【分类模型】详解时间卷积神经网络的时序分类预测模型_matlabtcn代码

目录

MATLAB实现TCN时间卷积神经网络的时序分类预测

MATLAB实现TCN时间卷积神经网络的时序分类预测

以下是采用 MATLAB 实现 TCN 时间卷积神经网络的时序分类预测的代码示例。

首先，我们需要导入相关的库和数据。本示例采用的是 UCI 的电力负载数据集。

% 导入相关库和数据
clear all;
addpath(genpath(pwd))
load('ElectricDevices_TRAIN.mat')
load('ElectricDevices_TEST.mat')
X_train = permute(X,[2 3 1]);
Y_train = Y + 1;
X_test = permute(X,[2 3 1]);
Y_test = Y + 1;

接着，我们定义 TCN 网络模型的结构，并编写相应的训练和测试函数。

%% 定义 TCN 网络结构
function layers = tcnLayers(inputSize,numFilters,filterSize,dilationFactor)
 
inputLayer = sequenceInputLayer(inputSize,'Name','input');
 
layers = [inputLayer];
 
for i=1:numel(numFilters)
    for j=1:numel(filterSize)
        if j == 1 && i == 1
            convLayer = convolution1dLayer(filterSize(j),numFilters(i),'Name',"conv" + int2str(i) + "_" + int2str(j),'Padding',0,"DilationFactor", dilationFactor(j));
        elseif j == numel(filterSize)
            convLayer = convolution1dLayer(filterSize(j),numFilters(i),'Name',"conv" + int2str(i) + "_" + int2str(j),'Padding',0);
        else
            convLayer = convolution1dLayer(filterSize(j),numFilters(i),'Name',"conv" + int2str(i) + "_" + int2str(j),'Padding',0,"DilationFactor", dilationFactor(j));
        end
        layers = [layers; convLayer; batchNormalizationLayer("Name","bn" + int2str(i) + "_" + int2str(j))];             
        layers = [layers; reluLayer("Name","relu" + int2str(i) + "_" + int2str(j))];             
    end
end
 
% global average pooling layer 
gapLayer = averagePooling1dLayer(sum(dilationFactor),"Name","gap");
layers = [layers; gapLayer];
 
% fully connected layer
fcLayer = fullyConnectedLayer(8,'Name','fc');
layers = [layers; fcLayer]; 
 
% softmax layer
softmaxLayer = softmaxLayer('Name','softmax');
layers = [layers; softmaxLayer];
 
% classification layer
classificationLayer = classificationLayer('Name','classification');
layers = [layers; classificationLayer];
 
%% 训练网络
function [net,info] = trainTcnNet(X_train,Y_train,X_val,Y_val,numFilters,filterSize,dilationFactor,maxEpochs,miniBatchSize)
    
    numFeatures = size(X_train,2);
    outputSize = size(Y_train,1);
 
    % define the network architecture
    layers = tcnLayers(numFeatures,numFilters,filterSize,dilationFactor);
 
    % training options
    options = trainingOptions('adam', ...
        'MaxEpochs',maxEpochs, ...
        'MiniBatchSize',miniBatchSize, ...
        'Shuffle','never', ...
        'ValidationData',{X_val,Y_val}, ...
        'ValidationFrequency',10, ...
        'Plots','training-progress');
 
    % train the network
    [net, info] = trainNetwork(X_train,categorical(Y_train),layers,options);
    
end
 
%% 测试网络
function [YPred,acc] = testTcnNet(net,X_test,Y_test)
 
    % predict the labels of test data
    YPred = classify(net,X_test);
    acc = sum(YPred == categorical(Y_test))/numel(Y_test);
    
end

接下来，我们定义并调用训练和测试函数，并输出模型准确率作为最终结果。

% 训练 TCN 网络
numFilters = [16 32 64];
filterSize = [3 3 3];
dilationFactor = [1 2 4];
maxEpochs = 10;
miniBatchSize = 64;
[net,info] = trainTcnNet(X_train,Y_train,X_test,Y_test,numFilters,filterSize,dilationFactor,maxEpochs,miniBatchSize);
 
% 测试 TCN 网络
[YPred,acc] = testTcnNet(net,X_test,Y_test);
 
% 输出模型准确率
fprintf('Accuracy: %.2f%%.\n', acc*100);