基于CNN-GRU/CNN-LSTM神经网络的刀具剩余使用寿命预测（matlab代码实现）

最新推荐文章于 2024-01-17 11:21:56 发布

向秃头前进中

最新推荐文章于 2024-01-17 11:21:56 发布

阅读量1.6k

点赞数 1

文章标签：神经网络 cnn gru

本文链接：https://blog.csdn.net/weixin_49890890/article/details/130635679

版权

1、代码实现

% clear
% clc
% load('data1.mat')

% % 数据集构建（列为特征，行为样本数目
% for i =1:7
%     l(:,i) = feature_1.p1(:,i) ;
%     l(:,i+7) = feature_1.p2(:,i);
%     l(:,i+14) = feature_1.p3(:,i)  ;
%     l(:,i+21) = feature_1.p4(:,i)  ;
%     l(:,i+28) = feature_1.p5(:,i)  ;
%     l(:,i+35) = feature_1.p6(:,i)  ;
%     l(:,i+42) = feature_1.p7(:,i)  ;
%     l(:,i+49) = feature_1.p8(:,i)  ;
%     l(:,i+56) = feature_1.p9(:,i)  ;
%     l(:,i+63) = feature_1.p10(:,i)  ;
%     l(:,i+70) = feature_1.f1(:,i)  ;
%     l(:,i+77) = feature_1.f2(:,i)  ;
%     l(:,i+84) = feature_1.f3(:,i)  ;
%     l(:,i+91) = feature_1.f4(:,i)  ;
%     l(:,i+98) = feature_1.f5(:,i)  ;
%     l(:,i+105) = feature_1.s1(:,i)  ;
%     l(:,i+112) = feature_1.s2(:,i)  ;
%     l(:,i+119) = feature_1.s3(:,i)  ;
%     l(:,i+126) = feature_1.s4(:,i)  ;
%     l(:,i+135) = feature_1.t1(:,i)  ;
%     l(:,i+140) = feature_1.t2(:,i)  ;
%     l(:,i+147) = feature_1.t3(:,i)  ;
%     l(:,i+154) = feature_1.t4(:,i)  ;
%     l(:,i+161) = feature_1.t5(:,i)  ;
%     l(:,i+168) = feature_1.t6(:,i)  ;
%     l(:,i+175) = feature_1.t7(:,i)  ;
%     l(:,i+182) = feature_1.t8(:,i)  ;
% end 
% 
%% 数据集降维提取 相关性较高的特征值。
% 加载数据
clear
clc
load('D.mat')

for f= 1:size(data,2)
    Ti(:,f) = normalize(data(:,f)); 
    d(1,f) = corr(life,Ti(:,f),'type','Spearman');
end
% 相关性系数
     d = abs(d);           %   将相关系数绝对值
    [j main_feature] = find(d>=0.8);
%    % ±0.80-±1.00  高度相关   
     M = length(main_feature);
 for k = 1:M  
     p = main_feature(1,k);
     Data1(:,k) = data1(:,p);    %    提取降维特征
     Data2(:,k) = data2(:,p);
     Data3(:,k) = data3(:,p);
     Data4(:,k) = data4(:,p);
     Data5(:,k) = data5(:,p);
     Data6(:,k) = data6(:,p);     
 end   

%% 设置训练集及测试集
y = [life1;life1];
y = [y;life1];

x = [Data1;Data1];
x = [x;Data1];
[xnorm,xopt] = mapminmax(x',0,1);
[ynorm,yopt] = mapminmax(y',0,1);
x = x';

% 转换成2-D image
for i = 1:size(Data1,1)
    Train_xNorm{i} = reshape(xnorm(:,i),M,1,1);
    Train_yNorm(i) = ynorm(i);
end
Train_yNorm= Train_yNorm';

% 测试集设置
%% 测试集
ytest = life2;
xtest = Data2;
[xtestnorm] = mapminmax('apply', xtest',xopt);
[ytestnorm] = mapminmax('apply',ytest',yopt);
xtest = xtest';
% 转换成2-D image
for i = 1:315
    Test_xNorm{i} = reshape(xtestnorm(:,i),M,1,1);
    Test_yNorm(i) = ytestnorm(i);
    Test_y(i) = ytest(i);
end
Test_yNorm = Test_yNorm';
%% 随机种子数
% % % % % % % % 设定随机种子数
w=1;
s=1;

% for w =1:100
%    for s =1:10
% %寻找最优隐含层
rng('default');
% rand(27,1);

% LSTM 层设置，参数设置
inputSize = size(Train_xNorm{1},1);   %数据输入x的特征维度
outputSize = 1;  %数据输出y的维度  
numhidden_units1=41;
numhidden_units2=20;
% numhidden_units3=8;
% numhidden_units4=5;
%


opts = trainingOptions('adam', ...
    'MaxEpochs',20, ...
    'GradientThreshold',1,...
    'ExecutionEnvironment','cpu',...
    'InitialLearnRate',0.00005, ...
    'LearnRateSchedule','piecewise', ...
    'LearnRateDropPeriod',10, ...   %2个epoch后学习率更新
    'LearnRateDropFactor',0.5, ...
    'Shuffle','once',...  % 时间序列长度
    'MiniBatchSize',1,...     
     'Verbose',0,...
     'Plots','training-progress' );           % 打印训练进度            
%    

 layers = [ ...
     
             sequenceInputLayer([inputSize,1,1],'name','input')   %输入层设置
             sequenceFoldingLayer('name','fold')
             convolution2dLayer([1,1],16,'Stride',[1,1],'name','conv1')
             batchNormalizationLayer('name','batchnorm1')
             reluLayer('name','relu1')
             maxPooling2dLayer([1,3],'Stride',1,'Padding','same','name','maxpool')
             convolution2dLayer([2,1],10,'Stride',[1,1],'name','conv2')
             batchNormalizationLayer('name','batchnorm2')
             reluLayer('name','relu2')
             maxPooling2dLayer([1,3],'Stride',1,'Padding','same','name','maxpoo2')
%              convolution2dLayer([1,1],6,'Stride',[1,1],'name','conv3')
%              batchNormalizationLayer('name','batchnorm3')
%              reluLayer('name','relu3')
%              maxPooling2dLayer([1,3],'Stride',1,'Padding','same','name','maxpoo3')
             sequenceUnfoldingLayer('name','unfold')
             flattenLayer('name','flatten')
             gruLayer(numhidden_units1,'Outputmode','sequence','name','hidden1')
             dropoutLayer(0.5,'name','dropout_1')
             gruLayer(numhidden_units2,'Outputmode','last','name','hidden2')
             dropoutLayer(0.5,'name','dropout_2')
%              gruLayer(numhidden_units3,'Outputmode','last','name','hidden3')
%              dropoutLayer(0.5,'name','dropout_3')
%              gruLayer(numhidden_units4,'Outputmode','last','name','hidden4')
%              dropoutLayer(0.5,'name','dropout_4')
             fullyConnectedLayer(4,'name','fc1')   % 全连接层设置（影响输出维度）（cell层出来的输出层） %
             reluLayer('name','relu3')
             fullyConnectedLayer(outputSize,'name','fc2')   % 全连接层设置（影响输出维度）（cell层出来的输出层） %
             tanhLayer('name','sm')
             regressionLayer('name','output')];

            lgraph = layerGraph(layers);
            lgraph = connectLayers(lgraph,'fold/miniBatchSize','unfold/miniBatchSize');

%
% 网络训练

net = trainNetwork(Train_xNorm,Train_yNorm,lgraph,opts);


% 测试
close all

Predict_Ynorm{w,s} = net.predict(Test_xNorm,'MiniBatchSize',1);    % 测试集
YPred = net.predict(Train_xNorm,'MiniBatchSize',1);                % 训练集

% 反归一化
Predict_Y{w,s}  = mapminmax('reverse',Predict_Ynorm{w,s}',yopt);   % 训练集
Predict_Y1{w,s}  = mapminmax('reverse',YPred,yopt);                % 测试集
Predict_Y{w,s}= Predict_Y{w,s}';
Predict_Y1{w,s}= Predict_Y1{w,s}';
%计算均方根误差 (RMSE)。
rmse1(w,s) = sqrt(mean((Predict_Y{w,s}-Test_y').^2));                 % 训练集
rmse2(w,s) = sqrt(mean((Predict_Y1{w,s}-life1').^2));                   % 测试集
%    end
% % end
% 可视化
   plot(Predict_Y1{w,s}) 
   hold on  
   plot(life1)
   
    figure
    plot(smooth(Predict_Y{1,1}))
    hold on
    plot(Test_y) 
   
    subplot(2,1,1)
    plot(Predict_Y/100)
    hold on 
    plot(Test_y/100,'r-')
    hold on
    plot(YPred{1,1}/100,'.-')
    hold on
    plot(YPred1{1,1}/100,'g')
    subplot(2,1,2)
    stem(Predict_Y-Test_y')
    xlabel("日期")
    ylabel("均方根误差")
    title("RMSE = " + rmse1);
%     
   
   

   subplot(2,1,1)
  
    subplot(2,1,2)
    stem(Predict_Y-Test_y')
    xlabel("日期")
    ylabel("均方根误差")
figure
    title("RMSE = " + rmse1);

 as=Predict_Ynorm';
  plot(as{1,1})
    hold on 
    plot(ytestnorm(1,:),'r-');
%    
% clearvars -except   Predict_Y %清除除DATA外的所有变量
% for i=5
% end