一个主要用于预测和分类的源程序.zip_动量因子_回归分析 matlab_误差建模 matlab_预测 matlab

%'因子数n1,预报对象数 n2,试报年数n,学习样本数 m,，mm为总样本数，学习因子a1，动量因子 a2,总体误差ee1 %个体误差ee2,迭代次数 num,隐层节点数 yj， close all clear all clc [file,path]=uigetfile('*.*','打开数据文件'); prompt={'包括Y的列数','样本总个数','建模样本数','预报因子数','预报对象数','学习因子','动量因子','训练次数','总体误差','隐节点数'}; def={'4','45','40','3','1','0.9','0.7','5','0.001','3'}; dlgTitle='参数输入'; lineNo=1; answer=inputdlg(prompt,dlgTitle,lineNo,def); answer=char(answer); h=str2num(answer(1,:)); mm=str2num(answer(2,:)); m=str2num(answer(3,:)); n1=str2num(answer(4,:)); n2=str2num(answer(5,:)); a1=str2num(answer(6,:)); a2=str2num(answer(7,:)); num=str2num(answer(8,:)); ee1=str2num(answer(9,:)); yj=str2num(answer(10,:)); filename1=strcat(path,file); fid1=fopen(filename1,'r'); a=fscanf(fid1,'%f',h*mm); fclose(fid1); a=reshape(a,h,mm); a=a'; p=a; [file,path]=uiputfile('*.*','保存'); filename2=strcat(path,file); fid2=fopen(filename2,'w'); n=mm-m; hdata=p; hda(3,(h))=0; fdata(mm,(h))=0; s1=min(hdata(1:m,:)); s2=max(hdata(1:m,:)); hda(1,1:h)=mean(hdata(1:m,1:(h))); mean(hdata(1:m,1:(h))); hda(2,1:h)=(9*s2-s1)./8; hda(3,1:h)=(9*s1-s2)./8; for i=1:mm for j=1:(n1+n2) if hdata(i,j)~=-999 fdata(i,j)=(hdata(i,j)-hda(3,j))/(hda(2,j)-hda(3,j)); else fdata(i,j)=(hda(1,j)-hda(3,j))/(hda(2,j)-hda(3,j)); end end end %-----------------------------------------fdata 中放的是标准化的资料，gdata 中放的是预报量资料 gdata=fdata(1:mm,(n1+1):h) fdata=fdata(:,1:n1) ggdata=p(1:mm,(n1+1):h) %---------------------------------------------------------- v=randn(n1,yj); vc=randn(yj); w=randn(yj,n2); rc=randn(n2); f=fdata; g=gdata; m_data(m+n,n2)=0; n_dd=0; ee=0.1; eee=0 %计算 n_dd=1 bar=waitbar(0,'please wait....'); for n_dd=1:num waitbar(n_dd/num,bar); %给连接权付初值0 v1=zeros(n1,yj); vc1=zeros(yj); w1=zeros(yj,n2); rc1=zeros(n2); for i=1:m % '计算隐含层的激活值 for j=1:yj xx=0 ; for k=1:n1 xx=xx+f(i,k)*v(k,j); end xx=xx+vc(j); b(j)=1/(1+exp(-xx)); end % 计算输出层单元的激活值 for j=1:n2 xx=0; for k=1:yj xx=xx+b(k)*w(k,j); end xx=xx+rc(j); c(j)=1/(1+exp(-xx)); mdata(i,j)=c(j); end % 计算输出层单元的一般化误差 for j=1:n2 d(j)=c(j)*(1-c(j))*(g(i,j)-c(j)); end % 计算隐含层对于每个dj误差 for j=1:yj xx=0; for k=1:n2 xx=xx+w(j,k)*d(k); end e(j)=b(j)*(1-b(j))*xx; end %调整隐含层单元 到输出层单元的连接权 for j=1:yj for k=1:n2 w1(j,k)=w1(j,k)+a1*b(j)*d(k); end end %调整 输出层单元的阈值 for j=1:n2 rc1(j)=rc1(j)+a1*d(j); end % 调整输入层单元 到隐含层单元的连接权 for j=1:n1 for k=1:yj v1(j,k)=v1(j,k)+a2*f(i,j)*e(k); end end % 计算隐含层单元的阈值 for j=1:yj vc1(j)=vc1(j)+a2*e(j); end end %------------------------------------- labuda=0.15; %调整连接权 for j=1:n1 for k=1:yj v(j,k)=v(j,k)+labuda*v1(j,k); end end for j=1:yj vc(j)=vc(j)+labuda*vc1(j); end for j=1:yj for k=1:n2 w(j,k)=w(j,k)+labuda*w1(j,k); end end for j=1:n2 rc(j)=rc(j)+labuda*rc1(j); end %计算总体误差 ee=0; for i=1:m for j=1:n2 ee=ee+(mdata(i,j)-g(i,j))^2/2; end end ee=ee/m; eee(n_dd)=ee; if ee<ee1 break end end close(bar); ee; plot(eee); fprintf(fid2,'包括预报量的列数：%1d,样本总数：%2d,建模数：%1d,预报因子数：%1d,预报量数：%1d,隐节点数：%1d\n',h,mm,m,n1,n2,yj); fprintf(fid2,'训练次数：%d:,动量因子：%4.1f,学习因子:%4.1f,总体误差：%f\n',num,a1,a2,ee1); fprintf(fid2,'收敛误差为：%8.6f\n\n',ee) ; sprintf('收敛误差为：%12.6f',ee) n_dd; %预报 for i=1:mm %计算隐含层新的激活值 for j=1:yj xx=0; for k=1:n1 xx=xx+f(i,k)*v(k,j); end xx=xx+vc(j); b(j)=1/(1+exp(-xx)); end % for j=1:n2 xx=0; for k=1:yj xx=xx+b(k)*w(k,j); end xx=xx+rc(j); c(j)=1/(1+exp(-xx)); mdata(i,j)=c(j); end end mdata; % transfering the original value for i=1:mm for j=1:n2 if mdata(i,j)~=-999 mdata(i,j)=mdata(i,j)*(hda(2,h)-hda(3,h))+hda(3,h) ; elseif mdata(i,j)==-999 mdata(i,j)=(hda(1,j)-hda(3,h))/(hda(2,h)-hda(3,h)); end end end mdata fprintf(fid2,'predictand fitting\n\n'); fprintf(fid2,' 实况 预测 误差 相对误差\n'); for i=1:m for j=1:n2 xd(i,j)=abs(ggdata(i,j)-mdata(i,j))/ggdata(i,j)*100; fprintf(fid2,' %12.4f%12.4f%12.4f%12.4f%%',ggdata(i,j),mdata(i,j),(ggdata(i,j)-mdata(i,j)),abs(ggdata(i,j)-mdata(i,j))/ggdata(i,j)*100); end fprintf(fid2,'\n'); end jd1=mean(abs(ggdata(1:m,:)-mdata(1:m,:))); fprintf(fid2,'average %12.4f%12.4f%12.4f%12.4f%%\n\n',mean(ggdata(1:m,:)),mean(mdata(1:m,:)),jd1,mean(xd(1:m,:))); fprintf(fid2,'\npredictand prediction\n\n'); for i=m+1:mm for j=1:n2 xd(i,j)=abs(ggdata(i,j)-mdata(i,j))/ggdata(i,j)*100; fprintf(fid2,' %12.4f%12.4f%12.4f%12.4f%%',ggdata(i,j),mdata(i,j),(ggdata(i,j)-mdata(i,j)),abs(ggdata(i,j)-mdata(i,j))/ggdata(i,j)*100); end fprintf(fid2,'\n'); end fprintf(fid2,'\n\n'); jd2=mean(abs(ggdata((m+1):mm,:)-mdata((m+1):mm,:))); fprintf(fid2,'\n\n average%12.4f%12.4f%12.4f%12.4f%%\n\n',mean(ggdata(1+m:mm,:)),mean(mdata(1+m:mm,:)),jd2,mean(xd(1+m:mm,1))); fprintf(fid2,'\n'); fprintf(fid2,'\n\n'); fprintf(fid2,'误差变化：\n'); ne=length(eee) for i=1:ne fprintf(fid2,'%4d %12.6f\n',i,eee(i)); end fclose(fid2);