custom simmulated annealing program doesn't produce desired result
1 visualización (últimos 30 días)
Mostrar comentarios más antiguos
I need to perform global optimization using simulated annealing algorithm. To do that I prepared a custom simulated annealing program which is basically optimized a distance function by varying few set of parameters. The program runs without any problem but it doesn't optimize the parameters. The code is attached in the following section.
if true
atype=a{1};
n=size(atype,1);
x= a{2};
y= a{3};
z= a{4};
qm_atom_charge=aa{3};
fclose(fileID1);
ah0=aaa{1};
bh0=aaa{2};
en0=aaa{3};
na=size(ah0,1);
nb=size(bh0,1);
bound=3.0;
vstep=1.0;
bound1(1:na)=bound;
bound2(1:nb)=bound;
ahu=ah0+bound1';
ahl=ah0-bound1';
bhu=bh0+bound2';
bhl=bh0-bound2';
enu=en0+bound1';
enl=en0-bound1';
Temp=100;
iter1=0;
acc=0.0;
rejc=0.0;
tmpstep=5;
mcstep=5;
maxit=tmpstep.*mcstep.*na.*na.*nb;
cost_function(1:maxit)=0.0;
ah=ah0;
bh=bh0;
en=en0;
ah1=ah0;
bh1=bh0;
en1=en0;
atom_charge=qeq_charge(n,x,y,z,atype,ah,bh,en);
distance=atom_charge-qm_atom_charge;
cost_current=norm(distance);
cost_tol=9999;
for itemp=1:tmpstep
Temp=Temp.*0.01;
for iter=1:mcstep
rv1=2.*bound.*rand(1:na)-bound;
rv2=2.*bound.*rand(1:nb)-bound;
rv3=2.*bound.*rand(1:na)-bound;
rv1=vstep.*rv1;
rv2=vstep.*rv2;
rv3=vstep.*rv3;
for m1=1:na
ah(m1)=ah1(m1)+rv1(m1);
if ah(m1)<=0.0
ah(m1)=ah(m1)-rv1(m1);
else
ind1=(ah(m1)>=ahl(m1)&ah(m1)<=ahu(m1));
if (ind1==0)
ah(m1)=ah(m1)-rv1(m1);
end
end
for m2=1:nb
bh(m2)=bh1(m2)+rv2(m2);
if bh(m2)<=0.0
bh(m2)=bh(m2)-rv2(m2);
else
ind2=(bh(m2)>=bhl(m2)&bh(m2)<=bhu(m2));
if (ind2==0)
bh(m2)=bh(m2)-rv2(m2);
end
end
for m3=1:na
en(m3)=en1(m3)+rv3(m3);
if en(m3)<=0.0
en(m3)=en(m3)-rv3(m3);
else
ind3=(en(m3)>=enl(m3)&en(m3)<=enu(m3));
if (ind3==0)
en(m3)=en(m3)-rv3(m3);
end
end
iter1=iter1+1;
chk(1)=isequal(ah,ah1);
chk(2)=isequal(bh,bh1);
chk(3)=isequal(en,en1);
if any(chk)==0
atom_charge=qeq_charge(n,x,y,z,atype,ah,bh,en);
distance=atom_charge-qm_atom_charge;
cost_old=cost_current;
cost_current=norm(distance);
if cost_current<=cost_old
ah1=ah;
bh1=bh;
en1=en;
acc=acc+1;
cost_function(acc)=cost_current;
if cost_current<=0.001
cost_tol=1;
break
else
end
else
p1=rand(1);
p=exp((cost_old-cost_current)/Temp);
if p1<p
acc=acc+1;
ah1=ah;
bh1=bh;
en1=en;
cost_function(acc)=cost_current;
else
rejc=rejc+1;
end
if cost_tol==1
break
end
end
if cost_tol==1
break
end
end
if cost_tol==1
break
end
end
if cost_tol==1
break
end
end
if cost_tol==1
break
end
end
if cost_tol==1
break
end
end
if cost_tol==1
break
end
end
end
Here, the function qeq_charge(n,x,y,z,atype,ah,bh,en) outputs an array, atom_charge and then I calculate the distance function. My intention is to minimize the distance function by varying ah,bh,en parameters. However, it doesn't work properly. I have been really puzzled in this issue since last week. Please help me out in this regard.
Thanks in advance sudipta
0 comentarios
Respuestas (0)
Ver también
Categorías
Más información sobre Simulated Annealing en Help Center y File Exchange.
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!