neural network NARX time series

I have six columns of data (See attached spreadsheet). I am trying to predict chamber pressure from motion responses in six degrees of freedom. As shown on the spreadsheet, I have created a liner regression model that performs better than I expected, but I was to improve this. I have tried the NARX neural net tool and get good results with the open loop, but when I close the loop my results are almost worse than the linear regression model.

I want to be able to predict unknown pressure outputs from new motion data so I cannot use an open loop network as this appears to need a known pressure output.

Any help would be brilliant.

Thanks

Answer 1

[Greg Heath]

It is not unusual for performance to degrade substantially when the loop is closed.

Solutions that that have worked for me

 1. Train the CL solution
 2. Additional training to reduce the OL error
 3. Obtain another OL solution

Hope this helps.

Thank you for formally accepting my answer

Greg