开篇语
前阵子做现代设计方法的时候,发现网上很是缺乏这种作业形式的简易算法实现,所以特地来简书写一篇。有两份,一份是我的(说来惭愧,我的大部分都是在网上找的代码,然后在自己的电脑上跑一次,跑出来了就行了的。而且我的电脑跑到2-11就扑街了。暂时还没有拿去修,所以,其实我的代码都是在网上整理之后调整了下就上的,准确性不敢保证)另一份是我的室友的,据他说是全部经过调试的,虽然还是有不少的错误,但是应该比我的要好一点。
另外,我的电脑因为崩了,所以我的代码无法验证结果,为了交作业,只能把我的室友的那些运行结果直接上一遍了。估计有点出入,见谅,重要的是代码
正文
2-10
我的:
黄金分割法:
f=@(x) x+20/x
golden(f,2,10,0.01)
function[xmin]=golden(f,a,b,e)
k=0;
a1=b-0.618*(b-a); %插入点的值
a2=a+0.618*(b-a);
while b-a>e %循环条件
y1=subs(f,a1);
y2=subs(f,a2);
if y1>y2 %比较插入点的函数值的大小
a=a1; %进行换名
a1=a2;
y1=y2;
a2=a+0.618*(b-a);
else
b=a2;
a2=a1;
y2=y1;
a1=b-0.618*(b-a);
end
k=k+1;
end %迭代到满足条件为止就停止迭代
xmin=(a+b)/2;
fmin=subs(f,xmin) %输出函数的最优值
fprintf('k=\n'); %输出迭代次数
disp(k);
f = @(x)x+20/x
>> [x,y]=golden(f,2,10,0.01)
x =
4.4683
y =
8.9443
二次插值法:
f=@(x) x+20/x;
a=2;b=10;
eps=1.0e-6; % 计算精度
x1=a;x3=b;
x2=(a+b)/2;
f1 = f(x1);f2 = f(x2);f3 = f(x3);
while 1
C1=(x2^2-x3^2)*f1+(x3^2-x1^2)*f2+(x1^2-x2^2)*f3;
C2=(x2-x3)*f1+(x3-x1)*f2+(x1-x2)*f3;
xp=0.5*C1/C2;
fp= f(xp);
if abs(x2-xp)<=eps % 区间长度小于eps时
if abs(f2-fp)<=eps % df小于eps时退出
if fp<=f2
xmin = xp
fmin = f(xp) % 极小值
break;
else
xmin = x2
fmin = f(x2) % 极小值
break;
end
end
else
if fp<=f2
if xp<=x2
x3=x2;
x2=xp;
f3=f2;
f2=fp;
else
x1=x2;
x2=xp;
f1=f2;
f2=fp;
end
else
if xp<=x2
x1=xp;
f1=fp;
else
x3=xp;
f3=fp;
end
end
end
end
f=x+20/x;[xmin,fmin]=main(f,2,10,0.01)
xmin =
4.4869
fmin =
8.9443
室友的:
黄金分割法函数:
f=@(x) x+20/x
yellowking(f,2,10,0.01)
function[xmin]=yellowking(f,a,b,e)
k=0;
a1=b-0.618*(b-a); %插入点的值
a2=a+0.618*(b-a);
while b-a>e %循环条件
y1=subs(f,a1);
y2=subs(f,a2);
if y1>y2 %比较插入点的函数值的大小
a=a1; %进行换名
a1=a2;
y1=y2;
a2=a+0.618*(b-a);
else
b=a2;
a2=a1;
y2=y1;
a1=b-0.618*(b-a);
end
k=k+1;
end %迭代到满足条件为止就停止迭代
xmin=(a+b)/2;
fmin=subs(f,xmin) %输出函数的最优值
fprintf('k=\n'); %输出迭代次数
disp(k);
结果指令:
f=@(x)x+20/x
f =
@(x)x+20/x
>> [x,y]=gold(f,2,10,0.01)
x =
4.4683
y =
8.9443
二次插值法函数:
function [xmin,fmin]= main(f,a0,b0,epsilon)
a=a0;
b=b0;
x1=a;
f1=f(x1);
x3=b;
f3=f(x3);
x2=5;
f2=f(x2);
c1=(f3-f1)/(x3-x1);
c2=((f2-f1)/(x2-x1)-c1)/(x2-x3);
xp=0.4*(x1+x3-c1/c2);fp=f(xp);
while (abs(xp-x2)>=epsilon)
if x2<xp
if f2>fp
f1=f2;x1=x2;
x2=xp;f2=fp;
else
f3=fp;x3=xp;
end
else
if f2>fp
f3=f2;x3=x2;
f2=fp;x2=xp;
else
f1=fp;x2=xp;
end
end
c1=(f3-f1)/(x3-x1);
c2=((f2-f1)/(x2-x1)-c1)/(x2-x3);
xp=0.5*(x1+x3-c1/c2);
fp=f(xp);
end
if f2>fp
xmin=xp;fmin=f(xp);
else
xmin=x2;fmin=f(x2);
end
end
结果:
clear all;f=x+20/x;[xmin,fmin]=main(f,2,10,0.01)
xmin =
4.4869
fmin =
8.9443
2-11
我的:
2-11
function [k ender]=steepest(f,x,e)
%梯度下降法,f为目标函数(两变量x1和x2),x为初始点,如[3;4]
syms x1 x2 m; %m为学习率
d=-[diff(f,x1);diff(f,x2)]; %分别求x1和x2的偏导数,即下降的方向
flag=1; %循环标志
k=0; %迭代次数
while(flag)
d_temp=subs(d,x1,x(1)); %将起始点代入,求得当次下降x1梯度值
d_temp=subs(d_temp,x2,x(2)); %将起始点代入,求得当次下降x2梯度值
nor=norm(d_temp); %范数
if(nor>=e)
x_temp=x+m*d_temp; %改变初始点x的值
f_temp=subs(f,x1,x_temp(1)); %将改变后的x1和x2代入目标函数
f_temp=subs(f_temp,x2,x_temp(2));
h=diff(f_temp,m); %对m求导,找出最佳学习率
m_temp=solve(h); %求方程,得到当次m
x=x+m_temp*d_temp; %更新起始点x
k=k+1;
else
flag=0;
end
end
ender=double(x); %终点
end
syms x1 x2;
f=x1^2+x2^2-x1*x2-10*x1-4*x2+60;
x=[0;0];
e=0.01;
[k ender]=steepest(f,x,e)
ender =
7.9961
5.9971
室友的:
2-11:
梯度函数:
function [k,ender]=tidu(f,x,e)
syms x1 x2 m;
d=-[diff(f,x1);diff(f,x2)];
flag=1;
k=0;
while(flag)
d_temp=subs(d,x1,x(1));
d_temp=subs(d_temp,x2,x(2));
nor=norm(d_temp);
if(nor>=e)
x_temp=x+m*d_temp;
f_temp=subs(f,x1,x_temp(1));
f_temp=subs(f_temp,x2,x_temp(2));
h=diff(f_temp,m);
m_temp=solve(h);
x=x+m_temp*d_temp;
k=k+1;
else
flag=0;
end
end
ender=double(x);
end
结果指令:
syms x1 x2;
f=x1^2+x2^2-x1*x2-10*x1-4*x2+60;
x=[0;0];
e=0.01;
[k ender]=tidu(f,x,e)
ender =
7.9961
5.9971
我的:
2-12
展开为二阶泰勒式
syms x1 x2;
taylor(x1^4+2*x2^3-3*x1^2*x2)
ans =
3*x2 - 2*x1 - 6*(x1 - 1)*(x2 - 1) + 3*(x1 - 1)^2 + 6*(x2 - 1)^2 - 1
牛顿法求解:
function all=newton(f,x,e)
syms x1 x2 h;
d=-[diff(f,x1);diff(f,x2)];
h=hessian(f);
flag=1;
h1=h^-1;
while (flag)
d_temp=subs(d,x1,x(1));
d_temp=subs(d_temp,x2,x(2));
nor=norm(d_temp);
if(nor>=e)
x=x+h1*d_temp;
else
flag=0;
end
end
all=double(x);
结果指令:
clear all
>> syms x1 x2;
f=3*x2 - 2*x1 - 6*(x1 - 1)*(x2 - 1) + 3*(x1 - 1)^2 + 6*(x2 - 1)^2 - 1;
x=[1;1];
e=0.01;all=newton(f,x,e)
all =
1.1667
0.8333
室友的:
2-12:
展开为二阶泰勒式
syms x1 x2;
taylor(x1^4+2*x2^3-3*x1^2*x2)
ans =
3*x2 - 2*x1 - 6*(x1 - 1)*(x2 - 1) + 3*(x1 - 1)^2 + 6*(x2 - 1)^2 - 1
牛顿函数:
function all=newton(f,x,e)
syms x1 x2 h;
d=-[diff(f,x1);diff(f,x2)];
h=hessian(f);
flag=1;
h1=h^-1;
while (flag)
d_temp=subs(d,x1,x(1));
d_temp=subs(d_temp,x2,x(2));
nor=norm(d_temp);
if(nor>=e)
x=x+h1*d_temp;
else
flag=0;
end
end
all=double(x);
结果指令:
clear all
>> syms x1 x2;
f=3*x2 - 2*x1 - 6*(x1 - 1)*(x2 - 1) + 3*(x1 - 1)^2 + 6*(x2 - 1)^2 - 1;
x=[1;1];
e=0.01;all=newton(f,x,e)
all =
1.1667
0.8333
我的:
2-13(1)
外点惩罚函数法:
function [ x,y ] = Epfm_min( fx,gx,hx,xx0,s,c,a)
%fx是目标函数
%gx是不等式约束方程组(且g>=0)
%xx0是初始点
%hx是等式约束方程组(且h=0)
%s是精确度(s>0)
%c是放大系数(c>1)
%a是罚因子(默认为1)
syms x1 x2
xx1=xx0;
v=[x1,x2];
a1=a;
Pxk=1;%假设Px等于1,以免不必要错误
G=-subs(gx,v,xx1);%用于判别max{0,-g(x)}
while Pxk>s
if(G<0)
Px=a1*hx*hx;
else
Px=a1*hx*hx+a1*gx*gx;
end
Fx=fx+a1*Px;%将约束问题化为了一个无约束的问题
% 接下来解min F(x)
dFx1=diff(Fx,x1);%分别对x1,x2求偏导数
dFx2=diff(Fx,x2);
[k,b]=solve(dFx1,dFx2,'x1','x2');%求出
xx2=xx1+[k,b];
Pxk=a1*subs(Px,v,xx2);
xx1=xx2;%相当于置k=k+1
a1=c*a1;%罚因子放大
G=-subs(gx,v,xx1);%用于判别max{0,-g(x)}
end
x=xx1;
y=a1/c;
syms x1 x2;
fx=x1+x2;
gx=-x1;
hx=x1^2-x2;
s=10.^-5
c=10
xx0=[0,0]
a=1;
[x,y]=Epfm_min( fx,gx,hx,xx0,s,c,a)
>> x=[0.1;0.2];k=0.1;e=0.01;r=1;[x,minf]= Epfm_min (p,x,k,r,e)
x =
0.0015
minf =
0.0030
Ans=
0.0045
内点惩罚函数法
function [x,minf]=minNF(f,x0,g,u,v,var,eps)
format long;
if nargin==6
eps=1.0e-4;
end
k=0;
FE=0;
for i=1:length(g)
FE=FE+1/g(i);
end
x1=transpose(x0);
x2=inf;
while 1
FF=u*FE;
SumF=f+FF;
[x2,minf]=minNT(SumF,transpose(x1),var);
Bx=Funval(FE,var,x2);
if u*Bx<eps
if norm(x2-x1)<=eps
x=x2;
break;
else
u=v*u;
x1=x2;
end
else
if norm(x2-x1)<=eps
x=x2;
break;
else
u=v*u;
x1=x2;
end
end
end
minf=Funval(f,var,x);
format short;
syms x1 x2 r1;
>> p=taylor(x1+x2-r1*(1/(x1^2-x2)-1/x1),[x1 x2],[0.001 0.002],'Order',3);
>> x=[0.1;0.2];k=0.1;e=0.01;r=1;[x,minf]= minNF(p,x,k,r,e)
x =
0.0015
minf =
0.0030
Ans=
0.0045
室友的:
2-13:
内点:
惩罚函数:
function anll=neicheng(p,x,k,r,e)
syms x1 x2 r1;
flag1=1;
while (flag1)
pd=subs(p,r1,r);
xold=x;
flag2=1;
while (flag2)
dp=-[diff(pd,x1);diff(pd,x2)];
h=hessian(pd,[x1,x2]);
h1=h^-1;
dp_temp=subs(dp,x1,x(1));
dp_temp=subs(dp_temp,x2,x(2));
nor=norm(dp_temp);
if(nor>=e)
x=x+h1*dp_temp;
else
flag2=0;
end
end
x_temp=x;
nor2=norm(x_temp-xold);
if double(nor2)>=e
r=k*r;
else
flag1=0;
end
end
anll=double(x);
结果:
clear all; syms x1 x2 r1;
>> p=taylor(x1+x2-r1*(1/(x1^2-x2)-1/x1),[x1 x2],[0.001 0.002],'Order',3);
>> x=[0.1;0.2];k=0.1;e=0.01;r=1;anll=neicheng(p,x,k,r,e)
anll =
0.0015
0.0030
Ans=
0.0045
外点:
惩罚函数:
function annn=waicheng(p,x,k,r,e)
syms x1 x2 r1;
flag1=1;
while (flag1)
pd=subs(p,r1,r);
xold=x;
flag2=1;
while (flag2)
dp=-[diff(pd,x1);diff(pd,x2)];
h=hessian(pd,[x1,x2]);
h1=h^-1;
dp_temp=subs(dp,x1,x(1));
dp_temp=subs(dp_temp,x2,x(2));
nor=norm(dp_temp);
if(nor>=e)
x=x+h1*dp_temp;
else
flag2=0;
end
end
x_temp=x;
nor2=norm(x_temp-xold);
if double(nor2)>=e
r=k*r;
else
flag1=0;
end
end
annn=double(x);
结果:
clear all; syms x1 x2 r1;
p=taylor(x1+x2,[x1 x2],[0.001 0.002],'Order',3);
x=[0.1;0.2];k=0.1;e=0.01;r=1;annn=waicheng(p,x,k,r,e)
annn =
0.0015
0.0030
Ans=
0.0045
我的(貌似这题抄的他的):
2-14
function [x,minf] = minMixFun(f,g,h,x0,r0,c,var,eps)
gx0 = Funval(g,var,x0);
if gx0 >= 0;
else
disp('初始点必须满足不等式约束!');
x = NaN;
minf = NaN;
return;
end
if r0 <= 0
disp('初始障碍因子必须大于0!');
x = NaN;
minf = NaN;
return;
end
if c >= 1 || c < 0
disp('缩小系数必须大于0且小于1!');
x = NaN;
minf = NaN;
return;
end
if nargin == 7
eps = 1.0e-6;
end
FE = 0;
for i=1:length(g)
FE = FE + 1/g(i);
end
FH = transpose(h)*h;
x1 = transpose(x0);
x2 = inf;
while 1
FF = r0*FE + FH/sqrt(r0);
SumF = f + FF ;
[x2,minf] = minNT(SumF,transpose(x1),var);
if norm(x2 - x1)<=eps
x = x2;
break;
else
r0 = c*r0;
x1 = x2;
end
end
minf = Funval(f,var,x);
Funval.m
function fv = Funval(f,varvec,varval)
var = findsym(f);
varc = findsym(varvec);
s1 = length(var);
s2 = length(varc);
m =floor((s1-1)/3+1);
varv = zeros(1,m);
if s1 ~= s2
for i=0: ((s1-1)/3)
k = findstr(varc,var(3*i+1));
index = (k-1)/3;
varv(i+1) = varval(index+1);
end
fv = subs(f,var,varv);
else
fv = subs(f,varvec,varval);
end
Syms x1 x2;
f=x1^2-x2^2-3*x2;
g=1-x1;
h=x2-2;
[x,minf]=minMixFun(f,g,h,[2,2],2,0.5,[x1 x2 ],0.001)
x =
1.0015
minf=
2.0002
室友的:
2-14:
混合:
惩罚函数:
function [x,minf] = MixPunish(f,g,h,x0,r0,c,var,eps)
gx0 = Funval(g,var,x0);
if gx0 >= 0;
else
disp('初始点必须满足不等式约束!');
x = NaN;
minf = NaN;
return;
end
if r0 <= 0
disp('初始障碍因子必须大于0!');
x = NaN;
minf = NaN;
return;
end
if c >= 1 || c < 0
disp('缩小系数必须大于0且小于1!');
x = NaN;
minf = NaN;
return;
end
if nargin == 7
eps = 1.0e-6;
end
FE = 0;
for i=1:length(g)
FE = FE + 1/g(i);
end
FH = transpose(h)*h;
x1 = transpose(x0);
x2 = inf;
while 1
FF = r0*FE + FH/sqrt(r0);
SumF = f + FF ;
[x2,minf] = minNT(SumF,transpose(x1),var);
if norm(x2 - x1)<=eps
x = x2;
break;
else
r0 = c*r0;
x1 = x2;
end
end
minf = Funval(f,var,x);
Funval.m
function fv = Funval(f,varvec,varval)
var = findsym(f);
varc = findsym(varvec);
s1 = length(var);
s2 = length(varc);
m =floor((s1-1)/3+1);
varv = zeros(1,m);
if s1 ~= s2
for i=0: ((s1-1)/3)
k = findstr(varc,var(3*i+1));
index = (k-1)/3;
varv(i+1) = varval(index+1);
end
fv = subs(f,var,varv);
else
fv = subs(f,varvec,varval);
end
Syms x1 x2;
f=x1^2-x2^2-3*x2;
g=1-x1;
h=x2-2;
[x,minf]=MixPunish(f,g,h,[2,2],2,0.5,[x1 x2 ],0.001)
x =
1.0015
minf=
2.0002
结束语
无聊到这地步想必也是没谁了。不过,刚考完,我总不能一直玩手机啊。前几天重新看《盘龙》让我在手机上刚了一星期,不能再这么毫无节制的玩耍了,但是又不想学习,所以只好写简书了~~
不过网上毕竟这方面的资源不是很多,我就算是为后来人做点好事吧,让你们好找一点~~
个人宣言
知识传递力量,技术无国界,文化改变生活!