P(C}E)=P(E!C)*P(C)/(P(E|C)*P(C) +
P(E} “C)*(1-P(C) (7)
and
P(C{"E)=(1-P(E;C))#P(C)/((1-P(E}C))+*P(C) +
(1-P(Ei^C))*(1-P(C) (8)
If evidence E expressed as E1 AND E2 AND ...
AND En, then
Decision Programme
Plan & Decision experience
| |
Type-1 Type-2 e oe
| |
[IF] Pmax < minium of threshold value
[THEN] C is not true, exit;
[END]
The three level inference make up a
reasonning net, which can be illustrated as
figure 2.
Advanced
Inference
Classification rules
Factor-2 ....
Factor-1
D enu d Sr
Geographical
Type-n
| Medium
Inference
|
Factor-m
T TI Basic
1 2....mm Inference
lod. 1
Data
Fig. 2 The Reasonning Net of MCGES-GIE
P(ElC)zmin(P(EilS)), i=0, 1, 2, ..., n (9)
The advanced inference of MCGES-GIE uses the
algorithm of backward chaining,which is shown
as following:
bkwdchain(knowledge base: KB, goal: C)
[BEGIN]
(1) scan KB to find rules with C as
conclusion,set NOTUSED tag to those rules and
put them into rule set IRSET;
(2) calculate P(C}E) of all rules in IRSET,
and sequence them according to their P(C}E);
(3) select the most important rule with
NOTUSED tag from IRSET, choose the evidence
with the biggest P(ElC)as new goal C1;
(4) [IF] C1 is unknow
[THEN] manage to calculate C1 or call
bkwdchain(KB,C1);
[ELSE]
a. calcute the probability of C if all
other evidence without C1 support it,
and record it as Pmax;
b. calcute the probability of C if all
other evidence without C1 do not
support it, and record it as Pmin;
Pmin <= maxium of threshold value,
Pmax >= minium of threshold value
(5) [IF]
[THEN]
a. change the NOTUSED tag to HASUSED in
current rule;
b. GOTO (3);
(6) [IF] Pmin > maxium of threshold value
[THEN] C is true, exit;
268
4, THE EXPLAINING MOTHODS OF MCGES-GIE
The explaining module is another important
part of MCGES-GIE.In the inference procedure,
MCGES-GIE allocate a buffer to store all of
the reasonning nodes,reasonning direction and
temoprary results. With user's different
requirement, MCGES-GIE presents three level
explanation based on those information,
(1) explanation for inference results,
including what the results mean, how the
system to get them, which rules are involved;
(2) explanation for rules involved in the
inference procedure;
(3) explanation for all geographical factors.
5. CONCLUSIONS
MCGES-GIE can iminate the inexact inference
of geographers to solve a great deal of
inexact geographical problems, and can
conveniently present explanation in different
levels to meet the user's requirements.MCGES-
GIE has been successfully used in Kouhe Soil
and Water Conservation Expert System.
REFERENCE
[1] Deng Julong, 1982, "Control Problems of
Grey System",Systems & Control Letters,Vol.1,
No.5.
[2] Ma Ainai, 1988,"A Geo-code Model for the
Use of GIS",The 16th ISPRS,Kyoto,1988,Vo1.27,
B4, pp.585-591.
