《BP神经网络matlab实例(简单而经典).doc》由会员分享，可在线阅读，更多相关《BP神经网络matlab实例(简单而经典).doc（7页珍藏版）》请在金锄头文库上搜索。
1、学习神经网络的好助手，可以仿照其中的代码，只需修改个别参数便可以轻易实现自己需要完成的任务。p=p1;t=t1; pn,minp,maxp,tn,mint,maxt=premnmx(p,t); %原始数据归一化% premnmx Preprocesses data so that minimum is -1 and maximum is =newff(minmax(pn),5,1,tansig,purelin,traingdx); %设置网络,建立相应的BP网络% minmax returns the Rx2 matrix PR of minimum and maximum values fo
2、r each row of P.net.trainParam.show=2000; % 训练网络net.trainParam.lr=0.01;net.trainParam.epochs=100000;net.trainParam.goal=1e-5;net,tr=train(net ,pn,tn); %调用TRAINGDM算法训练BP网络pnew=pnew1; pnewn=tramnmx(pnew,minp,maxp);% tramnmx TRAMNMX transforms the network input set using % minimum and maximum values th
3、at were previously % computed by PREMNMX. This function needs to be %used when a network has been trained using data normalized %by PREMNMX. All subsequent inputs to the network need to %be transformed using the same normalization.anewn=sim(net,pnewn); %对BP网络进行仿真anew=postmnmx(anewn,mint,maxt); %还原数据
4、y=anew;% Postprocesses data which has been preprocessed by PREMNMX. POSTMNMX postprocesses %the network training % set which was preprocessed by PREMNMX. It converts % the data back into unnormalized units.1、BP网络构建（1）生成BP网络：由维的输入样本最小最大值构成的维矩阵。：各层的神经元个数。：各层的神经元传递函数。：训练用函数的名称。（2）网络训练（3）网络仿真tansig,pure
5、lin,trainrpBP网络的训练函数 训练方法训练函数梯度下降法traingd有动量的梯度下降法traingdm自适应lr梯度下降法traingda自适应lr动量梯度下降法traingdx弹性梯度下降法trainrpFletcher-Reeves共轭梯度法traincgfPloak-Ribiere共轭梯度法traincgpPowell-Beale共轭梯度法traincgb量化共轭梯度法trainscg拟牛顿算法trainbfg一步正割算法trainossLevenberg-MarquardttrainlmBP网络训练参数训练参数参数介绍训练函数net.trainParam.epochs最大
6、训练次数（缺省为10）traingd、traingdm、traingda、traingdx、trainrp、traincgf、traincgp、traincgb、trainscg、trainbfg、trainoss、trainlmnet.trainParam.goal训练要求精度（缺省为0）traingd、traingdm、traingda、traingdx、trainrp、traincgf、traincgp、traincgb、trainscg、trainbfg、trainoss、trainlmnet.trainParam.lr学习率（缺省为0.01）traingd、traingdm、trai
7、ngda、traingdx、trainrp、traincgf、traincgp、traincgb、trainscg、trainbfg、trainoss、trainlmnet.trainParam.max_fail最大失 败次数（缺省为5）traingd、traingdm、traingda、traingdx、trainrp、traincgf、traincgp、traincgb、trainscg、trainbfg、trainoss、trainlmnet.trainParam.min_grad最小梯度要求（缺省为1e-10）traingd、traingdm、traingda、traingdx、tra
8、inrp、traincgf、traincgp、traincgb、trainscg、trainbfg、trainoss、trainlmnet.trainParam.show显示训练迭代过程（NaN表示不显示，缺省为25）traingd、traingdm、traingda、traingdx、trainrp、traincgf、traincgp、traincgb、trainscg、trainbfg、trainoss、trainlmnet.trainParam.time最大训练时间（缺省为inf）traingd、traingdm、traingda、traingdx、trainrp、traincgf、tr
9、aincgp、traincgb、trainscg、trainbfg、trainoss、trainlmnet.trainParam.mc动量因子（缺省0.9）traingdm、traingdxnet.trainParam.lr_inc学习率lr增长比（缺省为1.05）traingda、traingdxnet.trainParam.lr_dec学习率lr下降比（缺省为0.7）traingda、traingdxnet.trainParam.max_perf_inc表现函数增加最大比（缺省为1.04）traingda、traingdxnet.trainParam.delt_inc权值变化增加量（缺省为
10、1.2）trainrpnet.trainParam.delt_dec权值变化减小量（缺省为0.5）trainrpnet.trainParam.delt0初始权值变化（缺省为0.07）trainrpnet.trainParam.deltamax权值变化最大值（缺省为50.0）trainrpnet.trainParam.searchFcn一维线性搜索方法（缺省为srchcha）traincgf、traincgp、traincgb、trainbfg、trainossnet.trainParam.sigma因为二次求导对权值调整的影响参数（缺省值5.0e-5）trainscgnet.trainPara
11、m.lambdaHessian矩阵不确定性调节参数（缺省为5.0e-7）trainscgnet.trainParam.men_reduc控制计算机内存/速度的参量，内存较大设为1，否则设为2（缺省为1）trainlmnet.trainParam.mu的初始值（缺省为0.001）trainlmnet.trainParam.mu_dec的减小率（缺省为0.1）trainlmnet.trainParam.mu_inc的增长率（缺省为10）trainlmnet.trainParam.mu_max的最大值（缺省为1e10）trainlm2、BP网络举例举例1、%traingdclear;clc;P=-1
12、 -1 2 2 4;0 5 0 5 7;T=-1 -1 1 1 -1;%利用minmax函数求输入样本范围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%隐层权值net.b1%隐层阈值net.lw2,1%输出层权值net.b2%输出层阈值sim(net,P)举例2、利用三层BP神经网络来
13、完成非线性函数的逼近任务，其中隐层神经元个数为五个。样本数据：输入X输出D输入X输出D输入X输出D-1.0000-0.9602-0.30000.13360.40000.3072-0.9000-0.5770-0.2000-0.20130.50000.3960-0.8000-0.0729-0.1000-0.43440.60000.3449-0.70000.37710-0.50000.70000.1816-0.60000.64050.1000-0.39300.8000-0.3120-0.50000.66000.2000-0.16470.9000-0.2189-0.40000.46090.3000-0
14、.09881.0000-0.3201解：看到期望输出的范围是，所以利用双极性Sigmoid函数作为转移函数。程序如下：clear;clc;X=-1:0.1:1;D=-0.9602 -0.5770 -0.0729 0.3771 0.6405 0.6600 0.4609. 0.1336 -0.2013 -0.4344 -0.5000 -0.3930 -0.1647 -.0988. 0.3072 0.3960 0.3449 0.1816 -0.312 -0.2189 -0.3201;figure;plot(X,D,*); %绘制原始数据分布图（附录：1-1）net = newff(-1 1,5 1,tansig,tansig);net.trainParam.epochs = 100; %训练的最大次数net.trainParam.goal = 0.005; %全局最小误差net = trai
