clear
clc
close all
%% 数据预处理
% 数据读取
data = xlsread('数据1.xlsx','B:B');
figure;
plot(data);
title('原始序列');
ylabel('数据');
xlabel('样本数');
set(gcf,'color','w')
% 时间窗口划分
f_ = 10;
data_trans=add_timestep(data,f_);
input=data_trans(:,1:end-1);
output=data_trans(:,end);
% 数据集划分
m=round(size(data_trans,1)*0.7);
P_train=input(1:m,:)';
T_train=output(1:m,:)';
P_test=input(m+1:end,:)';
T_test=output(m+1:end,:)';
N = size(P_test, 2); % 测试集样本数
M = size(P_train, 2); % 训练集样本数
outdim= 1; % 输出的维度
% 数据归一化
[P_train, ps_input] = mapminmax(P_train, 0, 1);
P_test = mapminmax('apply', P_test, ps_input);
[t_train, ps_output] = mapminmax(T_train, 0, 1);
t_test = mapminmax('apply', T_test, ps_output);
% 数据平铺
P_train = double(reshape(P_train, f_, 1, M));
P_test = double(reshape(P_test , f_, 1, N));
t_train = t_train';
t_test = t_test' ;
% 数据格式转换
for ii = 1 : M
p_train{ii, 1} = P_train(:, 1, ii);
end
for ii = 1 : N
p_test{ii, 1} = P_test( :, 1, ii);
end
%% 构建网络并训练
% 网络构建
numChannels = 1;
maxPosition = 256;
numHeads = 8;
numKeyChannels = numHeads*32;
layers = [
sequenceInputLayer(10,Name="input")
convolution1dLayer(3,32, Name="Conv1_1", Padding="causal")
reluLayer
positionEmbeddingLayer(numChannels,maxPosition,Name="pos-emb");
additionLayer(2, Name="add")
selfAttentionLayer(numHeads,numKeyChannels,'AttentionMask','causal')
selfAttentionLayer(numHeads,numKeyChannels)
indexing1dLayer("last")
fullyConnectedLayer(1)
regressionLayer];
lgraph = layerGraph(layers);
lgraph = connectLayers(lgraph,"relu","add/in2");
% 参数设置
maxEpochs = 100;
miniBatchSize = 64;
learningRate = 0.001;
solver = 'adam';
shuffle = 'every-epoch';
gradientThreshold = 10;
executionEnvironment = "auto"; % chooses local GPU if available, otherwise CPU
options = trainingOptions(solver, ...
'Plots','training-progress', ...
'MaxEpochs', maxEpochs, ...
'MiniBatchSize', miniBatchSize, ...
'Shuffle', shuffle, ...
'InitialLearnRate', learningRate, ...
'GradientThreshold', gradientThreshold, ...
'ExecutionEnvironment', executionEnvironment);
% 开始训练
net = trainNetwork(p_train, t_train, lgraph, options);
% 数据预测
t_sim1 = predict(net, p_train);
t_sim2 = predict(net, p_test);
% 数据反归一化
T_sim1_imf_temp = mapminmax('reverse', t_sim1, ps_output);
T_sim2_imf_temp = mapminmax('reverse', t_sim2, ps_output);
% 数据格式转换
T_sim1 = double(T_sim1_imf_temp');
T_sim2 = double(T_sim2_imf_temp');
% 划分训练集测试集以供对比
data_trans=add_timestep(data,f_);
input=data_trans(:,1:end-1);
output=data_trans(:,end);
m=round(size(data_trans,1)*0.7);
P_train=input(1:m,:)';
T_train=output(1:m,:)';
P_test=input(m+1:end,:)';
T_test=output(m+1:end,:)';
%% 效果评估
figure
plot(1: M, T_train, 'r-', 1: M, T_sim1, 'g-*', 'LineWidth', 1)
legend('真实值','预测值')
xlabel('样本数')
ylabel('数据')
title('训练集预测结果对比')
xlim([1, M])
grid off
set(gcf,'color','w')
figure
plot(1: N, T_test, 'r-', 1: N, T_sim2, 'g-*', 'LineWidth', 1.2)
legend('真实值','预测值')
xlabel('样本数')
ylabel('数据')
title('测试集预测结果对比')
xlim([1, N])
grid off
set(gcf,'color','w')
% 测试集结果
figure;
plotregression(T_test,T_sim2,'回归图');
set(gcf,'color','w')
% 相关指标计算
disp('transformer训练集数据误差：')
[mae_train,mse_train,rmse_train,mape_train,error_train,errorPercent_train,R_train]=cal_metrics(T_train,T_sim1);
disp('transformer测试集数据误差：')
[mae_test,mse_test,rmse_test,mape_test,error_test,errorPercent_test,R_test]=cal_metrics(T_test,T_sim2);
