【LSTM回归预测】基于matlab主成分分析结合LSTM数据回归预测【含Matlab源码 9884期】.zip

clc; clear; close all warning off; data = xlsread('data.xlsx', 'Sheet1', 'A3:M1250'); x=data(:,1:12); y=data(:,13); % 求因子维度 [r,c]=size(x); % 求因子均值 aver=mean(x); % 求因子标准差 stdcov=std(x); % 求目标均值 y_aver=mean(y); % 求目标方差 y_stdcov=std(y); % 数据标准化 stdarr=(x-aver(ones(r,1),:))./stdcov(ones(r,1),:); std_y=(y-y_aver(ones(r,1)))./y_stdcov(ones(r,1)); % 计算相关系数矩阵 covArr=cov(stdarr); %% 相关系数矩阵进行主成分分析 [pc1,latent,explain]=pcacov(covArr); d=repmat(sign(sum(pc1)),size(pc1,1),1); pc=pc1.*d; sumcontr=0; i=1; % 选取贡献率大于90%的主成分 er=90; while sumcontr<er sumcontr=sumcontr+explain(i); i=i+1; end pcNum=i-1; % 计算各个主成分 F=stdarr*pc(:,1:pcNum); FF=[ones(r,1),F]; Fcoeff=regress(std_y,FF); Recoeff=pc(:,1:pcNum)*Fcoeff(2:pcNum+1,:); result(1)=Fcoeff(1); result(2:c+1)=Recoeff; y_p=Fcoeff(1)+stdarr*Recoeff; y_p=y_p.*y_stdcov(ones(r,1))+y_aver(ones(r,1)); %% BiLSTM %% % 输入数据 input =F'; output=y'; nwhole =size(data,1); temp=randperm(nwhole); train_ratio=0.9; ntrain=round(nwhole*train_ratio); ntest =nwhole-ntrain; % 准备输入和输出训练数据 input_train =input(:,temp(1:ntrain)); output_train=output(:,temp(1:ntrain)); % 准备测试数据 input_test =input(:, temp(ntrain+1:ntrain+ntest)); output_test=output(:,temp(ntrain+1:ntrain+ntest)); %% 归一化（全部特征 均归一化） [inputn_train,inputps] =mapminmax(input_train); [outputn_train,outputps]=mapminmax(output_train); inputn_test=mapminmax('apply',input_test,inputps); %% BiLSTM层设置，参数设置 inputSize = size(inputn_train,1); %数据输入x的特征维度 outputSize = size(outputn_train,1); %数据输出y的维度 numhidden_units=240; %% BiLSTM %输入层设、学习层、全连接层 layers = [ ... sequenceInputLayer(inputSize) fullyConnectedLayer(2*inputSize) lstmLayer(numhidden_units) dropoutLayer(0.2) fullyConnectedLayer(outputSize) regressionLayer]; %% trainoption(BiLSTM) %优化算法、训练次数、梯度阈值、运行环境、学习率、学习计划 opts = trainingOptions('adam', ... 'MaxEpochs',800, ... 'MiniBatchSize',48,... 'GradientThreshold',1,... 'ExecutionEnvironment','gpu',... 'InitialLearnRate',0.005, ... 'LearnRateSchedule','piecewise', ... 'LearnRateDropPeriod',100, ... 'LearnRateDropFactor',0.8, ... 'Verbose',0, ... 'Plots','training-progress'... ); %% BiLSTM网络训练 tic BiLSTMnet = trainNetwork(inputn_train,outputn_train,layers,opts); toc; [BiLSTMnet,BiLSTMoutputr_train]= predictAndUpdateState(BiLSTMnet,inputn_train); BiLSTMoutput_train = mapminmax('reverse',BiLSTMoutputr_train,outputps); %网络测试输出 [BiLSTMnet,BiLSTMoutputr_test] = predictAndUpdateState(BiLSTMnet,inputn_test); %网络输出反归一化 BiLSTMoutput_test= mapminmax('reverse',BiLSTMoutputr_test,outputps); %% BiLSTM数据输出 BiLSTMerror_test=BiLSTMoutput_test'-output_test'; BiLSTMpererror_test=BiLSTMerror_test./output_test'; RMSEtest = sqrt(sumsqr(output_test - BiLSTMoutput_test)/length(output_test)); MAPEtest = mean(abs(BiLSTMpererror_test)); %-------------------------------------------------------------------------- disp('PCA-BiLSTM神经网络测试平均绝对误差百分比MAPE'); disp(MAPEtest) disp('PCA-BiLSTM神经网络测试均方根误差RMSE') disp(RMSEtest) %-------------------------------------------------------------------------- %% PCA可视化 %% % 主成分贡献率分析 figure percent_explained = 100*explain / sum(explain); [H,ax] =pareto(percent_explained); set(ax,'linewidth',2,'FontName','Times New Roman') set(H(1),'FaceColor',[180 60 0]./255) set(H(2),'linestyle','--','Marker','s','Color',[255 0 0]./255,'linewidth',2,'Markersize',6,'MarkerFaceColor',[255 0 0]./255) xlabel('主成分') ylabel('贡献率(%)') title('主成分贡献率') h = gca; h.FontName = '黑体'; h.FontSize = 12; %-------------------------------------------------------------------------- figure plot(1:r,y,'k-','linewidth',0.8); hold on plot(1:r,y_p,'r-.','linewidth',1); title('结果比较'); legend('原始数据','协方差主成分回归预测值','Location','Best'); legend box off h = gca; h.FontName = '黑体'; h.FontSize = 10; axis tight %------------------------------------------------------------------------------------- %% PCA-BiLSTM可视化 %------------------------------------------------------------------------------------- figure plot(BiLSTMoutput_test,'-.','Color',[0 0 255]./255,'linewidth',1,'Markersize',3,'MarkerFaceColor',[0 0 255]./255) hold on plot(output_test,'-.','Color',[0 0 0]./255,'linewidth',0.8,'Markersize',3,'MarkerFaceColor',[0 0 0]./255) legend( '测试数据','实际数据','Location','NorthWest','FontName','黑体'); title('PCA-BiLSTM网络模型测试结果及真实值','fontsize',12,'FontName','黑体') xlabel('时间','fontsize',12,'FontName','黑体'); ylabel('数值','fontsize',12,'FontName','黑体'); axis tight %------------------------------------------------------------------------------------- figure plot(BiLSTMpererror_test,'-.','Color',[128 0 255]./255,'linewidth',1) legend('PCA-BiLSTM网络测试相对误差','Location','NorthEast','FontName','黑体') title('PCA-BiLSTM网络模型测试相对误差','fontsize',12,'FontName','黑体') ylabel('误差','fontsize',12,'FontName','黑体') xlabel('样本','fontsize',12,'FontName','黑体') axis tight