学步园

应用了归一化的预测，在归一化的过程中使用了premnmx和postmnmx，并在最后给出了这两个函数的应用情况。

%应用了归一化的预测<br /> clc;<br /> clear;<br /> originalData=rands(20,14);<br /> inputSampledata=originalData';<br /> outputData=rands(20,1);<br /> outputSampledata=outputData';<br /> % creating neural network and setting trainging parameters<br /> gwwnet=newff(minmax(inputSampledata),[4,1],{'tansig','purelin'},'traingdm');<br /> gwwnet.trainParam.show = 50;<br /> gwwnet.trainParam.lr = 0.05;<br /> gwwnet.trainParam.epochs = 5000;<br /> gwwnet.trainParam.goal = 1e-3;<br /> [input,mininput,maxinput,output,minoutput,maxoutput] = premnmx(inputSampledata,outputSampledata);<br /> % Preprocess data so that minimum is -1 and maximum is 1<br /> %<br /> % Examples<br /> %<br /> % Here is the code to normalize a given data set so that the inputs<br /> % and targets will fall in the range [-1,1].<br /> % p = [-10 -7.5 -5 -2.5 0 2.5 5 7.5 10];<br /> % t = [0 7.07 -10 -7.07 0 7.07 10 7.07 0];<br /> % [pn,minp,maxp,tn,mint,maxt] = premnmx(p,t);<br /> %<br /> % If you just want to normalize the input, [pn,minp,maxp] = premnmx(p);<br /> %<br /> % Algorithm<br /> % pn = 2*(p-minp)/(maxp-minp) - 1;<br /> [gwwnet,tr]=train(gwwnet,input,output);<br /> y=sim(gwwnet,input);<br /> %data offset (converting the network output data to it original unit)<br /> nnoutput = postmnmx(y,minoutput,maxoutput);<br /> %<br /> % Postprocess data that has been preprocessed by premnmx<br /> % Syntax<br /> %<br /> % [P,T] = postmnmx(PN,minp,maxp,TN,mint,maxt)<br /> % [p] = postmnmx(PN,minp,maxp)<br /> % Description<br /> %<br /> % postmnmx postprocesses the network training set<br /> % that was preprocessed by premnmx. It converts the data<br /> % back into unnormalized units.<br /> % postmnmx takes these inputs,<br /> % PN -- R x Q matrix of normalized input vectors<br /> % minp -- R x 1 vector containing minimums for each P<br /> % maxp -- R x 1 vector containing maximums for each P<br /> % TN -- S x Q matrix of normalized target vectors<br /> % mint -- S x 1 vector containing minimums for each T<br /> % maxt -- S x 1 vector containing maximums for each T<br /> % and returns,<br /> % P -- R x Q matrix of input (column) vectors<br /> % T -- R x Q matrix of target vectors<br /> % Examples<br /> %<br /> % In this example we normalize a set of training data with premnmx,<br /> % create and train a network using the normalized data, simulate the network,<br /> % unnormalize the output of the network using postmnmx,<br /> % and perform a linear regression between<br /> % the network outputs (unnormalized) and the targets<br /> % to check the quality of the network training.<br /> % p = [-0.92 0.73 -0.47 0.74 0.29; -0.08 0.86 -0.67 -0.52 0.93];<br /> % t = [-0.08 3.4 -0.82 0.69 3.1];<br /> % [pn,minp,maxp,tn,mint,maxt] = premnmx(p,t);<br /> % net = newff(minmax(pn),[5 1],{'tansig' 'purelin'},'trainlm');<br /> % net = train(net,pn,tn);<br /> % an = sim(net,pn);<br /> % [a] = postmnmx(an,mint,maxt);<br /> % [m,b,r] = postreg(a,t);<br /> %<br /> % Algorithm<br /> % p = 0.5(pn+1)*(maxp-minp) + minp;<br /> %plot<br /> time=1:1:20;<br /> plot(time,outputSampledata,'-v',time,nnoutput,'o');<br /> legend('actual output','NN output');<br /> xlabel('time');ylabel('Learning fitting curve');</p> <p>%scenario1 forecasting process<br /> column=10;<br /> for i=1:column;<br /> SceInput(1,i)=inputSampledata(1,20)*(1.0464^i);<br /> SceInput(2,i)=inputSampledata(2,20)*(1.0631^i);<br /> SceInput(3,i)=inputSampledata(3,20)*(1.0872^i);<br /> SceInput(4,i)=inputSampledata(4,20)*(1.2044^i);<br /> SceInput(5,i)=inputSampledata(5,20)*(1.2326^i);<br /> SceInput(6,i)=inputSampledata(6,20)*(1.0605^i);<br /> SceInput(7,i)=2*(1.01^i);<br /> SceInput(8,i)=42*(1.02^i);<br /> SceInput(9,i)=inputSampledata(9,20)*(1.1426^i);<br /> SceInput(10,i)=inputSampledata(10,20)*(1.017^i);<br /> SceInput(11,i)=inputSampledata(11,20)*(1.0205^i);<br /> SceInput(12,i)=inputSampledata(12,20)*(1.1336^i);<br /> SceInput(13,i)=inputSampledata(13,20)*(1.1599^i);<br /> SceInput(14,i)=inputSampledata(14,20)*(1.1783^i);<br /> end<br /> for j=1:20;<br /> for i=1:14;<br /> recalldata(i,j)=inputSampledata(i,j);<br /> end<br /> end<br /> for j=21:30;<br /> for i=1:14;<br /> recalldata(i,j)=SceInput(i,j-20)<br /> end<br /> end<br /> [alterinput,mininput,maxinput] = premnmx(recalldata);</p> <p>%training<br /> fvalue=sim(gwwnet,alterinput);<br /> %data offset (converting the network output data to it original unit)<br /> forecastvalue = postmnmx(fvalue,minoutput,maxoutput);<br /> %plot<br /> % waitforbuttonpress;<br /> %<br /> % Wait for key or mouse button press<br /> % Syntax<br /> % k = waitforbuttonpress<br /> %<br /> % Description<br /> %<br /> % k = waitforbuttonpress blocks the caller's execution stream<br /> % until the function detects that the user has pressed a mouse button<br /> % or a key while the figure window is active. The function returns 0<br /> % if it detects a mouse button press 1 if it detects a key press<br /> %<br /> % Example<br /> %<br /> % These statements display text in the Command Window<br /> % when the user either clicks a mouse button or<br /> % types a key in the figure window:<br /> % w = waitforbuttonpress;<br /> % if w == 0<br /> % disp('Button press')<br /> % else<br /> % disp('Key press')<br /> % end<br /> % clf;<br /> figure;<br /> time1=1:1:30;<br /> time2=1:1:20;<br /> plot(time1,forecastvalue,'r-',time2,outputSampledata,'-');<br /> legend('预测曲线','实际曲线');<br /> title('曲线');<br /> xlabel('时间');ylabel('量');<br /> %%%%%%%premnmx及postmnmx使用说明<br /> % clc;<br /> % clear;<br /> % p = [1 2 3 4;<br /> % 5 6 8 9 ];<br /> % [p1 pmin pmax]=premnmx(p);<br /> % p2=postmnmx(p1,pmin,pmax);<br /> % p1<br /> % pmin<br /> % pmax<br /> % p2<br /> %<br /> % p1 =<br /> %<br /> % -1.0000 -0.3333 0.3333 1.0000<br /> % -1.0000 -0.5000 0.5000 1.0000<br /> %<br /> %<br /> % pmin =<br /> %<br /> % 1<br /> % 5<br /> %<br /> %<br /> % pmax =<br /> %<br /> % 4<br /> % 9<br /> %<br /> %<br /> % p2 =<br /> %<br /> % 1 2 3 4<br /> % 5 6 8 9<br />