Escolar Documentos
Profissional Documentos
Cultura Documentos
, 0 < , ();
933 933 941 941 941 944 944 944
= 1 means a
i
connects with a
j
,
= 0 means not
connect, then
= =
0 1 0 1 1
0 0 1 1 1
0 1 0 1 1
1 1 1 0 0
1 1 1 0 0
.
3) Definition3: PA.
= , ,
, , , , ,
,
,
, ,
: ();
: () ();
: () ();
: () ;
: (
) () ;
: (
) ()
.
Among them, PIn is the set of input data state; POut is
the set of output data state; R
io
is the set of possible data
pretreatment program, DL is the collection of data state in
data level; CL is the collection of data state in feature level;
DeL is the collection of data state in decision level; See
shows the process of perception of Agent in initial state; F
cl
shows the collection of program transforming from the
collection of DL to the collection of CL; F
del
shows the
collection of program transforming from the collection of
CL to the collection of DeL; F
do
shows the collection of
program transforming from the collection of DeL to the
collection of POut; Est shows the process of determining
the output data results, through evaluation and learning
feedback for the program of data pretreatment; Act means
the process of focusing on selecting a specific output data
from specific POut, according to the corresponding set of a
program set; () shows the power set of DL, and the
other is also similar.
4) Definition4: RA.
= , , ,
, , , , ,
= {(,
)| ,
()};
= {(
, )|
(), ()};
: () ;
: () () ;
: () () ;
: () () .
Among them, PRn is the input decision-making data ;
RIn is the set consisting of all possible action program of
Agent; ROut is the set of all possible corresponding results
state to the all possible action program; S
pn
shows the
correspondence between sets of decision-making data and
action program; S
io
shows the correspondence between set
of program and set of result state; AIRule expresses the
intelligent selection strategy of micro-intelligent decision
module; See shows the process of determining the current
program set of the corresponding pair, according to the
input decision-making data and AIRule; Est shows the
process of determining possible result state, after ensuring
the corresponding program set through evaluation and
learning feedback; Adjust shows the process of determining
final result state, after ensuring the corresponding program
set through control and coordination according to AIRule;
Act shows the process of focusing on selecting a specific
program from selected programs, according to AIRule and
the corresponding result state set of a program set.
5) Definition5: EXA.
= {, , ,
, , , }
= {(,
)| ,
()};
= {(
, )|
(), ()};
: () ;
: () () ;
: () () .
Among them, ERIn is the input set of decision-making
program; ExIn is the set of all possible control command of
Agent; Exout is the set of all possible corresponding results
state to the all possible control command; S
re
shows the
correspondence between sets of decision-making program
and control command; S
eo
shows the correspondence
between set of control command and set of result state; See
shows the process of determining the current control
command set of the corresponding pair, according to the
input decision-making program; Est shows the process of
determining possible result state, after ensuring the
corresponding control command set through evaluation and
learning feedback; Act shows the process of focusing on
selecting a specific control command from selected control
commands, according to the corresponding result state set of
a control command set.
6) Definition6: EVA.
= {, , ,
, , , }
= {(,
)| ,
()};
= {(
, )|
(), ()};
: () ;
: () () ;
: () () .
Among them, In is the input set of evaluation data; EvIn
is the set of all evaluation program of Agent; EvOut is the
set of all possible corresponding results state to the all
evaluation program; S
ie
shows the correspondence between
sets of evaluation input and possible evaluation program; S
ev
shows the correspondence between set of evaluation
program and set of evaluation result state; See shows the
process of determining the current corresponding set of
evaluation program, according to the input set of evaluation
data; Est shows the process of determining possible result
state, after ensuring the corresponding set of evaluation
program through repeated evaluation and adjustment; Act
shows the process of focusing on selecting a evaluation
program from specific evaluation programs, according to
the corresponding result state set of a evaluation program.
7) Definition7: LFA.
= , , ,
, , ,
= {(,
)| ,
()};
= {(
, )|
(), ()};
: () ;
: () () ;
934 934 942 942 942 945 945 945
: () () .
Among them, In is the input set of learning feedback data;
LFIn is the set of all learning feedback of Agent; LFOut is
the set of all possible corresponding results state to the all
learning feedback program; S
il
shows the correspondence
between sets of learning feedback input and possible learning
feedback program; S
lf
shows the correspondence between set
of learning feedback program and set of result state; See
shows the process of determining the current corresponding
set of learning feedback program, according to the input set
of learning feedback data; Est shows the process of
determining possible result state, after ensuring the
corresponding set of learning feedback program through
repeated adjustment and optimization; Act shows the process
of focusing on selecting a learning feedback program from
specific learning feedback programs, according to the
corresponding result state set of a learning feedback program.
III. THE APPLICATION OF FORMAL MODEL
According to the above proposed model, ADAM
conducts the formal analysis for the process of an
autonomous decision-making. Assuming the process of the
simplification of general autonomous decision-making is as
shown below.
ScnsorN Scnsor1 Scnsor2
Decision1 Decision2 DecisionN
Auiononous
Dccision-naling
...
...
Figure3. Simplified Process of Autonomous Decision-making
Supposing P
i
are the obtained datas from Sensor1,
Sensor2...SensorN, C
i
are the received control commands
from Decision1, Decision2...DecisionN. So the process of
autonomous decision-making can be formalized as:
= {
0,
1
};
: ();
: () ();
: () ();
: () ;
: (
) () ;
= {
, } , ;
: () ;
: () () ;
: () () ;
= {
, } , ;
: () ;
: () () ;
= {
0,
1
} ,
.
As is shown above, the data gets the final set of control
command of autonomous decision-making, that is Con, after
through the data pretreatment of PA, the autonomous
decision-making response of RA and the decision-making
execution of EXA. The process of autonomous decision-
making also gets the constant learning feedback and
evaluation and adjustment from EVA and LFA. Among them,
the key of PA lies in determining appropriate program set
R
io
, optimizing the result setPOut; the key of RA lies in
determining proper decision-making programRIn,
optimizing the set of decision-making resultROut through
control and coordination. While the key of EXA lies in
determining appropriate set of control commandExIn,
optimizing the result set of execution ExOut.
IV. CONCLUSION AND FUTURE WORK
Autonomous decision-making, as the intelligent
decision-making at a higher sense, directly affects the
intelligent level of intelligent system. This paper proposes
an agent formal model for autonomous decision-making,
aiming at the autonomous decision-making process of the
current design and development of autonomous decision-
making system, analyzing the feature of autonomous
decision-making and combining the formal method of Agent.
The ADAM model includes five parts: data pretreatment,
decision-making response, decision-making execution,
decision-making evaluation and learning feedback, this
paper gives their own formal expression, aiming at the
features of each part. Finally, this paper uses ADAM model
to conduct formal analysis for the process of an autonomous
decision-making, and points out the realized critical point of
the model and preliminary demonstrates validity of the
model, providing certain theoretical basis for the design and
development of the system of autonomous decision-making.
The future work includes: further perfect the ADAM model;
theoretically demonstrate the consistency and completeness
of the model; study Multi-Agent model for autonomous
decision-making under the distributed situation.
REFERENCES
[1] N R Jennings. Agent-based computing: Promise and perils[C]. The
16th Int'1 Conf on Artificial Intelligence, Stockholm, Sweden, 1999.
[2] A.Kakas and P.Moraitis, Argumentation based decision making for
autonomous agents, Proceedings of the second international joint
conference on Autonomous agents and multiagent systems, (AAMAS
'03), ACM Press, 2003, 883-890.
[3] Cai Yuanli, Yu Zhenhua and Zhang Xinman. Formal modeling
methodology for Multi-Agent systems[J]. Journal of System
Simulation, 2007, 19(4): 3151-3157.
[4] Clarke E M, Wing J M. Formal methods: state of the art and future
directions[J]. ACM Computing Surveys(S03360-0300), 1996, 28(4):
626-643.
[5] Rao A S, Georgeff M P. BDI Agents: From theory to practice.
In:Proc. of the 1st Int'1 Conf. on Multi-Agent Systems(ICMSA-95).
San Francisco, 1995, 312-319.
[6] Yu Jiangtao. Rearch and application of Multi-agent models, learning
and collaborative[D]. University of ZheJiang, 2003,17-23.
[7] Hu Xiaojian, Yang Shanglin and Luo He. A formalized model on
Agent with probability and Utility operator[J]. Journal of Computer
Rearch and Development, 2006, 43(zl): 56-59.
[8] David Lee Hall. Mathematical Techniques in Multisensor Data
Fusion[M]. Artech House, Norwood, 1992.
935 935 943 943 943 946 946 946