5-Days Lead forecasting with LSTM

Hello, I have worked on a problem and used LSTM explained following.

https://www.mathworks.com/help/deeplearning/ug/time-series-forecasting-using-deep-learning.html?s_tid=answers_rc2-2_p5_MLT

To be clear with my question let me explain the problem I am working on. I have 200 water level timeseries data of river. And I have successfully run the LTSM code which can predict one step ahead forecasting (i.e. as per my knowledge when testing, if I give todays water level data it will predict the next days value).

But I want to it to predict 5 step ahead (i.e. when testing, if I give todays water level data it will predict the value for 5th day)

Answer 1

[Anay]

Hi Foysol,

I understand that you want to train a LSTM model to predict output 5 steps ahead. The example you are currently referring to shows how to train an LSTM model to predict a single element ahead in the sequence.

You can predict next 5 outputs by utilizing the “Closed loop forecasting” method where in order to predict output for time steps t to t + k, you need to have true value for the time step t - 1 only. Then, to make prediction for step i, use the predicted value for time step i – 1 as input.

numPredictionTimeSteps = 5; %Predict for next 5 time steps 
Y = zeros(numPredictionTimeSteps,numChannels); 
Y(1,:) = Z(end,:); 

for t = 2:numPredictionTimeSteps 
    [Y(t,:),state] = predict(net,Y(t-1,:)); 
    net.State = state; 
end 

You can refer to the “Closed loop forecasting” section of the “Sequence Forecasting Using Deep Learning Example” by following the below link:

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

I hope this resolves the issue!