BP神經(jīng)網(wǎng)絡(luò)matlab實(shí)例簡(jiǎn)單而經(jīng)典

1、%調(diào)用TRAINGDM算法訓(xùn)練pn,minp,maxp,tn,mint,maxt=premnmx(p,t);%原始數(shù)據(jù)歸一化net=newff(minmax(pn),5,1,tansig,purelin,traingdx);%設(shè)置網(wǎng)絡(luò)，建立相應(yīng)的BP網(wǎng)絡(luò)net.trainParam.show=2000;%訓(xùn)練網(wǎng)絡(luò)net.trainParam.lr=0.01;net.trainParam.epochs=100000;net.trainParam.goal=1e-5;BP網(wǎng)絡(luò)pnew=pnew1;pnewn=tramnmx(pnew,minp,maxp);anewn=sim(net,pnewn);

2、%對(duì)BP網(wǎng)絡(luò)進(jìn)行仿真anew=postmnmx(anewn,mint,maxt);%還原數(shù)據(jù)y=anew;1、BP網(wǎng)絡(luò)構(gòu)建(1)生成BP網(wǎng)絡(luò)netnewff(PR,S1S2.SNl,TF1TF2.TFNI,BTF,BLF,PF)net,tr=train(net,pn,tn);PR：由R維的輸入樣本最小最大值構(gòu)成的R2維矩陣。S1S2.SN1：各層的神經(jīng)元個(gè)數(shù)。TF1TF2.TFN1：各層的神經(jīng)元傳遞函數(shù)。BTF:訓(xùn)練用函數(shù)的名稱。(2)網(wǎng)絡(luò)訓(xùn)練net,tr,Y,E,Pf,Aftrain(net,P,T,Pi,Ai,VV,TV)(3)網(wǎng)絡(luò)仿真Y,Pf,Af,E,perfsim(net,P,Pi,

3、Ai,T)tansig,purelin,trainrpBP網(wǎng)絡(luò)的訓(xùn)練函數(shù)訓(xùn)練方法訓(xùn)練函數(shù)梯度卜降法traingd有動(dòng)量的梯度卜降法traingdm自適應(yīng)lr梯度卜降法traingda自適應(yīng)lr動(dòng)量梯度卜降法traingdx彈性梯度卜降法trainrpFletcher-Reeves共軻梯度法traincgfPloak-Ribiere共軻梯度法traincgpPowell-Beale共軻梯度法traincgb量化共輾梯度法trainscg擬牛頓算法trainbfg一步正割算法trainossLevenberg-MarquardttrainlmBP網(wǎng)絡(luò)訓(xùn)練參數(shù)訓(xùn)練參數(shù)參數(shù)介紹訓(xùn)練函數(shù)net.trai

4、nParam.epochs最大訓(xùn)練次數(shù)（缺省為10）traingd、traingdm、traingda、traingdx、trainrp、traincgf、traincgp、traincgb、trainscg、trainbfg、trainoss、trainlmnet.trainParam.goal訓(xùn)練要求精度（缺省為0）traingd、traingdm、traingda、traingdx、trainrp、traincgf、traincgp、traincgb、trainscg、trainbfg、trainoss、trainlmnet.trainParam.lr學(xué)習(xí)率（缺省為0.01）traing

5、d、traingdm、traingda、traingdx、trainrp、traincgf、traincgp、traincgb、trainscg、trainbfg、trainoss、trainlmnet.trainParam.max_fail最大失敗次數(shù)（缺省為5）traingd、traingdm、traingda、traingdx、trainrp、traincgf、traincgp、traincgb、trainscg、trainbfg、trainoss、trainlmnet.trainParam.min_grad最小梯度要求（缺省為traingd、traingdm、1e-10)traingd

6、a、traingdx、trainrp、traincgf、traincgp、traincgb、trainscg、trainbfg、trainoss、trainlmnet.trainParam.show顯示訓(xùn)練迭代過程（NaN表traingd、traingdm、布/、顯不，缺省為25）traingda、traingdx、trainrp、traincgf、traincgp、traincgb、trainscg、trainbfg、trainoss、trainlmnet.trainParam.time最大訓(xùn)練時(shí)間（缺省為inf）traingd、traingdm、traingda、traingdx、trai

7、nrp、traincgf、traincgp、traincgb、trainscg、trainbfg、trainoss、trainlmnet.trainParam.mc動(dòng)量因子（缺省0.9）traingdm、traingdxnet.trainParam.lr_inc學(xué)習(xí)率lr增長(zhǎng)比（缺省為1.05）traingda、traingdxnet.trainParam.lr_dec學(xué)習(xí)率lr下降比（缺省為0.7）traingda、traingdxnet.trainParam.max_perf_inc表現(xiàn)函數(shù)增加最大比（缺省為1.04）traingda、traingdxnet.trainParam.delt

8、_inc權(quán)值變化增加量（缺省為1.2）trainrpnet.trainParam.delt_dec權(quán)值變化減小量（缺省為0.5）trainrpnet.trainParam.delt0初始權(quán)值變化（缺省為0.07）trainrpnet.trainParam.deltamax權(quán)值變化最大值（缺省為50.0）trainrpnet.trainParam.searchFcn一維線性搜索方法（缺省為srchcha）traincgf、traincgp、traincgb、trainbfg、trainossnet.trainParam.sigma因?yàn)槎吻髮?dǎo)對(duì)權(quán)值調(diào)整的影響參數(shù)（缺省值5.0e-5）trains

9、cgnet.trainParam.lambdaHessian矩陣小確定性調(diào)節(jié)trainscg參數(shù)（缺省為5.0e-7）net.trainParam.men_reduc控制計(jì)算機(jī)內(nèi)存/速度的參量，內(nèi)存較大設(shè)為1,否則設(shè)為2（缺省為1）trainlmnet.trainParam.mu的初始值（缺省為0.001）trainlmnet.trainParam.mu_dec的減小率（缺省為0.1）trainlmnet.trainParam.mu_inc的增長(zhǎng)率（缺省為10）trainlmnet.trainParam.mu_max的最大值（缺省為1e10）trainlm2、BP網(wǎng)絡(luò)舉例舉例1、%traing

10、dclear;clc;P=-1-1224;05057;T=-1-111-1;%利用minmax函數(shù)求輸入樣本范圍net=newff(minmax(P),5,1,tansig,purelin,trainrp);net.trainParam.show=50;%net.trainParam.lr=0.05;net.trainParam.epochs=300;net.trainParam.goal=1e-5;net,tr=train(net,P,T);net.iw1,1%隱層權(quán)值net.b1%隱層閾值net.lw2,1%輸出層權(quán)值net.b2%輸出層閾值sim(net,P)舉例2、利用三層BP神經(jīng)網(wǎng)絡(luò)

11、來(lái)完成非線性函數(shù)的逼近任務(wù)，其中隱層神經(jīng)元個(gè)數(shù)為五個(gè)。樣本數(shù)據(jù)：輸入X輸出D輸入X輸出D輸入X輸出D-1.0000-0.9602-0.30000.13360.40000.3072-0.9000-0.5770-0.2000-0.20130.50000.3960-0.8000-0.0729-0.1000-0.43440.60000.3449-0.70000.37710-0.50000.70000.1816-0.60000.64050.1000-0.39300.8000-0.3120-0.50000.66000.2000-0.16470.9000-0.2189-0.40000.46090.3000-

12、0.09881.0000-0.3201解:看到期望輸出的范圍是1,1,所以利用雙極性Sigmoid函數(shù)作為轉(zhuǎn)移函數(shù)。程序如下：clear;clc;X=-1:0,1:1;D=-0.9602-0.5770-0.07290.37710.64050.66000.4609.0.1336-0.2013-0.4344-0.5000-0.3930-0.1647-.0988.0.30720.39600.34490.1816-0.312-0.2189-0.3201;figure;plot(X,D,*);%繪制原始數(shù)據(jù)分布圖(附錄：1-1)net=newff(-11,51,tansig,tansig);net.tr

13、ainParam.epochs=100;%訓(xùn)練的最大次數(shù)net.trainParam.goal=0.005;%全局最小誤差net=train(net,X,D);O=sim(net,X);figure;plot(X,D,*,X,O);%繪制訓(xùn)練后得到的結(jié)果和誤差曲線(附錄：1-2、1-3)V=net.iw1,1%輸入層到中間層權(quán)值theta1=net.b1%中間層各神經(jīng)元閾值W=net.lw2,1%中間層到輸出層權(quán)值theta2=net.b2%輸出層各神經(jīng)元閾值所得結(jié)果如下：輸入層到中間層的權(quán)值：V-9.16697.34487.37614.89663.5409)中間層各神經(jīng)元的閾值：6.5885

14、-2.4019-0.99621.53033.273T中間層到輸出層的權(quán)值：W0.34270.21350.2981-0.88401.9134輸出層各神經(jīng)元的閾值：T-1.5271舉例3、利用三層BP神經(jīng)網(wǎng)絡(luò)來(lái)完成非線性函數(shù)的逼近任務(wù)，其中隱層神經(jīng)元個(gè)數(shù)為五個(gè)。樣本數(shù)據(jù)：輸入X輸出D輸入X輸出D輸入X輸出D00448211539322621043371解:看到期望輸出的范圍超出1,1,所以輸出層神經(jīng)元利用線性函數(shù)作為轉(zhuǎn)移函數(shù)。程序如下：clear;clc;X=012345678910;D=01234321234;figure;plot(X,D,*);%繪制原始數(shù)據(jù)分布圖net=newff(010,

15、51,tansig,purelin)net.trainParam.epochs=100;net.trainParam.goal=0.005;net=train(net,X,D);O=sim(net,X);figure;plot(X,D,*,X,O);%繪制訓(xùn)練后得到的結(jié)果和誤差曲線(附錄：2-2、2-3)V=net.iw1,1%輸入層到中間層權(quán)值theta1=net.b1%中間層各神經(jīng)元閾值W=net.lw2,1%中間層到輸出層權(quán)值theta2=net.b2%輸出層各神經(jīng)元閾值所得結(jié)果如下：T輸入層到中間層的權(quán)值：V0.85842.0890-1.21660.2752-0.3910中間層各神經(jīng)元

16、的閾值：-14.0302-9.83407.4331-2.01350.561CT中間層到輸出層的權(quán)值：W-0.4675-1.12342.32084.6402-2.2686輸出層各神經(jīng)元的閾值：T1.7623問題：以下是上證指數(shù)2009年2月2日到3月27日的收盤價(jià)格，構(gòu)建一個(gè)三層BP神經(jīng)網(wǎng)絡(luò)，利用該組信號(hào)的6個(gè)過去值預(yù)測(cè)信號(hào)的將來(lái)值。日期價(jià)格日期價(jià)格2009/02/022011.6822009/03/022093.4522009/02/032060.8122009/03/032071.4322009/02/042107.7512009/03/042198.1122009/02/052098.02

17、12009/03/052221.0822009/02/062181.2412009/03/062193.0122009/02/092224.7112009/03/092118.7522009/02/102265.1612009/03/102158.5722009/02/112260.8222009/03/112139.0212009/02/122248.0922009/03/122133.8812009/02/132320.7922009/03/132128.8512009/02/162389.3922009/03/162153.2912009/02/172319.4422009/03/172

18、218.3312009/02/182209.8622009/03/182223.7312009/02/192227.1322009/03/192265.7612009/02/202261.4822009/03/202281.0912009/02/232305.7822009/03/232325.4812009/02/242200.6522009/03/242338.4212009/02/252206.5722009/03/252291.5512009/02/262121.2522009/03/262361.7012009/02/272082.8522009/03/272374.44loaddata3_1.txt;m,n=size(data3_1);tsx=data3_1(1:m-1,1);tsx=tsx;ts=data3_1(2:m,1);ts=ts;TSX,TSXps=mapminmax(tsx,1,2);TS,TSps=mapmi