基于前馈神经网络 (FNN)时间序列预测(Matlab完整源码和数据）

close all; clear; %FNN model to forecast the next 20 steps %train and test data setting data=xlsread('200014pressure--.csv'); x=data(:,1); y=data(:,2:end); n=length(y); max=zeros(1,size(y,2)); min=zeros(1,size(y,2)); for i=1:length(max) max(1,i)=y(1,i); min(1,i)=y(1,i); for j=1:n tem=y(j,i); if tem>max(1,i) max(1,i)=tem; end if tem<min(1,i) min(1,i)=tem; end end end y1=zeros(n,length(max)); y2=zeros(n-2,2*length(max)); y3=zeros(n-2,length(max)); %data normalized for i=1:n for j=1:length(max) y1(i,j)=(0.1*(max(1,j)-y(i,j))+0.9*(y(i,j)-min(1,j)))/(max(1,j)-min(1,j)); end end for i=1:n-2 y2(i,1:length(max))=y1(i,:); y2(i,length(max)+1:end)=y1(i+1,:); y3(i,:)=y1(i+2,:); end %prediction model setting size=int16(n*0.7); train_x=y2(1:size,:); train_y=y3(1:size,:); test_x=y2(size+1:end,:); test_y=y3(size+1:end,:); net1=newff(y2',y3',8*length(max)); net1.trainParam.epochs=5000; net1.trainParam.goal=0.0001; net1.trainParam.lr=0.001; %net1.performFcn='mse'; net1.divideParam.trainRatio=0.7; net1.divideParam.testRatio=0.3; net1=train(net1,train_x',train_y'); %prediction on test set for the next 20 steps %rolling forecasting method m=20; len=length(max); output1(1:len,:)=sim(net1,test_x'); input1(1:2*len,:)=test_x'; if m>1 for i=2:m input1(len*i+1:len*(i+1),:)=output1(len*(i-2)+1:len*(i-1),:); output1(len*(i-1)+1:len*i,:)=sim(net1,input1(len*(i-1)+1:len*(i+1),:)); end end %data inverse normalized for i=1:n-size-2 for j=1:len test_x(i,j)=(test_x(i,j)*(max(1,j)-min(1,j))+0.9*min(1,j)-0.1*max(1,j))/0.8; test_x(i,j+len)=(test_x(i,j+len)*(max(1,j)-min(1,j))+0.9*min(1,j)-0.1*max(1,j))/0.8; test_y(i,j)=(test_y(i,j)*(max(1,j)-min(1,j))+0.9*min(1,j)-0.1*max(1,j))/0.8; end end for i=1:m for j=1:len for k=1:n-size-2 output1(len*(i-1)+j,k)=(output1(len*(i-1)+j,k)*(max(1,j)-min(1,j))+0.9*min(1,j)-0.1*max(1,j))/0.8; end end end %evaluation indicators calculated measure1=zeros(m,60); for i=1:m measure1(i,1)=mae(abs(test_y(:,1)-output1(1+len*(i-1),:)')); measure1(i,2)=mse(test_y(:,1),output1(1+len*(i-1),:)'); measure1(i,3)=sqrt(mse(test_y(:,1),output1(1+len*(i-1),:)')); measure1(i,4)=mae(abs(test_y(:,2)-output1(2+len*(i-1),:)')); measure1(i,5)=mse(test_y(:,2),output1(2+len*(i-1),:)'); measure1(i,6)=sqrt(mse(test_y(:,2),output1(2+len*(i-1),:)')); end %save prediction values and evaluation indicators % tempdata1=xlsread('bp514test1.xlsx'); % clear size; % n11=size(tempdata1,2); % clear size; % n12=size(output1,1); % temdata1(:,1:n11)=tempdata1; % temdata1(:,n11+1:n11+n12)=output1'; % xlswrite('bp514test1.xlsx',temdata1); % tempdata3=xlsread('bp514test0.xlsx'); % clear size; % n15=size(tempdata3,1); % clear size; % n16=size(measure1,1); % temdata3(1:n15,:)=tempdata3; % temdata3(n15+1:n15+n16,:)=measure1; % xlswrite('bp514test0.xlsx',temdata3); %prediction on typhoon cases % number=[201509,201513,201521,201614,201622,201713,201720,201808,201822]; % for i=1:1 % %data processing % ddata=strcat(num2str(number(i)),'pressure--.csv'); % data1=xlsread(ddata); % ss=data1(:,2:end); % nn=length(data1); % max1=zeros(1,len); % min1=zeros(1,len); % for r=1:len % max1(1,r)=ss(1,r); % min1(1,r)=ss(1,r); % for s=1:nn % temp=ss(s,r); % if temp>max1(1,r) % max1(1,r)=temp; % end % if temp<min1(1,r) % min1(1,r)=temp; % end % end % end % %data normalized % for r=1:nn % for s=1:len % ss(r,s)=(0.1*(max1(1,s)-ss(r,s))+0.9*(ss(r,s)-min1(1,s)))/(max1(1,s)-min1(1,s)); % end % end % %prediction on typhoon case for the next 20 steps % %rolling forecasting method % clear y4; % clear input3; % y4(:,1:len)=ss(1:end-1,:); % y4(:,len+1:2*len)=ss(2:end,:); % result1=zeros(m*len,nn-1); % result1(1:len,:)=sim(net1,y4'); % input3(1:2*len,:)=y4'; % if m>1 % for r=2:m % input3(len*i+1:len*(i+1),:)=result1(len*(r-2)+1:len*(r-1),:); % result1(len*(r-1)+1:len*r,:)=sim(net1,input3(len*(i-1)+1:len*(i+1),:)); % end % end % %data inverse normalized % for r=1:nn-1 % for s=1:len % y4(r,s)=(y4(r,s)*(max1(1,s)-min1(1,s))+0.9*min1(1,s)-0.1*max1(1,s))/0.8; % y4(r,s+len)=(y4(r,s+len)*(max1(1,s)-min1(1,s))+0.9*min1(1,s)-0.1*max1(1,s))/0.8; % end % end % for r=1:nn % for s=1:len % ss(r,s)=(ss(r,s)*(max1(1,s)-min1(1,s))+0.9*min1(1,s)-0.1*max1(1,s))/0.8; % end % end % for r=1:m % for s=1:len % for t=1:nn-1 % result1(len*(r-1)+s,t)=(result1(len*(r-1)+s,t)*(max1(1,s)-min1(1,s))+0.9*min1(1,s)-0.1*max1(1,s))/0.8; % end % end % end % %evaluation indicators calculated % for r=1:m % measure1(r,6*i+1)=mae(abs(ss(r+2:end,1)-result1(1+len*(r-1),1:end-r)')); % measure1(r,6*i+2)=mse(ss(r+2:end,1),result1(1+len*(r-1),1:end-r)'); % measure1(r,6*i+3)=sqrt(mse(ss(r+2:end,1),result1(1+len*(r-1),1:end-r)')); % measure1(r,6*i+4)=mae(abs(ss(r+2:end,2)-result1(2+len*(r-1),1:end-r)')); % measure1(r,6*i+5)=mse(ss(r+2:end,2),result1(2+len*(r-1),1:end-r)'); % measure1(r,6*i+6)=sqrt(mse(ss(r+2:end,2),result1(2+len*(r-1),1:end-r)')); % end % %save the prediction values and evaluation indicators % clear temdata5; % clear temdata6; % name1=strcat('bp514',num2str(number(i)),'1.xlsx'); % name2=strcat('bp514',num2str(number(i)),'0.xlsx'); % tempdata5=xlsread(name1); % tempdata6=xlsread(name2); % clear size; % n21=size(tempdata5,2); % n31=size(tempdata6,1); % clear size; % n22=size(result1,1); % n32=size(measure1,2); % temdata5(:,1:n21)=tempdata5; % temdata5(:,n21+1:n21+n22)=result1'; % temdata6(1:n31,:)=tempdata6; % temdata6(n31+1:n31+n32,:)=measure1; % xlswrite(name1,temdata5); % xlswrite(name2,temdata6); % end