【LSTM时间序列预测】深度学习的长短期记忆网络LSTM时间序列预测未来【含Matlab源码 2345期】.zip

clear, clc, close all; data=xlsread('西班牙电价.csv'); data1=data; numTimeStepsTrain = floor(89);%89数据训练 ，3个用来验证 dataTrain = data(1:numTimeStepsTrain+1,:);% 训练样本 dataTest = data(numTimeStepsTrain:end,:); %测试样本 %训练数据标准化处理 mu = mean(dataTrain,'ALL'); sig = std(dataTrain,0,'ALL'); dataTrainStandardized = (dataTrain - mu) / sig; XTrain = dataTrainStandardized(1:end-1,:);% 训练输入 YTrain = dataTrainStandardized(2:end,:);% 训练输出 %% define the Deeper LSTM networks numFeatures= 1;%输入节点 numResponses = 1;%输出节点 numHiddenUnits = 500;%隐含层神经元节点数 %构建 LSTM网络 layers = [sequenceInputLayer(numFeatures) lstmLayer(numHiddenUnits) %lstm函数 dropoutLayer(0.2)%丢弃层概率 reluLayer('name','relu')% 激励函数 RELU fullyConnectedLayer(numResponses) regressionLayer]; XTrain=XTrain'; YTrain=YTrain'; %% LSTM结构参数 options = trainingOptions('adam', ... % adam优化算法 自适应学习率 'MaxEpochs',500,...% 最大迭代次数 'MiniBatchSize',10, ...%最小批处理数量 'GradientThreshold',1, ...%防止梯度爆炸 'InitialLearnRate',0.005, ...% 初始学习率 'LearnRateSchedule','piecewise', ... 'LearnRateDropPeriod',125, ...%125次后 ，学习率下降 'LearnRateDropFactor',0.2, ...%下降因子 0.2 'ValidationData',{XTrain,YTrain}, ... 'ValidationFrequency',5, ...%每五步验证一次 'Verbose',1, ... 'Plots','training-progress'); %% 训练 LSTM网络 net = trainNetwork(XTrain,YTrain,layers,options); %% 测试样本标准化处理 dataTestStandardized = (dataTest - mu) / sig; XTest = dataTestStandardized(1:end-1,:)%测试输入 YTest = dataTest(2:end,:);%测试输出 XTest=XTest'; YTest=YTest'; net = predictAndUpdateState(net,XTrain); YPred = [];%预测结果清零 %% 验证 numTimeStepsTest = numel(XTest); for i = 1:numTimeStepsTest [net,YPred(:,i)] = predictAndUpdateState(net,XTest(:,i),'ExecutionEnvironment','cpu');%仿真预测3天结果 end YPred = sig*YPred + mu;%预测结果去标准化 ，恢复原来的数量级 error=mean(abs((YPred-YTest))./YTest)*100% 误差百分比 %% 预测未来 net1 = resetState(net); net1 = predictAndUpdateState(net1,XTrain); YPred1 = [];%预测结果清零 [net1,YPred1] = predictAndUpdateState(net1,YTrain(end)); numTimeStepsTest1 = 10;% 预测未来数量，从 for i = 2:numTimeStepsTest1 [net1,YPred1(:,i)] = predictAndUpdateState(net1,YPred1(:,i-1),'ExecutionEnvironment','cpu'); end YPred1 = sig*YPred1 + mu; %% 作图 figure(1) plot(dataTrain(1:end-1,1));hold on; idx = (numTimeStepsTrain):(numTimeStepsTrain+numTimeStepsTest1); plot(idx,[(data(numTimeStepsTrain,1))' YPred1(1,:)],'--d'); xlabel("时间/天") ylabel("电价") legend('实际值','预测值') figure(2) subplot(2,1,1) plot(YTest(1,:)) hold on plot(YPred(1,:),'.-') hold off legend(["实际值" "预测值"]) xlabel("时间/天") ylabel("电价") ylim([5000 20000]) subplot(2,1,2) stem(YPred(1,:) - YTest(1,:)) xlabel("时间/天") ylabel("误差") title("误差百分数 = " + error)