MBC model fitting has more than one result, why?

Hi there,

I am using MBC-Model to fit some data using for example Gaussian Process Model / RationalQuadratic / PureQuadratic, and every time it gave me different results. I wonder why the answer is NOT one and only one?

Look forward to a quick answer.

Thanks!

[John D'Errico]

I don't know what MBC-Model is, so I cannot answer with certainty. But there are a variety of reasons why a problem may produce multiple solutions in a modeling context.

1. If the results are all similar, but not exactly equal, then this is usually due to convergence tolerances on the result, perhaps based on a random start point, so you get subtly differnt results each time, but they may be all effectively the same. Or some solutions may be better than others. Hard to know which it is. Often you can set a fixed start point of your own.
2. If the results tend to cluster in a set of distinct points, then there are likely several local solutions. Again, which solution arises can be based on a random start point.
3. If the results appear to lie along a line or curve in the parameter space, this may simply be due to insufficient data to resolve all necessary parameters, or it may reduce to a poor choice of model.

Answer 1

[York Zhu]

mbcmodel is model based calibration toolbox in MATLAB. Your explanation seems to make sense, but how to choose or set up starting point? Thanks