AN ADAPTIVE SYSTEM TO CONTROL CONSUMER
ELECTRONICS BASED ON EARTHQUAKE EARLY WARNING
Takashi Kokawa, Ryota Sakamoto, Hitoshi Ogawa, Victor V. Kryssanov
Faculty of Information Science and Engineering, Ritsumeikan University, Kusatsu, Shiga, Japan
Keywords: Consumer electronics network, Earthquake early warning, Constraint satisfaction problem.
Abstract: In Japan, earthquakes happen on a regular basis. There is an earthquake early warning system operated by
the Japan Meteorological Agency. By sensing P-waves, it can notify people a few seconds before the
earthquake reaches them. Since only few seconds are left, it is often difficult for people to act with
consideration on who is where at the moment, what consumer electronics should be turned off, and how to
ensure a family’s safety in their house. We developed a system that supports people with control of
network-connected home appliances when an earthquake occurs. The system utilizes three types of control
rules: general, family-related, and personalized for each type of consumer electronics. As there can often be
operational conflicts if applying the rules simultaneously, a conflict-resolving mechanism is implemented,
based on an optimisation algorithm for weighted constraint satisfaction problems. A pilot study of the
system deployment is described, and its results are briefly discussed in the light of the related work.
Conclusions are drawn, and future plans are outlined.
1 INTRODUCTION
About 2000 earthquakes happen every year in Japan,
so that this country is often called earthquake-ridden.
It is a serious problem to prevent the occurrence of
secondary earthquake-provoked disasters, such as
fires, short-circuits, etc. With the recent advent of
consumer electronics networks, real-time earthquake
information can be received by every household, and
it can thus be used to control consumer electronics
and reduce the risk of the secondary disasters.
Real-time information about the seismic
activity is provided by the earthquake early warning
system (Doi, 2002) operated by the Japan
Meteorological Agency (JMA). By sensing primary
waves (P-waves), this system can notify people, e.g.
via the Internet, several seconds before the
earthquake reaches them.
There has been developed an automatic consumer
electronics control system by the Japan Electronics
and Information Technology Industries Association
(JEITA, 2005). When an earthquake early warning is
received, the system provides services such as
alarming people, stopping gas, opening doors, etc.
This system does not, however, comply with the
specific situation of each particular household (i.e.
information about who live in, where they are, what
they currently do, etc. is not utilized by the system).
In this paper, we propose an intelligent consumer
electronics control system for earthquake-caused
disaster prevention. The system uses family-specific
knowledge and provides for generally a higher level
of safety for the family members than other systems
having similar goals.
The system realizes an agent-based architecture,
and it is thus quite reliable in earthquake conditions.
Agents are installed in every room (e.g. of a house).
Each room agent is autonomous, monitors electronic
appliances and the human status in the room, and
controls the electronics when an earthquake happens.
For the control, countermeasure agents processing
different types of rules are developed. During an
earthquake, a countermeasure agent receives
earthquake information and selects appropriate
constraints from an electronics control set. A room
agent operates the electronic appliances based on the
constraints. As there can often be conflicts when
applying control rules obtained from different
countermeasure agents simultaneously, a conflict-
resolving mechanism has been implemented, and the
system finds an optimised control set of rules by
solving a weighted constraint satisfaction problem
with achievement parameters.
250
Kokawa T., Sakamoto R., Ogawa H. and V. Kryssanov V. (2006).
AN ADAPTIVE SYSTEM TO CONTROL CONSUMER ELECTRONICS BASED ON EARTHQUAKE EARLY WARNING.
In Proceedings of the International Conference on e-Business, pages 250-256
DOI: 10.5220/0001426202500256
Copyright
c
SciTePress
In the following sections, the earthquake early
warning system is first outlined. Second, the
proposed system architecture is presented. Next, it is
explained how the system agents act. The control of
home appliances in an earthquake situation is
analysed through a pilot study. Finally, related work
is discussed, and conclusions are given.
2 THE EARTHQUAKE EARLY
WARNING SYSTEM
All earthquakes produce two types of shock waves:
primary waves (P-waves) and secondary waves (S-
waves). P-waves arrive first and rarely cause any
damages. S-waves arrive next and often result in
destruction and loss of lives. The earthquake early
warning system operated by JMA utilizes current
seismic data, such as the magnitude of an earthquake
and the place of its occurrence, obtained by sensing
and processing P-waves. Since P-waves are
normally propagated about twice as fast as S-waves
(excepting for the case of epicentral earthquakes),
the system can provide earthquake information
seconds to tens of seconds before the damaging
wave hits an area. The system is presently deployed
on an experimental basis, and various organizations
and companies participate in the experiment.
In the presented study, a program developed by
the Japan Weather Association and the Earthquake
Research Institute at the University of Tokyo is used
to calculate the expected seismic intensity and time
of the S-wave arrival at a specific location for a
given earthquake.
3 AGENT-BASED
ARCHITECTURE
To achieve a high level of operational robustness
and autonomy, the proposed system implements a
multi-agent architecture. There are several types of
agents.
A room agent monitors a room to get the personal
status and the consumer electronics status, and it
then saves the data to the household status database.
The method of obtaining information may differ for
each agent, e.g. who is in the room can be monitored
using either web-camera or RFID tags. A
countermeasure agent proposes countermeasure
rules for earthquake disaster prevention, which are
specific to the place of a room agent’s installation.
When an earthquake occurs, all countermeasure
agents propose control actions, which may be
inconsistent because of different knowledge
possessed by the agents. A conflict resolution
method is used to find a balanced solution for all the
agent surroundings.
The system operates as follows (
Figure 1). An
earthquake information agent (EIA) at each house is
a “JAVA wrapping” of the program processing
earthquake early warning data. An EIA receives an
earthquake early warning from JMA and calculates
the S-wave arrival time and the expected seismic
intensity. The EIA sends a message to three
countermeasure agents, which are an earthquake
countermeasure agent (ECA), a personal care agent
(PCA), and a precondition for consumer electronics
control agent (PCCA). The ECA utilizes the
knowledge of general countermeasures for
earthquake disaster prevention. The PCA applies the
knowledge of personalized countermeasures by
utilizing family-related information. The PCCA
makes use of the knowledge of specialized
countermeasures to control home appliances. The
countermeasure agents propose constraints to room
agents. After a room agent communicates or
attempts to the countermeasure agents to update its
rules, it generates, through resolving achievement-
weighted constraints, a set of instructions to control
the electronics set up in the room.
home
J
M
A
Eart hquake Early
Wa r n i n g
room
CE
control
count ermeasure agent s
eart hquake informat ion
inst r uct ions
Room Agent
ECA PCCA PCA
Eart hquake
Inf or mat ion Agent
Household St at us
Dat a Base
const raint s
CE and human
st at us
Consumer
Electronics
Figure 1: The system architecture.
AN ADAPTIVE SYSTEM TO CONTROL CONSUMER ELECTRONICS BASED ON EARTHQUAKE EARLY
WARNING
251
4 CONSTRAINTS FOR THE
ELECTRONICS CONTROL
A room agent decides on consumer electronics
control by solving an achievement-weighted
constraint satisfaction problem, AWCSP (Kokawa
and Ogawa, 2004). The AWCSP solver is an
enhanced reasoning method for constraint
satisfaction problems (CSP) that allows for
obtaining an optimised solution when constraints are
too strict. The AWCSP is represented with a set of
variables, a domain of values for each variable, and
a set of constraints in a way similar to the classic
CSP (Walliser et al., 2004). It also requires a set of
constraint weights and a set of constraint
achievement degrees defined.
Consumer electronics states and human actions
are represented with variables as follows: a set of
consumer electronics states, CE = {ce
1
, ce
2
, ce
3
, …,
ce
n
}, where n is the number of electronic appliances
installed; a set of human actions, ACT = {act
1
, act
2
,
…, act
m
}, where m is the number of the family
members. Domain sets of the variable values are
represented as follows: D
CE
= {
1
ce
d
,
2
ce
d
, …,
n
ce
d
},
and D
ACT
= {
1
act
d
,
2
act
d
, …,
m
act
d
}, respectively.
Constraints for the variables are produced by the
countermeasure agents by utilizing the relevant
knowledge. In a room agent, a set of constraints is
represented as C = {c
1
, c
2
, …, c
k
}, where k is the
number of constraints which are sent from
countermeasure agents.
The ECA handles general countermeasure rules
for earthquake disaster prevention, which are usually
pre-defined. Constraints of the agent are represented
as ECc
j
, where j is the rule number. The PCA deals
with family-related rules by utilizing information
about the current status of each family member (as
well as of other people in the rooms), and the
corresponding constraints are represented as PCc
m
.
The PCCA processes personalized rules based on a
control policy defined for the house, and its
constraints are represented as PCCc
m
. The policy
can be set up to allow for sending camera-recorded
data to a server; it would then be possible to use the
video for rescue operations or in an analysis.
Below, this is a fragment of the knowledge of
countermeasure agents:
state(A, B) means that A implies B.
ECA:
Rule1: IF (Seismic intensity >= 2)
Send a Constraint
ECc
1
{state(ce
i
, Speaker), ce
i
=
“Announce(earthquake
information)”, i = 1,…, n}
Rule2: IF (Seismic intensity >= 3)
Send a Constraint
ECc
2
{state(ce
i
, Heater), ce
i
=
“OFF”, i = 1,…, n}
Rule3: If (Seismic intensity >= 4)
Send a Constraint
ECc
3
{state(ce
i
, Door), ce
i
=
“Open”, i = 1,…, n}
Rule4: If (Seismic intensity >= 4)
Send Constraints
ECc
4
{act
h
= “Hide under
furniture”, h
is a person}
ECc
5
{state(ce
i
, all electronics),
ce
i
= “Off”, i
= 1,…, n}
PCA:
Rule1: IF (There is a person h)
Send a Constraint
PCc
1
{state(ce
j
, light), ce
j
= “On”,
ce
j
is near h}
Rule2: IF (Seismic intensity =< 2 and a
person h is sleeping)
Send a Constraint
PCc
2
{act
h
= “Keep sleeping”}
Rule3: If (Seismic intensity >= 3 and
“There is a child at home”)
Send a Constraint
PCc
3
{act
h
= “Accompany the child”}
PCCA:
Rule1: If (Seismic intensity =< 3 and
“Video is recording”)
Send a Constraint
PCCc
1
{state(ce
i
, Video), ce
i
!=
“On” , i = 1,…, n }
Rule2: If (Seismic intensity >= 5 and
“Agreement for recording”)
Send a Constraint
PCCc
2
{state(ce
i
, Camera), ce
i
=
“Record” , i =
1,…, n}
Owing to the different knowledge bases of the
countermeasure agents, constraints generated by the
agents may often conflict. For the conflict resolution,
a method that utilizes weights assigned to the
constraints has been proposed (Lau, 2002). A set of
constraint weights is represented as W = {w
1
, w
2
, …,
w
k
}, k is the number of constraints. When only the
constraint weights are used, some rules may get
completely ignored, e.g. when the seismic intensity
is over 4, the ECA sends ECc
4
for room agents. If
there is no furniture to hide under it in the room, the
ECc
4
is never satisfied, and the room agent produces
no instructions. The room agent should, however,
ICE-B 2006 - INTERNATIONAL CONFERENCE ON E-BUSINESS
252
recommend a substitute action for the person, such
as “Stay away from dangerous objects”. If one tries
to represent this agent behaviour with weighted
constraints, the rules become complicated, and it
becomes difficult to maintain the integrity of the
agent rules.
To cope with this problem, achievement degrees
are defined for the constraints. If a constraint is fully
satisfied, the room agent chooses a variable value
having the highest achievement degree. An
achievement degree is represented as a parameter
showing to what degree the constraint is achievable
in the given situation. The achievement parameter
thus consists of a variable and a threshold. The set of
achievement variables A = {a
1
, a
2
, …, a
k
} represents
the “satisfaction degree”. The set of achievement
thresholds F = {f
1
,f
2
, …, f
k
} gives thresholds that the
values of a
k
must achieve to make the constraints
satisfied.
The achievement degrees for ECc
4
are presented
in
Table 1. ECc
4
is to issue recommendations that
would help maintain a higher safety level for the
family members. A safety level is a mapping to the
achievement degree of the constraints; the safety
level of the action defined in a constraint makes the
achievement degree 100%, the lowest level
corresponds to the achievement degree 0% (i.e. no
action).
The achievement degrees for consumer
electronics -related constraints, such as ECc
2
, are
determined by the number of appliances actually
turned off.
Table 1: Safety levels and recommended actions.
Safety level Action group
5 Hide under furniture (table, etc.)
4 Stay away from dangerous objects
(window, vase, etc.)
3 Be accompanied.
2 Make a contact (via IP phone,
video phone, etc.)
1 Report location.
0 No action required.
Weights are assigned to the constraints at the time
of constraint definition. The general
countermeasures and maintaining the family
members’ safety have higher priorities and, hence,
higher weights, while other rules have lower weights
(see
Table 2). Achievement thresholds for the
constraints are dynamically calculated by the
countermeasure agents, depending on the time
remaining until the earthquake and current
information collected by room agents.
Table 2: Constraint Weights.
Constraint Weight
ECc
1
4
ECc
2
6
ECc
3
5
ECc
4
5
ECc
5
3
PCc
1
4
PCc
2
3
PCc
3
5
PCCc
1
1
PCCc
2
2
Assignments of values to the variables are made
via an optimisation procedure. Room agents produce
sets of assignment values corresponding to
achievement degrees higher than the achievement
threshold of a constraint. Room agents calculate M
an “optimisation degree” of the set as the following
sum:
∑
=
=
n
i
i
wM
1
(if f
i
= 100)
i
ii
i
f
fa
w
−
−
×
100
(else).
If a constraint c
i
is completely satisfied, the
corresponding summand is the constraint’s weight w
i
.
Otherwise, the added value is the product of the
constraint’s weight w
i
and the distance of the
achievement degree value a
i
to the achievement
threshold f
i
; n is the total number of constraints.
5 PILOT STUDY
In this section, we describe an example of the
application of the developed system to control the
situation in a room with a predefined environment
during an earthquake (also, see the definitions of the
constraints in the previous section).
The environment of room A:
State(ce
1
, TV),
1
ce
d
= {“Entertainment channel”, “News
channel”, “Sports channel”, “Off”};
State(ce
2
, Speaker),
AN ADAPTIVE SYSTEM TO CONTROL CONSUMER ELECTRONICS BASED ON EARTHQUAKE EARLY
WARNING
253
2
ce
d
= {“On”, “Announce(x)”, “Off”};
State(ce
3
, Room light),
3
ce
d
= {“On”, “Off”};
State(ce
4
, Corridor light),
4
ce
d
= {“On”, “Off”};
State(ce
5
, Camera),
5
ce
d
= {“Record”, “On”, “Off”};
State(ce
6
, Heater),
6
ce
d
= {“On”, “Off”};
State(ce
7
, IP phone),
7
ce
d
= {“Connect(room y)”, “Off”};
State(act
1
, Family member),
1
act
d
= {“Watch TV”, “Hide under
furniture”, “Stay away from
dangerous objects”, “Be accompanied”,
“Make a contact”, “Report location”,
“No action required”};
ce
1
= “Entertainment channel”,
ce
2
= “On”,
ce
3
= “On”,
ce
4
= “Off”,
ce
5
= ”Off”,
ce
6
= ”On”,
ce
7
= ”Off”,
act
1
= ”Watch TV”
There is a child in room B.
(Other details
of the surroundings of room B are omitted.)
Case 1:
The system processes an earthquake early
warning: the seismic intensity is 5 and the remaining
time is 15 seconds.
Table 3 lists constraints for the
room agent that are sent from countermeasure agents.
Achievement thresholds are set as follows. The
constraint c
2
“turn off all the heaters” must be
completely satisfied, and its achievement threshold
is therefore 100%. It is possible to simultaneously
announce both the earthquake information and
instructions to the people, and the achievement
threshold of “announce the earthquake information ”
is 100%. Since the domain of c
8
is binary, the
achievement threshold f
8
is 0%. c
5
does not have to
be fully satisfied if there is a higher priority
constraint, and the achievement threshold f
5
is 20%,
because there are only “less dangerous” electronic
appliances in the room (if the room is a kitchen, this
achievement threshold would however be more than
50%). As there is no door lock in room A, the
constraint c
3
does not have to be satisfied, and the
value of f
3
is 0%. There is a person in room A,
therefore lights must be turned on near the person,
and the value of f
6
is set 100%. The achievement
thresholds f
4
and f
7
are 40% and 60%, respectively,
because there is a child in room B, who should be
contacted via the IP phone or accompanied.
There is a conflict between c
4
and c
7
: there is no
furniture in room B. Depending on the remaining
time, the safety levels of the family member in room
A and of the child in room B are balanced by the
corresponding achievement thresholds. If there is
enough time to go to the child’s room, act
h
is
specified as follows: “Hide under furniture” and
“Accompany the child”. Based on the values of M
calculated for each possible scenario, the system
generates the following set of variable values (this
set corresponds to the greatest M obtained):
ce
1
= “Off”,
ce
2
= “Announce (earthquake
information, ‘Go to
room B’)”,
ce
3
= “Off”,
ce
4
= “On”,
ce
5
= “Record” (to record the result of
the family member’s action),
ce
6
= “Off”,
ce
7
= “Off”,
act
1
= “Accompany the child”
Achievement degrees assigned to the variables are
given in
Table 3. The optimisation degree M is equal
approximately to 24.
Table 3: Constraint parameters in Case 1.
c
i
Received
constraint
w
i
f
i
a
i
c
1
ECc
1
4 100 100
c
2
ECc
2
6 100 100
c
3
ECc
3
5 0 0
c
4
ECc
4
5 40 60
c
5
ECc
5
3 20 57
c
6
PCc
1
4 100 100
c
7
PCc
3
5 60 100
c
8
PCCc
2
2 0 100
Case 2:
Same as Case 1, but the remaining time is 5
seconds. The achievement threshold f
4
becomes 80%,
as for anyone in room A, it is now impossible to
reach room B before the earthquake hits. The action
recommended is to hide under the desk, and the
countermeasure to increase the safety level of the
child may, depending on the phone location, be to
talk to the child via the IP phone. The recommended
ICE-B 2006 - INTERNATIONAL CONFERENCE ON E-BUSINESS
254
actions to control the consumer electronics are then
as follows (also see
Table 4):
ce
1
= “Off”,
ce
2
= “Announce (earthquake
information, ‘Hide under the desk
and talk to the child via the IP
phone’)”,
ce
3
= “On”,
ce
4
= “Off”,
ce
5
= “Record”,
ce
6
= “Off”,
ce
7
= “Connect (room B)”,
act
1
= “Hide under the desk and use the
IP phone”
Table 4: Constraint parameters in Case 2.
c
i
Received
constraint
w
i
f
i
a
i
c
1
ECc
1
4 100 100
c
2
ECc
2
6 100 100
c
3
ECc
3
5 0 0
c
4
ECc
4
5 80 100
c
5
ECc
5
3 20 43
c
6
PCc
1
4 100 100
c
7
PCc
3
5 60 66
c
8
PCCc
2
2 0 100
The calculated optimisation degree M is equal
approximately to 23.
Case 3:
Same as Case 1, but the seismic intensity is
≤
2.
The expected effect of the earthquake is not severe.
The system only announces the earthquake
information and/or changes the TV channel to news.
There is only one constraint ECc
1
, and it can be fully
satisfied.
6 A LITTLE RELATED WORK
Most of the studies reported in the literature deal
with earthquake early warning systems to merely
provide for efficient and effective announcement of
the earthquake information (e.g. Doi, 2002; Wu et
al., 2004). There were, however, reports in the past 2
years about developed systems that have goals and
capabilities similar to the ones pursued in the
presented study.
The system proposed by the JEITA (JEITA, 2005)
is an automatic consumer electronics control system
developed in Japan. A similar system was developed
by the Seismic Warning Systems Incorporated
(SWS), using the earthquake early warning in the
west coast of the U.S (SWS, 2004). These systems
can control the electronics for the earthquake
secondary disaster prevention by using the
corresponding earthquake early warning systems:
e.g. they can shut off gas, issue warnings, open door
locks and so on. The systems have, however, to have
countermeasures defined for every possible scenario
in advance, and if the environment changes, the
recommended actions may become ineffective or
even dangerous. Besides, the consistency of the
system knowledge bases is hard to maintain due to
the changing surroundings.
The system developed in the presented study is
able to adapt to dynamic environments, as each
room agent monitors its room and updates the
system knowledge- and data-bases. Furthermore, the
proposed system ensures a high level of operational
transparency by providing access to its data- and
knowledge-bases for its users. This is a unique
feature that, to our knowledge, is not present in any
other systems having similar purposes.
7 CONCLUSIONS
In the presented study, a prototype of an intelligent
adaptive system to control consumer electronics
based on earthquake early warning has been
developed. The system has an agent-based
architecture, and it controls home electronic
appliances via solving an achievement-weighted
constraint satisfaction problem. A pilot-study of the
deployment of the system was outlined, and some
typical situations of the system operation were
considered.
Presently, the system works with a simulator to
examine various households and earthquake-related
situations. We are planning to test the system in a
real environment in the near future.
REFERENCES
Doi, K., 2002 Earthquake early warning system in Japan.
In: Jochen Zschau u., Andreas N., Kueppers (eds),
Eary Warning System for Natural Disaster Reduction,
pp. 447-452, Springer, Berlin.
Lau, H.C., 2002. A New Approach for Weighted
Constraint Satisfaction. Constraint Journal, 7, 150-
165.
AN ADAPTIVE SYSTEM TO CONTROL CONSUMER ELECTRONICS BASED ON EARTHQUAKE EARLY
WARNING
255
Yokoo, M., 2000. Distributed Constraint Satisfaction:
Foundations of Cooperation in Multi-agent Systems.
Springer-Verlag, New York .
Walliser, M., Branschen, S., Calisti, M., Hempfling, T.,
2004. Constraint Satisfaction Techniques for Agent-
Based Reasoning. Brinkhauser Verlag, Berlin.
Kokawa, T., Ogawa H., 2004. A decision-making system
in consideration of one's tastes. In: The Institute of
Electronics, Information and Communication
Engineers(IEICE) Technical Report, Vol. 104, No.
133, pp. 1-6, Tokyo, Japan, 2004.6.21, (In Japanese).
JEITA, 2005, “IT automatic disaster prevention system”,
second press release, 2005.12.7, (In Japanese).
SWS, 2004, “Corporate Backgrounder”, Corporate
Overview, 2004.7.30.
Wu, Y. M., T. L. Teng, N. C. Hsiao, T. C. Shin, W. H. K.
Lee and Y. B. Tsai, 2004, Progress on Earthquake
Rapid Reporting and Early Warning Systems in
Taiwan, In: Y. T. Chen, G. F. Panza, Z. L. Wu (eds),
Earthquake Hazard, Risk, and Strong Ground Motion,
pp. 463-486, Seismological Press, Berlin.
ICE-B 2006 - INTERNATIONAL CONFERENCE ON E-BUSINESS
256
