Using Data to determine input parameters

I am currently trying to run a computer experiment where I run a monte carlo of the time response of a lump mass heating up where I vary all of the different parameters in the model ( mass, heat flux, emissivity etc.). I then want to use this data to develop a model of the system using a neural network or another regression technique. Finally I then want to take experimental data and using the model of the system determine what the mass of the lump mass is.

My question is in the methodology of how to complete this problem. Could anyone propose a technique or toolbox that I should look into to do something like this. I was thinking neural networks as they are good at predicting nonlinear behavior given a lot of data, however I couldn't find a way to do what I proposed above. I know this question is open ended, so I am just trying to get an idea of where to look.

Answer 1

[Sreeram Mohan]

Hi Joshua,

Looks like in this case both System identification as well as neural networks toolboxes could be of viable approaches.

As far as the neural networks you could capture the system response (time series) and then train a neural network (either a pattern network or a time series neural network if your intention is to predict ) and then deploy it for use.

In one of the cases I have experimented capturing data from a simulink model and trained a neural network pattern recognizer and then been able to infer what are the components causing a particular kind of response and have seen good results. How ever I want to keep you informed that you may have to generate a lot of data (unique learnable features for the neural networks) and training to get this working !

Hope this helps !

Sreeram Mohan