Is there a method to predict one point in the future based on multiple point of past prediction? (LSTM univariate)

To explain the title above, how do i get prediction for i+1 based on i and i-1 combined?

I had already done the training and got the training info inside variable net.

This is an example of the code for LSTM trained net to predict 1 value into the future from Mathworks website.

for i=1:numTimeStepsTest
    [net,YPred(:,i)] = predictAndUpdateState(net,YPred(:,i-1));
end

I appreciate if anyone can help direct me to any forum/article regarding this matter if it has been answered previously.

Thank you.

Answer 1

[Aman Banthia]

Hi Revaldo,

I understand that you want to obtain a prediction for the next time step (i+1) based on the previous two time steps (i and i-1) using a trained LSTM network in MATLAB.

To get a prediction for ‘i+1’ based on ‘i’ and ‘i-1’ combined using a trained LSTM network in MATLAB, you can modify the code you provided as follows:

% Assuming you have already trained the LSTM network and have it stored in the variable 'net'

% Initialize variables
numTimeStepsTest = ...; % Specify the number of time steps for testing
YPred = zeros(...); % Initialize the prediction array

% Perform prediction for each time step
for i = 2:numTimeStepsTest
    % Combine inputs for prediction
    combinedInput = [YPred(:, i-1), YPred(:, i-2)]; % Combine previous predictions
 
    % Predict the next time step and update the network state
    [net, YPred(:, i)] = predictAndUpdateState(net, combinedInput);
    
end

Please refer to the following MATLAB File Exchange link for more information:

https://in.mathworks.com/matlabcentral/fileexchange/69506-multivariate-and-univariate-time-series-prediction

Please refer to the following MATLAB Documentation link to know more about LSTM:

https://in.mathworks.com/help/deeplearning/ug/time-series-forecasting-using-deep-learning.html

Hope the above solution helps.

Best Regards,

Aman Banthia