QUERY PROCESSING IN WIRELESS SENSOR NETWORKS - Managing and Mining Sensor Data - page 69

27 2.72

26

2.7

25

24 2.68

23

2.66

22

21 2.64

20

2.62

19

18 2.6

1600 1650 1700 1750 1800 1850 1900 1950 2000 2050 2100 1600 1650 1700 1750 1800 1850 1900 1950 2000 2050 2100

TIME STAMP INDEX TIME STAMP INDEX

(a) (b)

Figure 3.3. The correlation between different measurements

for the three join strategies, according to which the optimizer chooses
the optimal query plan.

6. Other Queries
This section further investigates the model-driven data acquisition,
probabilistic/approximation queries and event detection in WSNs.

6.1 Model-Driven Data Acquisition and
Probabilistic Queries
In WSNs, correlations exist between different kinds of measurements.
Figure 3.3(a) shows the temperature readings in a period of time, while
Figure 3.3(b) shows the voltage level of a sensor in the same duration
in the Intel Berkeley Data Set. It is obvious that there is strong cor-
relation between the two measurements. The two curves are so similar
that we can use one of them to predict the other. Another observation
is that the energy cost of sampling temperature is much larger than that
of retrieving the battery voltage. In order to reduce energy cost, in-
stead of sampling temperature directly, we may first acquire the battery
voltage and then predict the temperature reading. Motivated by this,
Deshpande et al. [31] propose the model-driven data acquisition system
called BBQ, a Tiny-Model Query System.
BBQ handles probabilistic queries. A probabilistic query typically
includes two more parameters than the general queries:

1 . An error bound indicating how much bias from the real value is
acceptable

2 . The confidence threshold of the returned value