大连理工大学优化方法上机大作业文档格式.docx
《大连理工大学优化方法上机大作业文档格式.docx》由会员分享,可在线阅读,更多相关《大连理工大学优化方法上机大作业文档格式.docx(27页珍藏版)》请在冰点文库上搜索。
f1=fun(x0+ak*dk);
slope=dot(gk,dk);
whilef1>
f0+0.1*ak*slope
ak=ak/4;
xk=x0+ak*dk;
f1=fun(xk);
k=k+1;
x0=xk;
gk=grad(xk);
res=norm(gk);
fprintf('
--The%d-thiter,theresidualis%f\n'
k,res);
x_star=xk;
end
>
clear
x0=[0,0]'
;
eps=1e-4;
x=steepest(x0,eps)
2.牛顿法:
functiong=grad2(x)
g=zeros(2,2);
g(1,1)=2+400*(3*x
(1)^2-x
(2));
g(1,2)=-400*x
(1);
g(2,1)=-400*x
(1);
g(2,2)=200;
functionx_star=newton(x0,eps)
bk=[grad2(x0)]^(-1);
dk=-bk*gk;
xk=x0+dk;
gk=grad(xk);
bk=[grad2(xk)]^(-1);
x1=newton(x0,eps)
--The1-thiter,theresidualis447.213595
--The2-thiter,theresidualis0.000000
x1=
1.0000
3.BFGS法:
functionx_star=bfgs(x0,eps)
g0=grad(x0);
gk=g0;
Hk=eye
(2);
=1000
dk=-Hk*gk;
fa0=xk-x0;
go=gk;
y0=gk-g0;
Hk=((eye
(2)-fa0*(y0)'
)/((fa0)'
*(y0)))*((eye
(2)-(y0)*(fa0)'
*(y0)))+(fa0*(fa0)'
*(y0));
End
x=bfgs(x0,eps)
4.共轭梯度法:
g=zeros(2,1);
functionx_star=CG(x0,eps)
gk=grad(x0);
g0=gk;
p=(gk/g0)^2;
dk1=dk;
dk=-gk+p*dk1;
x=CG(x0,eps)
上机大作业2:
functionf=obj(x)
f=4*x
(1)-x
(2)^2-12;
function[h,g]=constrains(x)
h=x
(1)^2+x
(2)^2-25;
g=zeros(3,1);
g
(1)=-10*x
(1)+x
(1)^2-10*x
(2)+x
(2)^2+34;
g
(2)=-x
(1);
g(3)=-x
(2);
functionf=alobj(x)%拉格朗日增广函数
%N_equ等式约束个数?
%N_inequ不等式约束个数
N_equ=1;
N_inequ=3;
globalr_alpena;
%全局变量
h_equ=0;
h_inequ=0;
[h,g]=constrains(x);
%等式约束部分?
fori=1:
N_equ
h_equ=h_equ+h(i)*r_al(i)+(pena/2)*h(i).^2;
end
%不等式约束部分
N_inequ
h_inequ=h_inequ+(0.5/pena)*(max(0,(r_al(i)+pena*g(i))).^2-r_al(i).^2);
%拉格朗日增广函数值
f=obj(x)+h_equ+h_inequ;
functionf=compare(x)
globalr_alpenaN_equN_inequ;
h_inequ=zeros(3,1);
%等式部分
1
h_equ=abs(h(i));
%不等式部分
3
h_inequ=abs(max(g(i),-r_al(i+1)/pena));
h1=max(h_inequ);
f=max(abs(h_equ),h1);
%sqrt(h_equ+h_inequ);
function[x,fmin,k]=almain(x_al)
%本程序为拉格朗日乘子算法示例算法%函数输入:
%x_al:
初始迭代点
%r_al:
初始拉格朗日乘子N-equ:
等式约束个数N_inequ:
不等式约束个数?
%函数输出
%X:
最优函数点FVAL:
最优函数值
%============================程序开始================================
globalr_alpena;
%参数(全局变量)
pena=10;
%惩罚系数
r_al=[1,1,1,1];
c_scale=2;
%乘法系数乘数
cta=0.5;
%下降标准系数
e_al=1e-4;
%误差控制范围
max_itera=25;
out_itera=1;
%迭代次数
%===========================算法迭代开始=============================
whileout_itera<
max_itera
x_al0=x_al;
r_al0=r_al;
%判断函数?
compareFlag=compare(x_al0);
%无约束的拟牛顿法BFGS
[X,fmin]=fminunc(@alobj,x_al0);
x_al=X;
%得到新迭代点
%判断停止条件?
ifcompare(x_al)<
e_al
disp('
wegettheoptpoint'
);
break
%c判断函数下降度?
cta*compareFlag
pena=1*pena;
%可以根据需要修改惩罚系数变量
else
pena=min(1000,c_scale*pena);
%%乘法系数最大1000
pena=2*pena'
%%?
更新拉格朗日乘子
[h,g]=constrains(x_al);
fori=1:
1
%%等式约束部分
r_al(i)=r_al0(i)+pena*h(i);
%%不等式约束部分
r_al(i+1)=max(0,(r_al0(i+1)+pena*g(i)));
out_itera=out_itera+1;
%+++++++++++++++++++++++++++迭代结束+++++++++++++++++++++++++++++++++
disp('
theiterationnumber'
k=out_itera;
thevalueofconstrains'
compare(x_al)
theoptpoint'
x=x_al;
fmin=obj(X);
x_al=[0,0];
[x,fmin,k]=almain(x_al)
上机大作业3:
1、
clearall
n=3;
c=[-3,-1,-3]'
A=[2,1,1;
1,2,3;
2,2,1;
-1,0,0;
0,-1,0;
0,0,-1];
b=[2,5,6,0,0,0]'
cvx_begin
variablex(n)
minimize(c'
*x)
subjectto
A*x<
=b
cvx_end
CallingSDPT34.0:
6variables,3equalityconstraints
------------------------------------------------------------
num.ofconstraints=3
dim.oflinearvar=6
*******************************************************************
SDPT3:
Infeasiblepath-followingalgorithms
versionpredcorrgamexponscale_data
NT10.00010
itpstepdsteppinfeasdinfeasgapprim-objdual-objcputime
-------------------------------------------------------------------
0|0.000|0.000|1.1e+01|5.1e+00|6.0e+02|-7.000000e+010.000000e+00|0:
0:
00|chol11
1|0.912|1.000|9.4e-01|4.6e-02|6.5e+01|-5.606627e+00-2.967567e+01|0:
01|chol11
2|1.000|1.000|1.3e-07|4.6e-03|8.5e+00|-2.723981e+00-1.113509e+01|0:
3|1.000|0.961|2.3e-08|6.2e-04|1.8e+00|-4.348354e+00-6.122853e+00|0:
4|0.881|1.000|2.2e-08|4.6e-05|3.7e-01|-5.255152e+00-5.622375e+00|0:
5|0.995|0.962|1.6e-09|6.2e-06|1.5e-02|-5.394782e+00-5.409213e+00|0:
6|0.989|0.989|2.7e-10|5.2e-07|1.7e-04|-5.399940e+00-5.400100e+00|0:
7|0.989|0.989|5.3e-11|5.8e-09|1.8e-06|-5.399999e+00-5.400001e+00|0:
8|1.000|0.994|2.8e-13|4.3e-11|2.7e-08|-5.400000e+00-5.400000e+00|0:
01|
stop:
max(relativegap,infeasibilities)<
1.49e-08
numberofiterations=8
primalobjectivevalue=-5.39999999e+00
dualobjectivevalue=-5.40000002e+00
gap:
=trace(XZ)=2.66e-08
relativegap=2.26e-09
actualrelativegap=2.21e-09
rel.primalinfeas(scaledproblem)=2.77e-13
rel.dual"
"
=4.31e-11
rel.primalinfeas(unscaledproblem)=0.00e+00
=0.00e+00
norm(X),norm(y),norm(Z)=4.3e+00,1.3e+00,1.9e+00
norm(A),norm(b),norm(C)=6.7e+00,9.1e+00,5.4e+00
TotalCPUtime(secs)=0.71
CPUtimeperiteration=0.09
terminationcode=0
DIMACS:
3.6e-130.0e+005.8e-110.0e+002.2e-092.3e-09
Status:
Solved
Optimalvalue(cvx_optval):
-5.4
2、
n=2;
c=[-2,-4]'
G=[0.5,0;
0,1];
A=[1,1;
-1,0;
0,-1];
b=[1,0,0]'
minimize(x'
*G*x+c'
7variables,3equalityconstraints
Forimprovedefficiency,SDPT3issolvingthedualproblem.
dim.ofsocpvar=4,num.ofsocpblk=1
dim.oflinearvar=3
0|0.000|0.000|8.0e-01|6.5e+00|3.1e+02|1.000000e+010.000000e+00|0:
1|1.000|0.987|4.3e-07|1.5e-01|1.6e+01|9.043148e+00-2.714056e-01|0:
2|1.000|1.000|2.6e-07|7.6e-03|1.4e+00|1.234938e+00-5.011630e-02|0:
3|1.000|1.000|2.4e-07|7.6e-04|3.0e-01|4.166959e-011.181563e-01|0:
4|0.892|0.877|6.4e-08|1.6e-04|5.2e-02|2.773022e-012.265122e-01|0:
5|1.000|1.000|1.0e-08|7.6e-06|1.5e-02|2.579468e-012.427203e-01|0:
6|0.905|0.904|3.1e-09|1.4e-06|2.3e-03|2.511936e-012.488619e-01|0:
7|1.000|1.000|6.1e-09|7.7e-08|6.6e-04|2.503336e-012.496718e-01|0:
8|0.903|0.903|1.8e-09|1.5e-08|1.0e-04|2.500507e-012.499497e-01|0:
9|1.000|1.000|4.9e-10|3.5e-10|2.9e-05|2.500143e-012.499857e-01|0:
10|0.904|0.904|4.7e-11|1.3e-10|4.4e-06|2.500022e-012.499978e-01|0:
00|chol22
11|1.000|1.000|2.3e-12|9.4e-12|1.2e-06|2.500006e-012.499994e-01|0:
12|1.000|1.000|4.7e-13|1.0e-12|1.8e-07|2.500001e-012.499999e-01|0:
13|1.000|1.000|2.0e-12|1.0e-12|4.2e-08|2.500000e-012.500000e-01|0:
14|1.000|1.000|2.6e-12|1.0e-12|7.3e-09|2.500000e-012.500000e-01|0:
00|
-------