| weighted_avg(Zout,Zmem,Wout,Wmem,tnorm,step)
|
%% weighted_avg.m
%% Function to extend the weighed average of crisp numbers to
%% the case where all the quantities involved are type-1 sets.
%% It implements the generalized centroid in Equation (9-11) in
%% the book Uncertain Rule-Based Fuzzy Logic Systems: Introduction
%% and New Directions, by Jerry M. Mendel, and published by
%% Prentice-Hall, 2000.
%% Written by Nilesh N. Karnik - July 22,1998
%% For use with MATLAB 5.1 or higher.
%% NOTE : This program uses an algorithm similar to
%% "extend_nary2.m"; therefore, it is slow, but uses less memory
%% than "extend_nary1.m".
%% Outputs : "out" and "mem" (column vectors) are, respectively,
%% the domain and the memberships of the result of the extended
%% weighted average operation.
%% Inputs : "Zout" and "Zmem" are "M x Nz" matrices, containing the
%% domain elements and memberships of the "M" type-1 sets "Z_i"s - see
%% Eq. (9-11). "Wout" and "Wmem" are "M x Nw" matrices, containing
%% the domain elements and memberships of the "M" type-1 "W_i"s - see
%% Eq. (9-11). The domain of each "Z_i" is assumed to contain "Nz"
%% elements and that of each "W_i" is assumed to contain "Nw" elements.
%% If "tnorm" (scalar) is < 0, minimum t-norm is used, otherwise
%% product t-norm is used. Here, "step" (scalar) is a required
%% parameter. In order to save memory, the domain of the output type-1
%% set is taken to be "[minz : step : maxz]", where
%% "minz = min(min(Zout))" and "maxz = max(max(Zout))". Note that
%% since this is a weighted average of the "Z_i"s, the smallest possible
%% domain point in the output set is "minz" and the largest possible is
%% "maxz".
%% Note 1 : The t-conorm used is maximum.
%% Note 2 : The notation in this program is somewhat different than
%% that in "extend_nary1.m" and "extend_nary2.m", but it conforms with
%% the notation in Eq. (9-11) .
%% Note 3 : Since type-reduction calculations can be represented as
%% extended weighted average calculations, this program can be used
%% for type-reduction calculations in Chapter 9.
function[out,mem] = weighted_avg(Zout,Zmem,Wout,Wmem,tnorm,step)
[M,Nz] = size(Zout) ;
Nw = size(Wout,2) ;
Zout1 = Zout ;
Zmem1 = Zmem ;
Wout1 = Wout ;
Wmem1 = Wmem ;
if tnorm < 0,
tnorm_op = 'min' ;
else
tnorm_op = 'prod' ;
end %%% if tnorm
%% Points for the output
minz = min(min(Zout)) ;
maxz = max(max(Zout)) ;
out = [minz : step : maxz] ;
mem = zeros(size(out)) ;
count_Z = Nz^(M-1) ;
count_W = Nw^(M-1) ;
temp1 = zeros(1,Nz) ;
temp2 = temp1 ;
temp3 = zeros(1,Nw) ;
temp4 = temp3 ;
temp5 = temp3 ;
for i6 = 1 : count_Z,
for i4 = 1 : Nz,
Zpoint = Zout1(:,i4)*ones(1,Nw) ;
memZ = feval(tnorm_op,Zmem1(:,i4)) ;
num_mat = Zpoint .* Wout1 ;
for i1 = 1 : count_W,
answer = sum(num_mat)./sum(Wout1) ;
ind = round((answer - out(1))/step) + 1 ;
memW = feval(tnorm_op,Wmem1) ;
if tnorm < 0,,
memout = min(memW,memZ) ;
else
memout = memZ.*memW ;
end %% if minflag
for i3 = 1 : Nw,
mem(ind(i3)) = max(mem(ind(i3)) , memout(i3)) ; % max t-conorm
end %% if i3
temp5(1:Nw-1) = num_mat(M,2:Nw) ;
temp5(Nw) = num_mat(M,1) ;
num_mat(M,:) = temp5 ;
temp3(1:Nw-1) = Wout1(M,2:Nw) ;
temp3(Nw) = Wout1(M,1) ;
Wout1(M,:) = temp3 ;
temp4(1:Nw-1) = Wmem1(M,2:Nw) ;
temp4(Nw) = Wmem1(M,1) ;
Wmem1(M,:) = temp4 ;
for i2 = 1 : M-1,
if mod(i1,Nw^i2) == 0,
temp5(1:Nw-1) = num_mat(M-i2,2:Nw) ;
temp5(Nw) = num_mat(M-i2,1) ;
num_mat(M-i2,:) = temp5 ;
temp3(1:Nw-1) = Wout1(M-i2,2:Nw) ;
temp3(Nw) = Wout1(M-i2,1) ;
Wout1(M-i2,:) = temp3 ;
temp4(1:Nw-1) = Wmem1(M-i2,2:Nw) ;
temp4(Nw) = Wmem1(M-i2,1) ;
Wmem1(M-i2,:) = temp4 ;
end %% if mod(i1,Nw^i2)
end %%% for i2
end %%% for i1
end %% for i4
temp1(1:Nz-1) = Zout1(M,2:Nz) ;
temp1(Nz) = Zout1(M,1) ;
Zout1(M,:) = temp1 ;
temp2(1:Nz-1) = Zmem1(M,2:Nz) ;
temp2(Nz) = Zmem1(M,1) ;
Zmem1(M,:) = temp2 ;
for i5 = 1 : M-1,
if mod(i6,Nz^i5) == 0,
temp1(1:Nz-1) = Zout1(M-i5,2:Nz) ;
temp1(Nz) = Zout1(M-i5,1) ;
Zout1(M-i5,:) = temp1 ;
temp2(1:Nz-1) = Zmem1(M-i5,2:Nz) ;
temp2(Nz) = Zmem1(M-i5,1) ;
Zmem1(M-i5,:) = temp2 ;
end %% if mod(i1,N^i5)
end %%% for i5
end %%% for i6
return ;
|
|