How can I use LSTM networks to solve a time series regression problem?

How can I use LSTM networks to solve a time series regression problem? I would like to train a LSTM regression network using all but the last 10 time steps and then test it to see how well the model has done using the last 10 steps of time series that were not used in the training.

Answer 1

[MathWorks Support Team]

To forecast the values of future time steps of a sequence, you can train a sequence-to-sequence regression LSTM network, where the responses are the training sequences with values shifted by one time step. That is, at each time step of the input sequence, the LSTM network learns to predict the value of the next time step.

Please refer to the attached example, "TimeSeriesForecastLSTM.mlx", which demonstrates how to forecast time-series data using a long short-term memory (LSTM) network.

This example trains an LSTM network to forecast the number of chickenpox cases given the number of cases in previous months. The training data contains a single time series, with time steps corresponding to months and values corresponding to the number of cases.

Further, you mentioned that you need to forecast the values for the last 10 steps. To forecast the values of multiple time steps in the future, you can use the "predictAndUpdateState" function to predict time steps one at a time and update the network state at each prediction. Please refer to the documentation of the "predictAndUpdateState" function for more information on how to use the function by typing the following command in the Command Window:

>> doc predictAndUpdateState

[Mohamed Nedal]

Hi, Could you demonstrate how to modify this code to be able to deal with a multivariable problem?

Like, for instance, we have 3 time-series inputs (X1, X2, X3) and we need to:

1. forecast a single time-series target (Y)

2. forecast 2 time-series targets (Y1, Y2)

based on those 3 time-series inputs.

[Viktor Deryabin]

Could you explain, please, how does this LSTM "chikenpox" net actually work? We have a pair of training sequence, i.e. XTrain = data(1:sequenceLength); YTrain = data(2:sequenceLength+1). We want to train the network to predict YTrain(i+1) for given XTrain(i), i=1,2,... That is the network will predict a value for i+1 time moment, if it knows a value for i moment, will not it?

If I want to use d previous values, I should to define XTrain = data(1:sequenceLength-d+1); YTrain = data(d+1:sequenceLength+1). Is it correct?

Generally, how can we define a number of tapped delayes in LSTM?

Thanks in advance

[Dan Hendrickson]

Hi Mohamed and Viktor,

If you have additonal questions about LSTM nets please contact MathWorths technical support and they can assist you further.

best,

Dan

Answer 2

[Abolfazl Nejatian]

here is my code,

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

this piece of code predicts time series data by use of deep learning and shallow learning algorithm.

best wishes

Abolfazl Nejatian

[israt fatema]

how to use this code for multivariate time series data?

Answer 3

[Ana Correia]

I can't seem to use this example with MATLAB r2017b. It says 'adam' is not a valid solver name and when I switch to 'sgdm' i get this error: "Regression is not supported for networks with LSTM layers."

Is this example only supported by r2018a?

Thanks in advance for your help.

Answer 4

[Said Benkaciali]

Dear

I have the honnor to ask you to explain the LSTM model for forecasting a global solar irradiation for one hour ahead forecast or if it is possible send me a matlab code.

Thank you

Answer 5

[Said Benkaciali]

Dear

I have the honnor to ask you to explain the LSTM model for forecasting a global solar irradiation for one hour ahead forecast or if it is possible send me a matlab code.

Thank you

Name: Said

Email: [e-mail]