代码原理
基于TCN（Temporal Convolutional Network）和BiGRU（Bidirectional Gated Recurrent Unit）的方法可以用于处理单输入单输出的数据时序预测任务。TCN是一种适用于时序数据的卷积神经网络结构，能够有效地捕捉时序数据中的长期依赖关系。BiGRU则是一种双向门控循环单元网络，能够同时考虑序列数据的过去和未来信息。以下是基于TCN-BiGRU的数据时序预测的一般步骤：
1. 数据准备：收集和整理单输入单输出的时序数据，确保数据整齐并且可供模型训练和测试。
2. 数据特征提取：根据时序预测的需求，可能需要对输入数据进行适当的特征提取，以便于模型的学习。
3. 数据划分：将数据集划分为训练集、验证集和测试集，以便于模型训练和评估。
4. 搭建TCN-BiGRU模型：构建一个堆叠了TCN和BiGRU层的模型结构，以便于综合利用卷积神经网络和双向循环神经网络的优势。
5. 模型训练：使用训练集数据对TCN-BiGRU模型进行训练，以学习输入数据和输出数据之间的映射关系。
6. 模型调参：调整模型的超参数，如学习率、层的数量、单元的数量等，以获得更好的性能。
7. 模型验证：使用验证集数据对训练好的模型进行验证，评估模型的泛化能力和性能。
8. 模型评估和预测：使用测试集数据对模型进行最终评估，并利用训练好的TCN-BiGRU模型对未来的时序数据进行预测。
TCN-BiGRU的结合将卷积神经网络和双向循环神经网络的特点结合起来，能够更好地处理时序数据，并且能够更好地捕捉数据中的长期依赖关系和时间特性。
部分代码
%% 清空环境变量来自公众号《KYXB》warning off % 关闭报警信息close all % 关闭开启的图窗clear % 清空变量clc%% 导入数据data = readmatrix('风电场预测.xlsx');data = data(5665:8640,12); %选取3月份数据[h1,l1]=data_process(data,24); %步长为24，采用前24个时刻的温度预测第25个时刻的温度res = [h1,l1];num_samples = size(res,1); %样本个数% 训练集和测试集划分outdim = 1; % 最后一列为输出num_size = 0.7; % 训练集占数据集比例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P_train = res(1: num_train_s, 1: f_)';T_train = res(1: num_train_s, f_ + 1: end)';M = size(P_train, 2);P_test = res(num_train_s + 1: end, 1: f_)';T_test = res(num_train_s + 1: end, f_ + 1: end)';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);% 格式转换for i = 1 : Mvp_train{i, 1} = p_train(:, i);vt_train{i, 1} = t_train(:, i);endfor i = 1 : Nvp_test{i, 1} = p_test(:, i);vt_test{i, 1} = t_test(:, i);end%% 优化算法优化前，构建优化前的TCN_BiGRU模型outputSize = 1; %数据输出y的维度numFilters = 64;filterSize = 5;dropoutFactor = 0.005;numBlocks = 2;layer = sequenceInputLayer(f_,Normalization="rescale-symmetric",Name="input");lgraph = layerGraph(layer);outputName = layer.Name;% 来自公众号《KYXB》for i = 1:numBlocksdilationFactor = 2^(i-1);layers = [convolution1dLayer(filterSize,numFilters,DilationFactor=dilationFactor,Padding="causal",Name="conv1_"+i)layerNormalizationLayerdropoutLayer(dropoutFactor)% spatialDropoutLayer(dropoutFactor)convolution1dLayer(filterSize,numFilters,DilationFactor=dilationFactor,Padding="causal")layerNormalizationLayerreluLayerdropoutLayer(dropoutFactor)additionLayer(2,Name="add_"+i)];% Add and connect layers.lgraph = addLayers(lgraph,layers);lgraph = connectLayers(lgraph,outputName,"conv1_"+i);% Skip connection.if i == 1% Include convolution in first skip connection.layer = convolution1dLayer(1,numFilters,Name="convSkip");lgraph = addLayers(lgraph,layer);lgraph = connectLayers(lgraph,outputName,"convSkip");lgraph = connectLayers(lgraph,"convSkip","add_" + i + "/in2");elselgraph = connectLayers(lgraph,outputName,"add_" + i + "/in2");end% Update layer output name.outputName = "add_" + i;endtempLayers = flattenLayer("Name","flatten");lgraph = addLayers(lgraph,tempLayers);tempLayers = gruLayer(35,"Name","gru1");lgraph = addLayers(lgraph,tempLayers);tempLayers = [FlipLayer("flip3")gruLayer(35,"Name","gru2")];lgraph = addLayers(lgraph,tempLayers);tempLayers = [concatenationLayer(1,2,"Name","concat")% selfAttentionLayer(1,50,"Name","selfattention") %Attention机制fullyConnectedLayer(outdim,"Name","fc")regressionLayer("Name","regressionoutput")];lgraph = addLayers(lgraph,tempLayers);% 来自公众号《KYXB》lgraph = connectLayers(lgraph,outputName,"flatten");lgraph = connectLayers(lgraph,"flatten","gru1");lgraph = connectLayers(lgraph,"flatten","flip3");lgraph = connectLayers(lgraph,"gru1","concat/in1");lgraph = connectLayers(lgraph,"gru2","concat/in2");% 参数设置options0 = trainingOptions('adam', ... % 优化算法Adam'MaxEpochs', 150, ... % 最大训练次数'GradientThreshold', 1, ... % 梯度阈值'InitialLearnRate', 0.01, ... % 初始学习率'LearnRateSchedule', 'piecewise', ... % 学习率调整'LearnRateDropPeriod',100, ... % 训练100次后开始调整学习率'LearnRateDropFactor',0.001, ... % 学习率调整因子'L2Regularization', 0.001, ... % 正则化参数'ExecutionEnvironment', 'cpu',... % 训练环境'Verbose', 1, ... % 关闭优化过程'Plots', 'none'); % 画出曲线
代码效果图
