Multi step ahead forecasting with LSTM

How to perform multi-step ahead forecasting with LSTM. I want to predict 2,3, and 4 time stesp ahead prediction with LSTM?

Please help. Thanks in advance.

[Nikolaos Efkarpidis]

Dear,

How did you finally perform multi-step ahead forecasting with LSTM and predictAndUpdateState ?
I am also struggling with this issue.

Best regards,
Nick 

[Raziye Ghasemi]

How did you finally perform multi-step ahead forecasting with LSTM and predictAndUpdateState ? I am also struggling with this issue.

Answer 1

[adel adel]

Forecasting is basicaly sequence-to-sequence regression, let suppos that your entire sequence is data,

1. You divide data into train and test parts, you can specify the proportion as you wish:

numTimeStepsTrain = floor(0.9*numel(data));% 90% for training 10%for testing
dataTrain = data(1:numTimeStepsTrain+1);
dataTest = data(numTimeStepsTrain+1:end);

2. Preparing training data and response sequences by shifting data by one time step, such as for data(t) the response will be data(t+1)

XTrain = dataTrain(1:end-1);
YTrain = dataTrain(2:end);

3. Preparing the network and training hyperparameters, then train the network using training data and training responses

numFeatures = 1;
numResponses = 1;
numHiddenUnits = 200;
layers = [ ...
    sequenceInputLayer(numFeatures)
    lstmLayer(numHiddenUnits)
    fullyConnectedLayer(numResponses)
    regressionLayer];
options = trainingOptions('adam', ...
    'MaxEpochs',250, ...
    'GradientThreshold',1, ...
    'InitialLearnRate',0.005, ...
    'Verbose',0, ...
    'Plots','training-progress');
net = trainNetwork(XTrain,YTrain,layers,options);

3. Now you can forecast 1, 2, 3 or 4 steps ahead using predictAndUpdateState function, since you use predicted values to update the network state and you don’t use actual values contained in dataTest for this, you can make predictions on any time step number

net = predictAndUpdateState(net,XTrain);
[net,YPred] = predictAndUpdateState(net,YTrain(end));
stepsAhead = 4; % you can use 1,2,3,4 on any value of steps ahead
for i = 2:stepsAhead+1
    [net,YPred(:,i)] = predictAndUpdateState(net,YPred(:,i-1));
end

[anurag kulshrestha]

Thanks. But your code is preciting only 4 steps ahead. I have a test set with 30 data points (lets assume). This code is using only the one predicted value to generate one step ahead predicitons. But, if i want to utilize past 2 or 3 or 4 predicted values to generate one step ahead forecasting like mutli-step? WIll this code work?

[Aniroodh Kumaraswamy]

Very valid question, I too am confused how multiple historical values can be used (say 10 values) to get a forecast into the future. Would appreciate a response from someone who knows

[Gordon Elgin]

I have a question. I know that you can predict multi-time steps. However, how do we input say 5 timesteps to predict 1 time step ahead.

[Kapila]

Any answer for this?

Answer 2

[David Willingham]

Hi,

Please see this updated example in R2022a that shows multi-step ahead (closed loop) predictions:

Time Series Forecasting Using Deep Learning

[Shawn Berry]

Thanks!

[nahed zemouri]

thanks