BP神經(jīng)網(wǎng)絡(luò)matlab源程序代碼

BP神經(jīng)網(wǎng)絡(luò)matlab源程序代碼）%******************************%學(xué)習(xí)程序%******************************%%======原始數(shù)據(jù)輸入========p=[284528334488;283344884554;448845542928;455429283497;292834972261;...349722616921;226169211391;692113913580;139135804451;358044512636;...445126363471;263634713854;347138543556;385435562659;355626594335;...265943352882;433528824084;433528821999;288219992889;199928892175;...288921752510;217525103409;251034093729;340937293489;372934893172;...348931724568;317245684015;]';%===========期望輸出=======t=[4554292834972261692113913580445126363471385435562659...4335288240841999288921752510340937293489317245684015...3666];ptest=[284528334488;283344884554;448845542928;455429283497;292834972261;...349722616921;226169211391;692113913580;139135804451;358044512636;...445126363471;263634713854;347138543556;385435562659;355626594335;...265943352882;433528824084;433528821999;288219992889;199928892175;...288921752510;217525103409;251034093729;340937293489;372934893172;...348931724568;317245684015;456840153666]';[pn,minp,maxp,tn,mint,maxt]=premnmx(p,t);%將數(shù)據(jù)歸一化NodeNum1=20;%隱層第一層節(jié)點(diǎn)數(shù)NodeNum2=40;%隱層第二層節(jié)點(diǎn)數(shù)TypeNum=1;%輸出維數(shù)TF1='tansig';TF2='tansig';TF3='tansig';net=newff(minmax(pn),[NodeNum1,NodeNum2,TypeNum],{TF1TF2TF3},'traingdx');%網(wǎng)絡(luò)創(chuàng)建traingdmnet.trainParam.show=50;net.trainParam.epochs=50000;%訓(xùn)練次數(shù)設(shè)置net.trainParam.goal=1e-5;%訓(xùn)練所要達(dá)到的精度net.trainParam.lr=0.01;%學(xué)習(xí)速率net=train(net,pn,tn);p2n=tramnmx(ptest,minp,maxp);%測(cè)試數(shù)據(jù)的歸一化an=sim(net,p2n);[a]=postmnmx(an,mint,maxt)%數(shù)據(jù)的反歸一化，即最終想得到的預(yù)測(cè)結(jié)果plot(1:length(t),t,'o',1:length(t)+1,a,'+');title('o表示預(yù)測(cè)值---*表示實(shí)際值')gridonm=length(a);%向量a的長(zhǎng)度t1=[t,a(m)];error=t1-a;%誤差向量figureplot(1:length(error),error,'-.')title('誤差變化圖')gridon中國(guó)振動(dòng)聯(lián)盟'sArchiver論壇?Matlab討論區(qū)?BP神經(jīng)網(wǎng)絡(luò)輸入向量和目標(biāo)向量的格式問(wèn)題wxywan368發(fā)表于2008-3-1511:46BP神經(jīng)網(wǎng)絡(luò)輸入向量和目標(biāo)向量的格式問(wèn)題P=[-1,

-2,

3,

1;

-1,

1,

5,

-3];

%T為目標(biāo)矢量

T=[-1,-1,1,1];

pause;

clc

%

創(chuàng)建一個(gè)新的前向神經(jīng)網(wǎng)絡(luò)

net=newff(minmax(P),[3,1],{'tansig','purelin'},'traingdm')

%

當(dāng)前輸入層權(quán)值和閾值

inputWeights=net.IW{1,1}

inputbias=net.b{1}

%

當(dāng)前網(wǎng)絡(luò)層權(quán)值和閾值

layerWeights=net.LW{2,1}

layerbias=net.b{2}

pause

clc

%

設(shè)置訓(xùn)練參數(shù)

net.trainParam.show=50;

net.trainParam.lr=0.05;

net.trainParam.mc=0.9;

net.trainParam.epochs=1000;

net.trainParam.goal=1e-3;

pause

clc

%

調(diào)用TRAINGDM算法訓(xùn)練BP網(wǎng)絡(luò)

[net,tr]=train(net,P,T);

pause

clc

%

對(duì)BP網(wǎng)絡(luò)進(jìn)行仿真

A=sim(net,P)

%

計(jì)算仿真誤差

E=T-A

MSE=mse(E)

pause

clc

echooff

上面這個(gè)程序運(yùn)行沒(méi)有問(wèn)題，但是下面我把

按行向量寫入輸入向量P=[0.3459,0.3548,0.475,0.6691,0.6903;0.3595,0.3705,0.4932,0.6848,0.7302;0.3574,0.3688,0.4914

0.6809,0.7258;0.3592,0.3695,0.493,0.6846,0.7298;0.3198,0.3492,0.4554,0.65250.6486;0.345,0.3542,0.4731,0.6687,0.6899;0.3481,0.3579,0.4776,0.6742,0.6957;0.3626,0.4094,0.52150.7033,0.7915;0.3646,0.4114,0.5235,0.7079,0.7975;0.344,0.3525,0.4728,0.6646,0.6854]

目標(biāo)向量Q=[206,34.5,7850,2600;226,38.5,7850,2600;206,34.5,7650,2500;226,34.5,7650,2500;186,30.5,

7850,2600;206,34.5,7450,2200;206,40.5,7450,2200;266,34.5,7850,2600;266,38.5,7850,2600;206,30.5,7450,2400]

輸入向量為10x5矩陣，輸出向量為10x4矩陣

當(dāng)程序運(yùn)行時(shí)，提示Matricesallhave1rows，請(qǐng)教下怎么實(shí)現(xiàn)上面輸入與輸出的bp實(shí)現(xiàn)。

望各位可以不吝賜教，先謝謝！F117_ren_0發(fā)表于2008-3-1618:32%

創(chuàng)建一個(gè)新的前向神經(jīng)網(wǎng)絡(luò)

net=newff(minmax(P),[3,1],{'tansig','purelin'},'traingdm')

%

當(dāng)前輸入層權(quán)值和閾值

inputWeights=net.IW{1,1}

inputbias=net.b{1}

你寫的(3,1)你還指望能輸出10*4的矩陣?

你寫的代碼有問(wèn)題,輸入和輸出有問(wèn)題.

[[i]本帖最后由F117_ren_0于2008-3-1618:35編輯[/i]]wxywan368發(fā)表于2008-3-1619:57問(wèn)題找出來(lái)了，net=newff(minmax(P),[3,1],{'tansig','purelin'},'traingdm')

newff()為建立BP神經(jīng)網(wǎng)絡(luò)的函數(shù)，minmax(pn)表示樣本數(shù)據(jù)經(jīng)預(yù)處理后的網(wǎng)絡(luò)輸入pn的取值范圍，[3，1]表示隱層節(jié)點(diǎn)數(shù)是3，輸出層節(jié)點(diǎn)數(shù)是1，{'tansig','purelin'}表示隱含層中的神經(jīng)元采用tansig轉(zhuǎn)換函數(shù)，輸出層采用purelin函數(shù)，'traingdm'表示選擇的學(xué)習(xí)算法。

我的矩陣輸出矩陣是10*4，輸出層節(jié)點(diǎn)數(shù)應(yīng)該為4，隱含層節(jié)點(diǎn)數(shù)可參考下式：m=(p+q)0.5+a,其中m為隱層節(jié)點(diǎn)數(shù)。隱層節(jié)點(diǎn)應(yīng)該為1-10之間的常數(shù)，我更改為10.

還有樓上仁兄提出的輸入和輸出問(wèn)題，后來(lái)經(jīng)多方面查證，是我的矩陣沒(méi)有做歸一化處理，誤差得不到收斂。最后對(duì)其進(jìn)行歸一化處理后，并把輸出矩陣改成10*5，