コード
全体的なコードシーケンスは次のとおりです。
1、
clc
close all
clear all
%训练数据
data = xlsread('load data3.xlsx',1,'A2:G2001');
input = data(:,1:6);
output = data(:,7);
numTimeStepsTrain = floor(0.7*numel(data(:,1)));
XTrain = input(1:numTimeStepsTrain,:);
YTrain = output(1:numTimeStepsTrain,:);
XTest = input(numTimeStepsTrain+1:end,:);
YTest = output(numTimeStepsTrain+1:end,:);
x = XTrain;
y = YTrain;
[xnorm,xopt] = mapminmax(x',0,1);
[ynorm,yopt] = mapminmax(y',0,1);
x = x';
% 转换成2-D image
for i = 1:length(ynorm)
Train_xNorm{
i} = reshape(xnorm(:,i),6,1,1);
Train_yNorm(:,i) = ynorm(:,i);
Train_y(i,:) = y(i,:);
end
Train_yNorm= Train_yNorm';
xtest = XTest;
ytest = YTest;
[xtestnorm] = mapminmax('apply', xtest',xopt);
[ytestnorm] = mapminmax('apply',ytest',yopt);
xtest = xtest';
for i = 1:length(ytestnorm)
Test_xNorm{
i} = reshape(xtestnorm(:,i),6,1,1);
Test_yNorm(:,i) = ytestnorm(:,i);
Test_y(i,:) = ytest(i,:);
end
Test_yNorm = Test_yNorm';
%% SSA优化参数设置
SearchAgents = 10; % 种群数量 50
Max_iterations = 30 ; % 迭代次数 10
lowerbound = [1e-10 0.001 10 ];%三个参数的下限
upperbound = [1e-2 0.01 200 ];%三个参数的上限
dimension = 3;%数量，即要优化的LSTM参数个数
%% SSA优化LSTM
% [Best_pos,Best_score,Convergence_curve]=SSA(SearchAgents,Max_iterations,lowerbound,upperbound,dimension,ObjFcn)
%% 参数设置
ST = 0.8; % 预警值
PD = 0.2; % 发现者的比列，剩下的是加入者
PDNumber = SearchAgents * PD; % 发现者数量
SDNumber = SearchAgents - SearchAgents * PD; % 意识到有危险麻雀数量
%% 判断优化参数个数
if(max(size(upperbound)) == 1)
upperbound = upperbound .* ones(1, dimension);
lowerbound = lowerbound .* ones(1, dimension);
end
%% 种群初始化
pop_lsat = initialization(SearchAgents, dimension, upperbound, lowerbound);
pop_new = pop_lsat;
%% 计算初始适应度值
fitness = zeros(1, SearchAgents);
for i = 1 : SearchAgents
fitness(i) = fun(pop_new(i,:),Train_xNorm,Train_yNorm,Test_xNorm,Test_y,yopt);
end
%% 得到全局最优适应度值
[fitness, index]= sort(fitness);
GBestF = fitness(1);
%% 得到全局最优种群
for i = 1 : SearchAgents
pop_new(i, :) = pop_lsat(index(i), :);
end
GBestX = pop_new(1, :);
X_new = pop_new;
%% 优化算法
for i = 1: Max_iterations
BestF = fitness(1);
R2 = rand(1);
for j = 1 : PDNumber
if(R2 < ST)
X_new(j, :) = pop_new(j, :) .* exp(-j / (rand(1) * Max_iterations));
else
X_new(j, :) = pop_new(j, :) + randn() * ones(1, dimension);
end
end
for j = PDNumber + 1 : SearchAgents
if(j > (SearchAgents - PDNumber) / 2 + PDNumber)
X_new(j, :) = randn() .* exp((pop_new(end, :) - pop_new(j, :)) / j^2);
else
A = ones(1, dimension);
for a = 1 : dimension
if(rand() > 0.5)
A(a) = -1;
end
end
AA = A' / (A * A');
X_new(j, :) = pop_new(1, :) + abs(pop_new(j, :) - pop_new(1, :)) .* AA';
end
end
Temp = randperm(SearchAgents);
SDchooseIndex = Temp(1 : SDNumber);
for j = 1 : SDNumber
if(fitness(SDchooseIndex(j)) > BestF)
X_new(SDchooseIndex(j), :) = pop_new(1, :) + randn() .* abs(pop_new(SDchooseIndex(j), :) - pop_new(1, :));
elseif(fitness(SDchooseIndex(j)) == BestF)
K = 2 * rand() -1;
X_new(SDchooseIndex(j), :) = pop_new(SDchooseIndex(j), :) + K .* (abs(pop_new(SDchooseIndex(j), :) - ...
pop_new(end, :)) ./ (fitness(SDchooseIndex(j)) - fitness(end) + 10^-8));
end
end
%% 边界控制
for j = 1 : SearchAgents
for a = 1 : dimension
if(X_new(j, a) > upperbound(a))
X_new(j, a) = upperbound(a);
end
if(X_new(j, a) < lowerbound(a))
X_new(j, a) = lowerbound(a);
end
end
end
%% 获取适应度值
for j = 1 : SearchAgents
fitness_new(j) = fun(X_new(j, :),Train_xNorm,Train_yNorm,Test_xNorm,Test_y,yopt);
end
%% 获取最优种群
for j = 1 : SearchAgents
if(fitness_new(j) < GBestF)
GBestF = fitness_new(j);
GBestX = X_new(j, :);
end
end
%% 更新种群和适应度值
pop_new = X_new;
fitness = fitness_new;
%% 更新种群
[fitness, index] = sort(fitness);
for j = 1 : SearchAgents
pop_new(j, :) = pop_new(index(j), :);
end
%% 得到优化曲线
curve(i) = GBestF;
avcurve(i) = sum(curve) / length(curve);
end
%% 得到最优值
Best_pos = GBestX;
Best_score = curve(end);
%% 得到最优参数
NumOfUnits =abs(round( Best_pos(1,3))); % 最佳神经元个数
InitialLearnRate = Best_pos(1,2) ;% 最佳初始学习率
L2Regularization = Best_pos(1,1); % 最佳L2正则化系数
%
inputSize = 6;
outputSize = 1; %数据输出y的维度
layers = [ ...
sequenceInputLayer([inputSize,1,1],'name','input') %输入层设置
sequenceFoldingLayer('name','fold')
convolution2dLayer([2,1],10,'Stride',[1,1],'name','conv1')
batchNormalizationLayer('name','batchnorm1')
reluLayer('name','relu1')
convolution2dLayer([1,1],10,'Stride',[1,1],'name','conv2')
batchNormalizationLayer('name','batchnorm2')
reluLayer('name','relu2')
maxPooling2dLayer([1,3],'Stride',1,'Padding','same','name','maxpool')
sequenceUnfoldingLayer('name','unfold')
flattenLayer('name','flatten')
lstmLayer(NumOfUnits ,'Outputmode','sequence','name','hidden1')
dropoutLayer(0.3,'name','dropout_1')
lstmLayer(NumOfUnits ,'Outputmode','sequence','name','hidden3')
dropoutLayer(0.3,'name','dropout_3')
lstmLayer(NumOfUnits ,'Outputmode','last','name','hidden2')
dropoutLayer(0.3,'name','drdiopout_2')
fullyConnectedLayer(outputSize,'name','fullconnect') % 全连接层设置（影响输出维度）（cell层出来的输出层） %
tanhLayer('name','softmax')
regressionLayer('name','output')];
lgraph = layerGraph(layers)
lgraph = connectLayers(lgraph,'fold/miniBatchSize','unfold/miniBatchSize');
% 参数设置
% options = trainingOptions('adam', ... % 优化算法Adam
% 'MaxEpochs', 2000, ... % 最大训练次数
% 'GradientThreshold', 1, ... % 梯度阈值
% 'InitialLearnRate', InitialLearnRate, ... % 初始学习率
% 'LearnRateSchedule', 'piecewise', ... % 学习率调整
% 'LearnRateDropPeriod', 850, ... % 训练850次后开始调整学习率
% 'LearnRateDropFactor',0.2, ... % 学习率调整因子
% 'L2Regularization', L2Regularization, ... % 正则化参数
% 'ExecutionEnvironment', 'cpu',... % 训练环境
% 'Verbose', 0, ... % 关闭优化过程
% 'Plots', 'training-progress'); % 画出曲线
opts = trainingOptions('adam', ...
'MaxEpochs',500, ...
'GradientThreshold',1,...
'ExecutionEnvironment','cpu',...
'InitialLearnRate',InitialLearnRate, ...
'LearnRateSchedule','piecewise', ...
'LearnRateDropPeriod',100, ... %2个epoch后学习率更新
'LearnRateDropFactor',0.5, ...
'Shuffle','once',... % 时间序列长度
'SequenceLength',1,...
'MiniBatchSize',128,...
'Verbose',1,...
'Plots','training-progress');
% 网络训练
tic
net = trainNetwork(Train_xNorm,Train_yNorm,lgraph,opts );
Predict_Ynorm_Train = net.predict(Train_xNorm);
Predict_Y_Train = mapminmax('reverse',Predict_Ynorm_Train',yopt);
Predict_Y_Train = Predict_Y_Train';
figure
hold on
plot(Predict_Y_Train,'r-','LineWidth',2.0)
plot(Train_y,'b-','LineWidth',2.0);
ylabel('kW')
legend('预测值','实际值')
title('训练')
Predict_Ynorm = net.predict(Test_xNorm);
Predict_Y = mapminmax('reverse',Predict_Ynorm',yopt);
Predict_Y = Predict_Y';
figure
hold on
plot(Predict_Y,'r-','LineWidth',2.0)
plot(Test_y,'b-','LineWidth',2.0)
legend('预测值','实际值')
ylabel('kW')
title('测试')
% 预测结果评价
ae= abs(Predict_Y - Test_y);
rmse = (mean(ae.^2)).^0.5;
mse = mean(ae.^2);
mae = mean(ae);
mape = mean(ae./Predict_Y);
disp('机组预测结果评价指标：')
disp(['RMSE = ', num2str(rmse)])
disp(['MSE = ', num2str(mse)])
disp(['MAE = ', num2str(mae)])
disp(['MAPE = ', num2str(mape)])
disp(['最佳神经元个数为：', num2str(NumOfUnits)])
disp(['最佳初始学习率为：', num2str(InitialLearnRate)])
disp(['最佳L2正则化系数为：', num2str(L2Regularization)])
2、
function Positions = initialization(SearchAgents_no, dim, ub, lb)
%% 初始化
%% 待优化参数个数
Boundary_no = size(ub, 2);
%% 若待优化参数个数为1
if Boundary_no == 1
Positions = rand(SearchAgents_no, dim) .* (ub - lb) + lb;
end
%% 如果存在多个输入边界个数
if Boundary_no > 1
for i = 1 : dim
ub_i = ub(i);
lb_i = lb(i);
Positions(:, i) = rand(SearchAgents_no, 1) .* (ub_i - lb_i) + lb_i;
end
end
3、
function y = fun(x,Train_xNorm,Train_yNorm,Test_xNorm,Test_y,yopt)
%函数用于计算粒子适应度值
%rng default;%固定随机数
numhidden_units1 = fix(x(3))+1; % 隐含层神经元数量 round为四舍五入函数；
numhidden_units2= fix(x(3))+1;
% 层设置，参数设置
inputSize = 6;
outputSize = 1; %数据输出y的维度
options = trainingOptions('adam', ...
'MaxEpochs',100, ...
'GradientThreshold',1,...
'ExecutionEnvironment','cpu',...
'InitialLearnRate',x(2), ...
'LearnRateSchedule','piecewise', ...
'LearnRateDropPeriod',100, ... %100个epoch后学习率更新
'LearnRateDropFactor',0.5, ...
'Shuffle','once',... % 时间序列长度
'SequenceLength',1,...
'MiniBatchSize',128,...
'L2Regularization', x(1), ... '
'Verbose',1);
%% lstm
layers = [ ...
sequenceInputLayer([inputSize,1,1],'name','input') %输入层设置
sequenceFoldingLayer('name','fold')
convolution2dLayer([2,1],10,'Stride',[1,1],'name','conv1')
batchNormalizationLayer('name','batchnorm1')
reluLayer('name','relu1')
convolution2dLayer([1,1],10,'Stride',[1,1],'name','conv2')
batchNormalizationLayer('name','batchnorm2')
reluLayer('name','relu2')
maxPooling2dLayer([1,3],'Stride',1,'Padding','same','name','maxpool')
sequenceUnfoldingLayer('name','unfold')
flattenLayer('name','flatten')
lstmLayer(numhidden_units1,'Outputmode','sequence','name','hidden1')
dropoutLayer(0.3,'name','dropout_1')
lstmLayer(numhidden_units1,'Outputmode','sequence','name','hidden3')
dropoutLayer(0.3,'name','dropout_3')
lstmLayer(numhidden_units2,'Outputmode','last','name','hidden2')
dropoutLayer(0.3,'name','drdiopout_2')
fullyConnectedLayer(outputSize,'name','fullconnect') % 全连接层设置（影响输出维度）（cell层出来的输出层） %
tanhLayer('name','softmax')
regressionLayer('name','output')];
lgraph = layerGraph(layers)
lgraph = connectLayers(lgraph,'fold/miniBatchSize','unfold/miniBatchSize');
%
% 网络训练
net = trainNetwork(Train_xNorm,Train_yNorm,lgraph,options);
Predict_Ynorm = net.predict(Test_xNorm);
Predict_Y = mapminmax('reverse',Predict_Ynorm',yopt);
Predict_Y = Predict_Y';
% rmse_without_update1 = sqrt(mean((Predict_Y(1,:)-Test_y(1,:))).^2,'ALL')
rmse_without_update1 = sqrt(mean((Predict_Y(1,:)-(Test_y(1,:))).^2,'ALL'))
y = rmse_without_update1 ;% cost为目标函数 ，目标函数为rmse
end
4、
clc
close all
clear all
data = xlsread('load data3.xlsx',1,'A2:G2001');
input = data(:,1:6);
output = data(:,7);
inputSize = 6;
outputSize = 1;
numTimeStepsTrain = floor(0.7*numel(data(:,1)));
XTrain = input(1:numTimeStepsTrain,:);
YTrain = output(1:numTimeStepsTrain,:);
XTest = input(numTimeStepsTrain+1:end,:);
YTest = output(numTimeStepsTrain+1:end,:);
x = XTrain;
y = YTrain;
[xnorm,xopt] = mapminmax(x',0,1);
[ynorm,yopt] = mapminmax(y',0,1);
x = x';
for i = 1:length(ynorm)
Train_xNorm{
i} = reshape(xnorm(:,i),inputSize,1,1);
Train_yNorm(:,i) = ynorm(:,i);
Train_y(i,:) = y(i,:);
end
Train_yNorm= Train_yNorm';
xtest = XTest;
ytest = YTest;
[xtestnorm] = mapminmax('apply', xtest',xopt);
[ytestnorm] = mapminmax('apply',ytest',yopt);
xtest = xtest';
for i = 1:length(ytestnorm)
Test_xNorm{
i} = reshape(xtestnorm(:,i),inputSize,1,1);
Test_yNorm(:,i) = ytestnorm(:,i);
Test_y(i,:) = ytest(i,:);
end
Test_yNorm = Test_yNorm';
NumOfUnits =100;
InitialLearnRate = 0.005;
%
layers = [ ...
sequenceInputLayer([inputSize,1,1],'name','input') %
sequenceFoldingLayer('name','fold')
convolution2dLayer([2,1],10,'Stride',[1,1],'name','conv1')
batchNormalizationLayer('name','batchnorm1')
reluLayer('name','relu1')
convolution2dLayer([1,1],10,'Stride',[1,1],'name','conv2')
batchNormalizationLayer('name','batchnorm2')
reluLayer('name','relu2')
maxPooling2dLayer([1,3],'Stride',1,'Padding','same','name','maxpool')
sequenceUnfoldingLayer('name','unfold')
flattenLayer('name','flatten')
lstmLayer(NumOfUnits ,'Outputmode','sequence','name','hidden1')
dropoutLayer(0.3,'name','dropout_1')
lstmLayer(NumOfUnits ,'Outputmode','sequence','name','hidden3')
dropoutLayer(0.3,'name','dropout_3')
lstmLayer(NumOfUnits ,'Outputmode','last','name','hidden2')
dropoutLayer(0.3,'name','drdiopout_2')
fullyConnectedLayer(outputSize,'name','fullconnect') % 鍏ㄨ繛鎺ュ眰璁剧疆锛堝奖鍝嶈緭鍑虹淮搴︼級锛坈ell灞傚嚭鏉ョ殑杈撳嚭灞傦級 %
tanhLayer('name','softmax')
regressionLayer('name','output')];
lgraph = layerGraph(layers)
lgraph = connectLayers(lgraph,'fold/miniBatchSize','unfold/miniBatchSize');
opts = trainingOptions('adam', ...
'MaxEpochs',500, ...
'GradientThreshold',1,...
'ExecutionEnvironment','cpu',...
'InitialLearnRate',InitialLearnRate, ...
'LearnRateSchedule','piecewise', ...
'LearnRateDropPeriod',100, ... %
'LearnRateDropFactor',0.5, ...
'Shuffle','once',... %
'SequenceLength',1,...
'MiniBatchSize',128,...
'Verbose',1,...
'Plots','training-progress');
net = trainNetwork(Train_xNorm,Train_yNorm,lgraph,opts );
Predict_Ynorm_Train = net.predict(Train_xNorm);
Predict_Y_Train = mapminmax('reverse',Predict_Ynorm_Train',yopt);
Predict_Y_Train = Predict_Y_Train';
figure
hold on
plot(Predict_Y_Train,'r-','LineWidth',2.0)
plot(Train_y,'b-','LineWidth',2.0);
ylabel('kW')
legend('预测值?','实际值')
title('训练')
Predict_Ynorm = net.predict(Test_xNorm);
Predict_Y = mapminmax('reverse',Predict_Ynorm',yopt);
Predict_Y = Predict_Y';
figure
hold on
plot(Predict_Y,'r-','LineWidth',2.0)
plot(Test_y,'b-','LineWidth',2.0)
legend('预测值','实际值')
ylabel('kW')
title('测试')
ae= abs(Predict_Y - Test_y);
rmse = (mean(ae.^2)).^0.5;
mse = mean(ae.^2);
mae = mean(ae);
mape = mean(ae./Predict_Y);
disp('误差评价指标?')
disp(['RMSE = ', num2str(rmse)])
disp(['MSE = ', num2str(mse)])
disp(['MAE = ', num2str(mae)])
disp(['MAPE = ', num2str(mape)])
データ
結果
コードとデータが必要な場合は、コメント欄にメールを送信してください。通常は 1 日以内に返信されます。いいね + 注目してください。ありがとうございます。！