POA-CNN-LSTM-Attention鹈鹕优化卷积长短期记忆神经网络注意力机制多变量回归预测（Matlab完整源码和数据)

%% 清空环境变量 warning off % 关闭报警信息 close all % 关闭开启的图窗 clear % 清空变量 clc % 清空命令行 %% 导入数据 res=xlsread('data.xlsx'); %% 数据分析 num_size = 0.7; % 训练集占数据集比例 outdim = 1; % 最后一列为输出 num_samples = size(res, 1); % 样本个数 res = res(randperm(num_samples), :); % 打乱数据集（不希望打乱时，注释该行） num_train_s = round(num_size * num_samples); % 训练集样本个数 f_ = size(res, 2) - outdim; % 输入特征维度 %% 划分训练集和测试集 P_train = res(1: num_train_s, 1: f_)'; T_train = res(1: num_train_s, f_ + 1: end)'; P_test = res(num_train_s + 1: end, 1: f_)'; T_test = res(num_train_s + 1: end, f_ + 1: end)'; %% 划分训练集和测试集 M = size(P_train, 2); N = size(P_test, 2); %% 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 % 将数据平铺成1维数据只是一种处理方式 % 也可以平铺成2维数据，以及3维数据，需要修改对应模型结构 % 但是应该始终和输入层数据结构保持一致 P_train = double(reshape(P_train, f_, 1, 1, M)); P_test = double(reshape(P_test , f_, 1, 1, N)); t_train = double(t_train)'; t_test = double(t_test )'; %% 数据格式转换 for i = 1 : M p_train{i, 1} = P_train(:, :, 1, i); end for i = 1 : N p_test{i, 1} = P_test( :, :, 1, i); end %% 优化算法参数设置 SearchAgents_no = 5; % 数量 Max_iteration = 10; % 最大迭代次数 dim = 3; % 优化参数个数 lb = [1e-3,50 1e-3]; % 参数取值下界(学习率，隐藏层节点，正则化系数) ub = [1e-2,60,1e-2]; % 参数取值上界(学习率，隐藏层节点，正则化系数) fitness = @(x)fical(x,f_); [Best_score,Best_pos,curve]=POA(SearchAgents_no,Max_iteration,lb ,ub,dim,fitness); Best_pos(1, 2) = round(Best_pos(1, 2)); best_hd = Best_pos(1, 2); % 最佳隐藏层节点数 best_lr= Best_pos(1, 1);% 最佳初始学习率 best_l2 = Best_pos(1, 3);% 最佳L2正则化系数 disp("最佳参数组合") disp("学习率，隐藏层节点，正则化系数") disp([best_lr best_hd best_l2]) %% 创建混合CNN-LSTM-SE-Attention网络架构 FiltZise =4; lgraph = layerGraph(); % 建立空白网络结构 tempLayers = [ sequenceInputLayer([f_, 1, 1], "Name", "sequence") % 建立输入层，输入数据结构 sequenceFoldingLayer("Name", "seqfold")]; % 建立序列折叠层 lgraph = addLayers(lgraph, tempLayers); % 将上述网络结构加入空白结构中 tempLayers = convolution2dLayer([3, 1], 32, "Name", "conv_1"); % 卷积层 卷积核[3, 1] 步长[1, 1] 通道数 32 lgraph = addLayers(lgraph,tempLayers); % 将上述网络结构加入空白结构中 tempLayers = [ reluLayer("Name", "relu_1") % 激活层 convolution2dLayer([3, 1], 64, "Name", "conv_2") % 卷积层 卷积核[3, 1] 步长[1, 1] 通道数 64 reluLayer("Name", "relu_2")]; % 激活层 lgraph = addLayers(lgraph, tempLayers); % 将上述网络结构加入空白结构中 tempLayers = [ globalAveragePooling2dLayer("Name", "gapool") % 全局平均池化层 fullyConnectedLayer(16, "Name", "fc_2") % SE注意力机制，通道数的1 / 4 reluLayer("Name", "relu_3") % 激活层 fullyConnectedLayer(64, "Name", "fc_3") % SE注意力机制，数目和通道数相同 sigmoidLayer("Name", "sigmoid")]; % 激活层 lgraph = addLayers(lgraph, tempLayers); % 将上述网络结构加入空白结构中 tempLayers = multiplicationLayer(2, "Name", "multiplication"); % 点乘的注意力 lgraph = addLayers(lgraph, tempLayers); % 将上述网络结构加入空白结构中 numHiddenUnits = best_hd; % 隐含层神经元个数 tempLayers = [ sequenceUnfoldingLayer("Name", "sequnfold") % 建立序列反折叠层 flattenLayer("Name", "flatten") % 网络铺平层 lstmLayer(numHiddenUnits, "Name", "lstm", "OutputMode", "last") % LSTM层 fullyConnectedLayer(1, "Name", "fc") % 全连接层 regressionLayer("Name", "regressionoutput")]; % 回归层 lgraph = addLayers(lgraph, tempLayers); % 将上述网络结构加入空白结构中 lgraph = connectLayers(lgraph, "seqfold/out", "conv_1"); % 折叠层输出 连接 卷积层输入; lgraph = connectLayers(lgraph, "seqfold/miniBatchSize", "sequnfold/miniBatchSize"); % 折叠层输出 连接 反折叠层输入 lgraph = connectLayers(lgraph, "conv_1", "relu_1"); % 卷积层输出 链接 激活层 lgraph = connectLayers(lgraph, "conv_1", "gapool"); % 卷积层输出 链接 全局平均池化 lgraph = connectLayers(lgraph, "relu_2", "multiplication/in2"); % 激活层输出 链接 相乘层 lgraph = connectLayers(lgraph, "sigmoid", "multiplication/in1"); % 全连接输出 链接 相乘层 lgraph = connectLayers(lgraph, "multiplication", "sequnfold/in"); % 点乘输出 %% 参数设置 options = trainingOptions('adam', ... % Adam 梯度下降算法 'MaxEpochs', 200,... % 最大训练次数 'MiniBatchSize',32,... % 批处理 'InitialLearnRate',best_lr,... % 学习率 'L2Regularization', best_l2,... % L2正则化参数 'LearnRateSchedule', 'piecewise',... % 学习率下降 'LearnRateDropFactor', 0.5,... % 学习率下降因子 0.5 'LearnRateDropPeriod', 50,... % 经过训练后 学习率为*0.1 'Shuffle', 'every-epoch',... % 每次训练打乱数据集 'ValidationPatience', Inf,... % 关闭验证 'Plots', 'training-progress',... % 画出曲线 'Verbose', 1); %% 训练 [net,traininfo] = trainNetwork(p_train, t_train, lgraph, options); %% 预测 t_sim1 = predict(net, p_train); t_sim2 = predict(net, p_test ); %% 数据反归一化 T_sim1 = mapminmax('reverse', t_sim1', ps_output); T_sim2 = mapminmax('reverse', t_sim2', ps_output); T_sim1 = double(T_sim1) ; T_sim2 = double(T_sim2) ; %% 绘图 figure plot(1 : length(curve), curve,'linewidth',1.5); title('POA-CNN-LSTM-Attention', 'FontSize', 10); xlabel('迭代次数', 'FontSize', 10); ylabel('适应度值', 'FontSize', 10); grid off %% 均方根误差 RMSE error1 = sqrt(sum((T_sim1 - T_train).^2)./M); error2 = sqrt(sum((T_test - T_sim2).^2)./N); %% 显示网络结构 analyzeNetwork(net) %% 绘图 figure plot(lgraph) title("CNN-LSTM-Attention模型结构") %% 绘图 figure plot(T_sim1,'-s','Color',[255 0 0]./255,'linewidth',1,'Markersize',5,'MarkerFaceColor',[250 0 0]./255) hold on plot(T_train,'-o','Color',[150 150 150]./255,'linewidth',0.8,'Markersize',4,'MarkerFaceColor',[150 150 150]./255) legend( 'POA-CNN-LSTM-Attention拟合训练数据','实际分析数据','Location','NorthWest','FontName','华文宋体'); title('POA-CNN-LSTM-Attention模型预测结果及真实值','fontsize',12,'FontName','华文宋体') xlabel('样本','fontsize',12,'FontName','华文宋体'); ylabel('数值','fontsize',12,'FontName','华文宋体'); xlim([1 M]) %------------------------------------------------------------------------------------- figure bar((T_sim1 - T_train)./T_train) legend('POA-CNN-LSTM-Attention模型训练集相对误差','Location','NorthEast','FontName','华文宋体') title('POA-CNN-LSTM-Attention模型训练集相对误差','fontsize',12,'FontName','华文宋体') ylabel('误差','fontsize',12,'FontName','华文宋体') xlabel('样本','fontsize',12,'FontName','华文宋体') xlim([1 M