Database Reference
In-Depth Information
A typically probabilistic query in BBQ is:
Q9:
SELECT
nodeID, temperature
0.1, confidence(0.95)
FROM
Pressure AS P
P.pressure > threshold
)
WHERE
In the above query, the user asks for the temperature readings of
nodes with ID from 1 to 8. The user specifies an error bound 0
.
1
◦
C
on temperature, meaning it is acceptable if the difference between the
result r and the real value r* is no more than 0
.
1
◦
. In addition, the user
requires the probability that r* falls in the range [
r
0
.
1
◦
C,r
+0
.
1
◦
C
]
is at least 95%, which is defined by the confidence parameter. BBQ
builds a probability density function (pdf), for each attribute, according
to the historical data. Suppose the sensor measures n attributes. The
pdf is a function with n variables with the form p(
X
1
,X
2
,,X
n
). When a
query on attribute
att
i
arrives, BBQ first marginalizes pdf with respect
to
att
i
. It uses the marginalized pdf to compute the most possible value
r
of
att
i
, i.e., the expected
att
i
value. Then it calculate the probability
pr
that the real value of
att
i
falls in the range [
r−,r
+
], where
is the
given error bound. If
pr
satisfies the confidence parameter,
r
is returned.
Otherwise, BBQ physically retrieves a new sample of
att
i
from sensors.
BBQ also handles probabilistic range queries. For instance, “check if
the temperature at the 2nd node falls in the range [20
◦
C,
25
◦
C
]”. The
procedure is similar. First, BBQ computes the marginalized pdf with
respect to the temperature attribute. It then computes the probability
that the expected temperature reading falls in the query range. If the
probability is high (low) enough, BBQ is confident to returns true (false).
Otherwise, BBQ is not confident enough to make the decision and it
physically retrieves new samples from sensors to answer the query.
−
6.2 Event Detection
Event detection is an important application in WSNs. In forests, peo-
ple use event detection to predict potential fire disasters. In factories,
event detection is used to monitor abnormal machinery activities. Ad-
vised by the historical data, users of a WSN gain knowledge on the range
of the sensor readings. For instance, in a factory, the temperature near
machines is normally from 30
◦
C
to 60
◦
C
andthehumidityinabarn
is from 30% to 50%. If sensors report temperatures higher than 60
◦
C
or humidity below 30%, then somewhere in the factory is in danger of
catching fire.
Condition-based
maintenance detects sensor readings fall
out of the normal range, where conditions define the normal ranges of
sensor readings. A condition consists of three attributes:
