7. Imagine when you
come home ...
... your front door opens on its own
... lights turn on automatically
... your fridge is filled
... the pets are already fed
7
12. Context-sensitive
• act application-specific
lighting for a party
• context triggered actions
the cake comes in
usually achieved using Machine Learning
12
13. Adaptive
• our habits change
summer vs. winter ...
• different persons have
different perceptions
male vs. female ...
usually achieved with Neural Networks
13
14. Capturing the
Environment
Time == 2pm
Month == September
Date == 21
Humidity == 35%
Luminosity == 100 lx
Location == 30.12 , 41.21, 8.51
...
we will end up with millions of rules!
14
21. invisible. It has removable floor and ceiling tiles, lots of space for equipment
and customized electrics, which allow us to reconfigure lights, wall sockets and
Example: Lighting Control System
switches as needed. A picture of the smart home is shown in figure 1.
21
22. Example: Lighting Control System
Inputs
outdoor light level
person activity
time
Outputs
ceiling light power
venetian blinds position
22
24. Example: Lighting Control System
at home
absent
1
0
0 255
Person activity
Sensor gives either 0 or 255 (binary)
24
25. Example: Lighting Control System
t1 t2 t3 t4 t5
1
...
0
-20 0 120 1440
Time
1440 minutes mapped on 50 ‘zones’
25
26. Example: Lighting Control System
on on
off off
1 1
0 0
0 255 0 255
Ceiling Blinds
Override: on/off Override: on/off
26
27. Example: Lighting Control System
quite small quite much
much
small normal
1
0
250
0 250
Output 1: Ceiling Light Power
Defuzzify using ‘Center of Gravity’
27
28. Example: Lighting Control System
down up
closed up
closed center
1
0
250
0 250
Output 2: Venetian Blinds Position
Defuzzify using ‘Center of Gravity’
28
29. event-based control.
Example: Lighting Control System
Table 1. An example of a rule table
Example Rule
Fuzzify input, map to output and defuzzify output
Table 2 shows all the possible types of rules used and the possible values
in the rule table with the used rules. In autonomous control, the override flags
of outputs on the input side are defined to be off, marked with number one.
The output states on the input side are marked with zeros, so that the state
of an output is ignored during the input aggregation. All the other values of 29
31. ... But this system can learn
its rule table without prior
knowledge!
31
32. Learning Process
Data
Fuzzification
Data
Filtering
Sensors Server
Rule Database
Update
Fuzzy
control
process
32
33. Automatic Data Gathering
• Monitor Input and Output devices
• Record their values periodically
• Reasonable Timer: 1 minute
Data
Fuzzification
Data
Filtering
Sensors Server
Rule Database
Update
Fuzzy
control
process
33
34. Data Fuzzification
• Read recorded input and output values.
• Determine membership function with
greatest degree of membership.
• Store fuzzy value for later use in learning
process.
Data
Fuzzification
Data
Filtering
Sensors Server
Rule Database
Update
Fuzzy
control
process
34
35. Data Filtering
• Search most common combinations of
inputs and outputs within a time period.
• Time period no longer than one fuzzy
time unit.
Data
Fuzzification
Data
Filtering
Sensors Server
Rule Database
Update
Fuzzy
control
process
35
36. Rule Base Updating
• Search database for input combinations
determined in previous step.
• If not found: add rule with small weight
• If found: increase/ decrease weights
• If weight becomes 0: remove
Data
Fuzzification
Data
Filtering
Sensors Server
Rule Database
Update
Fuzzy
control
process
36
37. Discussion
• System well suited for pro-active control
• Learns behavior quickly
• Needs tweaking of values and thresholds
• Timer too small: data explosion
• Timer too long: behavior not adaptive enough
37
38. Still there are many more
problems to solve...
Scale system up to hundreds of
sensors and thousands of rules?
Control Interfaces?
Interaction between
controller systems?
38
39. Research Work Covered
A.Vainio et al. : Learning and adaptive fuzzy
control system for smart home.
H.Sunghoi et al. : Adaptive Type-2 Fuzzy Logic
for Intelligent Home Environment.
Minkyoung Kim et al. : Behavior Coordination
Mechanism for Intelligent Home.
39