BP_BP分类_BP分类_多分类_matlabBP神经网络

clear all close all clc load databp.mat; % 将第一类1-30，第二类的60-95，第三类的131-153作为训练集 train_bpdata = [bpdata(1:400,2:end);bpdata(501:900,2:end);bpdata(1001:1400,2:end)]; train_bpdata_labels = [bpdata(1:400,1);bpdata(501:900,1);bpdata(1001:1400,1)]; % 将第一类31-59，第二类的96-130，第三类的154-178作为测试集 test_bpdata = [bpdata(300:500,2:end);bpdata(600:800,2:end);bpdata(1100:1300,2:end)]; test_bpdata_labels = [bpdata(300:500,1);bpdata(600:800,1);bpdata(1100:1300,1)]; [t,r] = size(train_bpdata); for i = 1:t if train_bpdata_labels(i,1) == 1 train_labels(i,:) = [1;0;0]; elseif train_bpdata_labels(i,1) == 2 train_labels(i,:) = [0;1;0]; else train_labels(i,:) = [0;0;1]; end end %对测试数据集输出类别重新设置 [t1,r1] = size(test_bpdata); for j = 1:t1 if test_bpdata_labels(j,1) == 1 test_labels(j,:) = [1;0;0]; elseif test_bpdata_labels(j,1) == 2 test_labels(j,:) = [0;1;0]; else test_labels(j,:) = [0;0;1]; end end %% normalize [train_data,ps] = mapminmax(train_bpdata',0,1); [test_data,ps1] = mapminmax(test_bpdata',0,1); %% network net = newff(train_data,train_labels', [10], { 'logsig' 'purelin' } , 'traingdx' , 'learngdm') ;%输入数据为 特征数*数据个数，输出为 类别向量*数据个数 net.trainparam.show = 50 ;%每间隔50步显示一次训练结果 net.trainparam.epochs = 1000 ;%允许最大训练步数500步 net.trainparam.goal = 0.000001 ;%训练目标最小误差0.01 net.trainParam.lr = 0.001 ;%学习速率0.05 tic; %% 开始训练 net = train( net, train_data , train_labels' ); toc; %% 仿真测试 Y = sim( net,test_data ) ; %统计识别正确率 [s1,s2] = size( Y ) ; hitNum = 0 ; count = 0; for i = 1:s2 [m,Index] = max(Y(:,i )); [m_c,Index_c] = max(test_labels(i,:)'); if( Index == Index_c) hitNum = hitNum + 1 ; else count = count + 1; end end sprintf('识别率是 %3.3f%%',100 * hitNum / s2 )