Prediction Time series Data with LSTM Network and multiple Inputs

Hello Together,

i am currently trying to use an LSTM Network to predict Time Series data.

So far the Training has worked out ok and now i am trying to predict data for every Input variable (12). I tried to modify and use the code which was given in the Examples, https://de.mathworks.com/help/deeplearning/ug/time-series-forecasting-using-deep-learning.html but it did not work out.

Is there anybody who has worked with this Type of Network and could help me out ? I am by no means an expert in matlab but have managed to come this far, which ist alread for my standards an achievment.

My goals for my project:

• I want to predict the Temperature and efficiency of electric motors by using an neural Network model in Simulink
• My Simulink model should contain an Neural network Block with an already Trained network based on real Trainingsdata of an electric Motor
• It should be able to predict the 12 Parameters, i will calculate the effiiciency of the Motor of the Predicted data
• It should be able to do the Predictions for any Operating point of the electrical machine
• i want to give the Simulation 3+1 inputs, so that it can calculate the rest itself
• Torque and RPM to determin the operating Point
• Time( time Steps) to determin the Operating Time of each Point
• Initial Tempreature value to let the system know in which condition its starts

I have made a small PDF with my Idea which is also attached.

i will shorty summarize what i did so far in my code that i attached

1. Collecting Data in a test Rig, the longest data Set is 24 hours long
2. Narrowed the input Parameters of the Network to 12 Inputs and 12 Outputs
3. Load in 3 different dat sets
4. Norming the Data with ist own max values
5. Transformed my table in to a matrix
6. created an numeric array with same sequence length of all 3 data sets
7. separated the data into Training and Test data
8. Created predictors and Responses with the Training Data
9. created for each an Datastore
10. Combined the Datastore
11. Define The Network and Training Opts
12. Trained the Network
13. Seperated the Test data in predictors and response
14. Tried to predict something with the predict function
15. tried closed loop prediction with the predict and update state function

I Hope i was bale to summarize what i did. I have no idea how i should keep going to make the Prediction part work.

I would like to give you the Datasets, but they are too big to upload them here. Do you know how i can provide thme to you?

Thank You for your Help!

Answer 1

[Udit]

Hi,

I understand that you want to want to train an LSTM network with 12 inputs given to the model and the trained model should predict 12 outputs. You can train a multi-output LSTM network using a custom training loop. You can refer to the following MathWorks documentations to understand more about how to train a deep learning model having multiple inputs and multiple outputs.

1. https://www.mathworks.com/help/deeplearning/ug/train-network-with-multiple-outputs.html
2. https://www.mathworks.com/help/deeplearning/ug/multiple-input-and-multiple-output-networks.html

I hope this helps.