Open loop Training performance and closed loop training performance are good but multi-step prediction is bad. Reason?

Dear All,

I am using NARNET to predict multi-step ahead values for temperature data for my laser set operation.

For standardized data (mean = 0 and Var = 1), performance values are following:

OpenloopMSEperformance = 5.451431363178139e-05
OpenloopPERFORMperformance = 5.505768944105006e-05
ClosedloopMSEperformance = 1.043409861314658e-04
ClosedloopPERFORMperformance = 1.053810133096861e-04

However, when I am predicting multi-step ahead values, for next 30 signals, results are really bad.

What can be the reason?

BR, Shashank

[Greg Heath]

Yellow is a poor choice for plots

Please explain your 4-colored plot

Greg

Answer 1

[Greg Heath]

You are not considering information from the autocorrelation fuction.

Hope this helps

Greg