Fitting vs. time-series for a Maglev-like problem?

Dear All, I try to model the behaviour of a solar chiller with a neural network which will be trained by results from a physical model. Basically it is similar to the Maglev example in the documentation: 6 inputs (i.e. ambient temperature, solar radiation, state of the system,...) determine 5 outputs (massflow, cooling power, surface temperature, ...) However, it is not linear -> the cooling happens during nighttime and depends on the state of charge that was achieved during daytime. So I guess a NARX-model needs to be used for this problem.

when I create a Narx net with the NNstart-tool I can plot the response of the system and see how it behaves. FIRST QUESTION: Is there an option to plot the resuts/errors for each output? I.e. feed inputs to the trained net and compare the correct outputs (from physical model) with the outputs calculated by the net? the standard plot for the tool is "Response of output element 1 for time-series 1", but i would like to have the response for output elements 1-5 in separate plots.

SECOND Question: When I use the fitting tool to generate a net for the same purpose as above I get an astonishing low error (R=0.99999) and MSE of 0.002! Which would mean that the net performs perfectly! This is hard to beleive, so I guess I did something wrong... Is the fitting-tool suitable for modelling an above described scenario?

Any help is highly appreciated!

Cheers, and thanks in advance, Philip

[Greg Heath]

What is the difference between the nets in the 2 questions?

How many input/output pairs?

How many input delays?

How many hidden nodes ?

Greg