base_predictor_LSTM.m

%% LSTM基预测器

%% 清空环境变量
clc;clear;close all;

%% 训练测试数据
load dataset.mat

trainXn=dataset.trainXn;
trainYn=dataset.trainYn;
testXn=dataset.validXn;
testY=dataset.validY;

outputps=dataset.outputps; 

%% LSTM网络构建
inputSize = size(trainXn,1);
layers = [...
    sequenceInputLayer(inputSize)
    lstmLayer(100,'OutputMode','sequence')
    lstmLayer(100,'OutputMode','last')
    fullyConnectedLayer(100)
    tanhLayer()
    fullyConnectedLayer(1)
    tanhLayer()
    regressionLayer()
    ];
options = trainingOptions('adam',...
    'ExecutionEnvironment', 'cpu',...
    'GradientThreshold',    1,...
    'MaxEpochs',            100,...
    'MiniBatchSize',        4,...
    'InitialLearnRate',     1e-3, ...
    'LearnRateSchedule',    'piecewise', ...
    'LearnRateDropPeriod',  10, ...
    'LearnRateDropFactor',  0.1, ...
    'SequenceLength',       'longest',...
    'Shuffle',              'never',...
    'Plots',                'training-progress',...
    'Verbose',              false,...
    'L2Regularization',     1e-6...
    );

%% LSTM网络训练
net = trainNetwork(con2seq(trainXn)',trainYn',layers,options);

%% LSTM网络预测
testYn_out = predict(net,con2seq(testXn)');
testOutput = mapminmax('reverse',testYn_out',outputps);

%% 误差分析
tValue=testY;
pValue=testOutput;
disp('LSTM');
myerror = myError(tValue,pValue,'show_less');

%% 输出预测结果
xlsName = '.\myResults\results_on_valid.xlsx';
xlswrite(xlsName,tValue',1,'A');
xlswrite(xlsName,pValue',1,'C');

%% 绘图
figure
plot(pValue)
hold on
plot(tValue)
title('测试样本预测结果','fontsize',14)
xlabel('样本','fontsize',14)
ylabel('输出','fontsize',14)
legend('预测值','实际值')
grid on

%% Save variables
% Environments
fullpath = mfilename('fullpath'); 
[~,name]=fileparts(fullpath);
save(['.\myStatus\',name]);
clear fullpath path name;

% Key parameters
lstm = net;
save('.\myModel\lstm','lstm');