CNN-LSTM Matlab源码，包括原数据

%% 导入数据 addpath('E:\Download\cnn-lstm\2022'); %文件存放路径 opts = detectImportOptions('2022.csv'); preview('2022.csv',opts); for i =8:14 opts.VariableTypes{i}='categorical'; end % opts.SelectedVariableNames = [1:15]; opts.DataLines = [2,17520]; load=readtable('2022.csv',opts); clearvars -except load % 预处理 load.month =dummyvar(load.month); % load.hour=dummyvar(load.hour); load.day_of_week =dummyvar(load.day_of_week); load.weekend =double(load.weekend)-1; load.holiday =double(load.holiday)-1; load.holiday_name =[]; load.day_of_year=[]; load.date=[]; load.year=[]; load.zone=[]; % 数据集划分 corrplot(load(:,2:4)) % 添加特征 (交互影响） load.crossmonth =load.drybulb.*load.month; load.crosshourweekday =load.hour.*load.weekend; load.crossdewpnt = load.dewpnt.*load.month; load.drybulb2 =load.drybulb.^2; load.dewpnt2 =load.dewpnt.^2; load =movevars(load,{'drybulb2','dewpnt2','trend'},'Before','month'); % 预测 ypred =CnnLstm(load); %% CNN+LSTM function ypred =CnnLstm(load) % 划分时间步长 for i =1:size(load,1)-24 x{i,1} = reshape(load{i:24+i-1,2:end}',11,9,1,24); y(i,1) = load.demand(i+24-1); end a =mean(y); b=std(y); y = (y-mean(y))/std(y); n =floor(0.8*17495); xtrain = x(1:n); ytrain =y(1:n); xtest = x(n+1:end); ytest = y(n+1:end); clear i n x y %% CNN+LSTM % 模型结构设置 inputSize = [11 9 1]; filterSize = 6; numFilters = 6; numHiddenUnits = 200; outputsize=1; layers = [ ... sequenceInputLayer(inputSize,'Name','input','Normalization','zerocenter') sequenceFoldingLayer('Name','fold') convolution2dLayer(filterSize,numFilters,'Name','conv') maxPooling2dLayer(3,'Stride',1,'Name','pool') batchNormalizationLayer('Name','bn') reluLayer('Name','relu') sequenceUnfoldingLayer('Name','unfold') flattenLayer('Name','flatten') lstmLayer(100,'OutputMode','sequence','Name','lstm1') lstmLayer(50,'OutputMode','last','Name','lstm2') %concatenationLayer(1,2,'name','concat') fullyConnectedLayer(outputsize, 'Name','fc') regressionLayer('Name','regression')]; lgraph = layerGraph(layers); lgraph = connectLayers(lgraph,'fold/miniBatchSize','unfold/miniBatchSize'); % layer2 = [imageInputLayer([1,2,1],'name','input2')]; % lgraph = addLayers(lgraph,layer2); % lgraph = connectLayers(lgraph,'input2/out','concat/in2'); figure plot(lgraph) analyzeNetwork(lgraph) %% 模型参数设置 opts = trainingOptions('adam', ... 'MaxEpochs',12, ... 'GradientThreshold',1,... 'ExecutionEnvironment','cpu',... 'InitialLearnRate',0.01, ... 'LearnRateSchedule','piecewise', ... 'LearnRateDropPeriod',2, ... %2个epoch后学习率更新 'LearnRateDropFactor',0.9, ... 'ValidationData',{xtest,ytest},... 'Shuffle','every-epoch',... % 时间序列长度 'SequenceLength',24,... 'MiniBatchSize',24*30,... 'Verbose',0, ... 'Plots','training-progress'... ); %net = assembleNetwork(lgraph) %% 训练 tic trainperiod =24; net = trainNetwork(xtrain,ytrain,lgraph,opts); % 滚动预测 toc; %% 预测下一天 ypred =net.predict(xtest(1:24)); plot(ypred) hold on plot(ytest(1:24)) legend('预测','实际') end