优化方法上机大作业
学 院: 姓 名: 学 号:
指导老师: 肖 现 涛
第一题
源程序如下:
function zy_x = di1ti(x)
%di1ti是用来求解优化作业第一题的函数。 x0=x; yimuxulong=0.000001; g0=g(x0);s0=-g0; A=2*ones(100,100); k=0; while k<100
lanmed=-(g0)'*s0/(s0'*A*s0); x=x0+lanmed*s0;
1
g=g(x); k=k+1;
if norm(g) break; end miu=norm(g)^2/norm(g0)^2; s=-g+miu*s0; g0=g; s0=s;x0=x; end function f=f(x) f=(x'*ones(100,1))^2-x'*ones(100,1); function g=g(x) g=(2*x'*ones(100,1))*ones(100,1)-ones(100,1); 代入x0,运行结果如下: >> x=zeros(100,1); >> di1ti(x) After 1 iterations,obtain the optimal solution. 2 The optimal solution is -0.250000. The optimal \"x\" is \"ans\". ans =0.005*ones(100,1). 第二题 1. 最速下降法。 源程序如下: function zy_x=di2titidu(x) %该函数用来解大作业第二题。 x0=x; yimuxulong=1e-5; k=0; g0=g(x0); s0=-g0; while k>=0 if norm(g0) break; else lanmed=10;c=0.1;i=0; while i>=0&i<100 x=x0+lanmed*s0; if f(x)>(f(x0)+c*lanmed*g0'*s0) lanmed=lanmed/2; i=i+1; else break; end end x=x0+lanmed*s0; x0=x; g0=g(x); s0=-g0; k=k+1; end end zy_x=x; 4 zyj=f(x); fprintf('after %d iterations,obtain the optimal solution.\\n\\nThe optimal solution is %f.\\n\\n The optimal \"x\" is \"ans\".\\n',k,zyj); function f=f(x) x1=[1 0 0 0]*x; x2=[0 1 0 0]*x; x3=[0 0 1 0]*x; x4=[0 0 0 1]*x; f=(x1-1)^2+(x3-1)^2+100*(x2-x1^2)^2+100*(x4-x3^2)^2; function g=g(x) x1=[1 0 0 0]*x; x2=[0 1 0 0]*x; x3=[0 0 1 0]*x; x4=[0 0 0 1]*x; g=[2*(x1-1)-400*x1*(x2-x1^2);200*(x2-x1^2);2*(x3-1)-400*x3*(x4-x3^2);200*(x4-x3^2)]; >> x=[-1.2 1 -1.2 1]'; 5 >> di2titidu(x) after 5945 iterations,obtain the optimal solution. The optimal solution is 0.000000. The optimal \"x\" is \"ans\". ans = 1.0000 1.0000 1.0000 1.0000 2. 牛顿法 源程序如下: function zy_x=di2tinewton(x) %该函数用来解大作业第二题。 x0=x; yimuxulong=1e-5; k=0; g0=g(x0); h0=h(x0);s0=-inv(h0)*g0; while k>=0&k<1000 if norm(g0) x0=x; g0=g(x); h0=h(x); s0=-inv(h0)*g0; k=k+1; end end zy_x=x; zyj=f(x); fprintf('after %d iterations,obtain the optimal solution.\\n\\nThe optimal solution is %f.\\n\\n The optimal \"x\" is \"ans\".\\n',k,zyj); function f=f(x) x1=[1 0 0 0]*x; x2=[0 1 0 0]*x; x3=[0 0 1 0]*x; x4=[0 0 0 1]*x; f=(x1-1)^2+(x3-1)^2+100*(x2-x1^2)^2+100*(x4-x3^2)^2; 7 function g=g(x) x1=[1 0 0 0]*x; x2=[0 1 0 0]*x; x3=[0 0 1 0]*x; x4=[0 0 0 1]*x; g=[2*(x1-1)-400*x1*(x2-x1^2);200*(x2-x1^2);2*(x3-1)-400*x3*(x4-x3^2);200*(x4-x3^2)]; function h=h(x) x1=[1 0 0 0]*x; x2=[0 1 0 0]*x; x3=[0 0 1 0]*x; x4=[0 0 0 1]*x; h=[2+1200*x1^2-400*x2 -400*x1 0 0;-400*x1 200 0 0;0 0 2+1200*x3^2-400*x4 -400*x3;0 0 -400*x3 200]; 代入初始值,运行结果如下: >> x=[-1.2 1 -1.2 1]'; >> di2tinewton(x) after 6 iterations,obtain the optimal solution. The optimal solution is 0.000000. The optimal \"x\" is \"ans\". 8 ans = 1.0000 1.0000 1.0000 1.0000 可以看出,用Newton法经过6次迭代就能求出最优解。 3. BFGS法 源程序如下: function zy_x=di2tiBFGS(x) %该函数用来解大作业第二题。 x0=x; yimuxulong=1e-5; k=0; g0=g(x0); H0=eye(4);s0=-H0*g0; while k>=0&k<100 if norm(g0) if f(x)>(f(x0)+c*lanmed*g0'*s0) 9 lanmed=lanmed/2; i=i+1; else break; end end x=x0+lanmed*s0; dete_x=x-x0; dete_g=g(x)-g0; miu=1+dete_g'*H0*dete_g/(dete_x'*dete_g); H=H0+(miu*dete_x*dete_x'-H0*dete_g*dete_x'-dete_x*dete_g'*H0)/(dete_x'*dete_g); s=-H*g(x); x0=x; s0=s; H0=H; g0=g(x); k=k+1; end end 10 zy_x=x; zyj=f(x); fprintf('after %d iterations,obtain the optimal solution.\\n\\nThe optimal solution is %f.\\n\\n The optimal \"x\" is \"ans\".\\n',k,zyj); function f=f(x) x1=[1 0 0 0]*x; x2=[0 1 0 0]*x; x3=[0 0 1 0]*x; x4=[0 0 0 1]*x; f=(x1-1)^2+(x3-1)^2+100*(x2-x1^2)^2+100*(x4-x3^2)^2; function g=g(x) x1=[1 0 0 0]*x; x2=[0 1 0 0]*x; x3=[0 0 1 0]*x; x4=[0 0 0 1]*x; g=[2*(x1-1)-400*x1*(x2-x1^2);200*(x2-x1^2);2*(x3-1)-400*x3*(x4-x3^2);200*(x4-x3^2)]; 11 代入初始值,计算结果如下: >> x=[-1.2 1 -1.2 1]'; >> di2tiBFGS(x) after 53 iterations,obtain the optimal solution. The optimal solution is 0.000000. The optimal \"x\" is \"ans\". ans = 1.0000 1.0000 1.0000 1.0000 第三题 1. 惩罚函数法 源程序如下: 12 function zy_x=di3ti(x) %该函数用来解大作业第三题。 x0=x; M=100; c=4; m=1; while m>0 g0=g(x0,M); yimuxulong=1e-5;k=0;s0=-inv(H(x0,M))*g0; while k>=0 if norm(g0) s0=-inv(H(x0,M))*g0; k=k+1; end end if max([abs(h(x)),g1(x),g2(x),g3(x)])<0.5 break; else M=M*c; 13 m=m+1; end end zy_x=x; zyj=f(x); fprintf('after %d iterations,obtain the optimal solution.\\n\\nThe optimal solution is %f.\\n\\n The optimal \"x\" is \"ans\".\\n',m,zyj); function F=F(x,M) x1=[1 0]*x; x2=[0 1]*x; F=4*x1-x2^2-12+M*(h^2+g1^2+g2^2+g3^2); function g=g(x,M) x1=[1 0]*x; x2=[0 1]*x; g=[4+M*(-4*(25-x1^2-x2^2)*x1+2*(10*x1-x1^2+10*x2-x2^2-34)*(10-2*x1)+2*x1);-2*x2+M*(-4*(25-x1^2-x2^2)*x2+2*(10*x1-x1^2+10*x2-x2^2-34)*(10-2*x2)+2*x2)]; function H=H(x,M) x1=[1 0]*x; 14 x2=[0 1]*x; H=[M*(24*x1^2-120*x1+8*x2^2-40*x2+238),M*(16*x1*x2-40*x1-40*x2+200);M*(16*x1*x2-40*x1-40*x2+200),-2+M*(24*x2^2-120*x2+8*x1^2-40*x1+238)]; function f=f(x) x1=[1 0]*x; x2=[0 1]*x; f=4*x1-x2^2-12; function h=h(x) x1=[1 0]*x; x2=[0 1]*x; h=25-x1^2-x2^2; function g1=g1(x) x1=[1 0]*x; x2=[0 1]*x; g=10*x1-x1^2+10*x2-x2^2-34; if g<0 g1=g; else g1=0; end 5 1 function g2=g2(x) x1=[1 0]*x; x2=[0 1]*x; if x1>=0 g2=0; else g2=x1; end function g3=g3(x) x1=[1 0]*x; x2=[0 1]*x; if x2>=0 g3=0; else g3=x2; end 代入任意初始值,运算结果如下。 >> x=rand(2,1); >> di3ticf(x) after 1 iterations,obtain the optimal solution. 16 The optimal solution is -31.490552. The optimal \"x\" is \"ans\". ans = 1.0024 4.8477 2. 乘子法 源程序如下: function [x,mu,lambda,output]=multphr(fun,hf,gf,dfun,dhf,dgf,x0) %功能: 用乘子法解一般约束问题: min f(x), s.t. h(x)=0, g(x).=0 %输入: x0是初始点, fun, dfun分别是目标函数及其梯度; % hf, dhf分别是等式约束(向量)函数及其Jacobi矩阵的转置; % gf, dgf分别是不等式约束(向量)函数及其Jacobi矩阵的转置; %输出: x是近似最优点,mu, lambda分别是相应于等式约束和不等式约束的乘子向量; % output是结构变量, 输出近似极小值f, 迭代次数, 内迭代次数等 maxk=500; 17 c=2.0; eta=2.0;theta=0.8; k=0;ink=0; epsilon=0.00001; x=x0;he=feval(hf,x);gi=feval(gf,x); n=length(x);l=length(he);m=length(gi); mu=zeros(l,1);lambda=zeros(m,1); btak=10;btaold=10; while(btak>epsilon&&k ink=ink+ik; he=feval(hf,x);gi=feval(gf,x); 18 子问题 [x,ival,ik]=bfgs('mpsi','dmpsi',x0,fun,hf,gf,dfun,dhf,dgf,mu,lambda,c); btak=0; for i=1:l btak=btak+he(i)^2; end %更新乘子向量 for i=1:m temp=min(gi(i),lambda(i)/c); btak=btak+temp^2; end btak=sqrt(btak); if btak>epsilon if k>=2&&btak>theta*btaold c=eta*c; 9 1 end for i=1:l mu(i)=mu(i)-c*he(i); end for i=1:m lambda(i)=max(0,lambda(i)-c*gi(i)); end k=k+1; btaold=btak; x0=x; end end f=feval(fun,x); 20 output.fval=f; output.iter=k; %增广拉格朗日函数 function psi=mpsi(x,fun,hf,gf,dfun,dhf,dgf,mu,lambda,c) f=feval(fun,x);he=feval(hf,x);gi=feval(gf,x); l=length(he);m=length(gi); psi=f;s1=0; for i=1:l psi=psi-he(i)*mu(i); s1=s1+he(i)^2; end psi=psi+0.5*c*s1; s2=0; 21 for i=1:m s3=max(0,lambda(i)-c*gi(i)); s2=s2+s3^2-lambda(i)^2; end psi=psi+s2/(2*c); %不等式约束函数文件g1.m function gi=g1(x) gi=10*x(1)-x(1)^2+10*x(2)-x(2)^2-34; %目标函数的梯度文件df1.m function g=df1(x) g=[4, -2*x(2)]'; %等式约束(向量)函数的Jacobi矩阵(转置)文件dh1.m function dhe=dh1(x) 22 dhe=[-2*x(1), -2*x(2)]' %不等式约束(向量)函数的Jacobi矩阵(转置)文件dg1.m function dgi=dg1(x) dgi=[10-2*x(1), 10-2*x(2)]'; function [x,val,k]=bfgs(fun,gfun,x0,varargin) maxk=500; rho=0.55;sigma=0.4;epsilon=0.00001; k=0;n=length(x0); Bk=eye(n); while(k if(norm(gk) 23 end dk=-Bk\\gk; m=0;mk=0; while(m<20) newf=feval(fun,x0+rho^m*dk,varargin{:}); oldf=feval(fun,x0,varargin{:}); if(newf break; end m=m+1; end x=x0+rho^mk*dk; 24 sk=x-x0; yk=feval(gfun,x,varargin{:})-gk; if(yk'*sk>0) Bk=Bk-(Bk*sk*sk'*Bk)/(sk'*Bk*sk)+(yk*yk')/(yk'*sk); end k=k+1;x0=x; end val=feval(fun,x0,varargin{:}); 结果 x=[2 2]'; [x,mu,lambda,output]=multphr('fun','hf','gf1','df','dh','dg',x0) x = 1.0013 25 4.8987 mu = 0.7701 lambda = 0 0 0.9434 output = fval: -31.9923iter: 4 26 因篇幅问题不能全部显示,请点此查看更多更全内容