
show the changes in power consumption for House 1 and its Refrigerator. The distribution of
power changes that classify the Refrigerator's
on
/
off
states are illustrated in Plot
e
and
f
edges in (c) the overall energy signal which do not belong to the Refrigerator. The
distribution of the edges that classify the Refrigerator's
states are illustrated
in Fig.
10.2
e, f. These distributions can serve as training input for most probabilistic
models.
on
/
off
10.4.2 Appliance State Classification
In this study, we aim at evaluating the appliance state classification task by means
of various machine learning techniques, including Naive Bayes (NB) classifier,
Factorial Hidden Markov Model (FHMM), Classification Tree (CT), and One-
Nearest-Neighbor (1NN) classifier.
We selected thesemodels based on their complementary characteristics and degree
of popularity regarding the energy disaggregation task. Table
10.1
shows typical
characteristics of the considered machine learning techniques [
16
], although the
Table 10.1
Characteristics of
algorithms
NB
FHMM CT
1NN
Fitting speed
Fast
Fast
Fast
Fast
Prediction speed
Fast
Fast
Fast
Medium
Memory usage
Low
Low
Low
High
Easy to interpret
Ye s
No
Ye s
No
Search WWH ::
Custom Search