Parameter estimation behavior by GA optimization toolbox

For the parameter estimation by GA optimization toolbox, is that possible that the estimated parameter deviates from the true values?

Thanks!

Answer 1

[Shashank Prasanna]

Do you know the true values? It is always entirely possible that the estimates are not the same as your true values that let you generate the data. After all you are only working with a sample dataset.

There are a number of ways to improve your results:

http://www.mathworks.com/help/gads/improve-results.html

You can use HybridFcn to try and improve the result using another solver from where GA terminated:

http://www.mathworks.com/help/gads/using-a-hybrid-function.html

[Khoo]

Great! I will try this later, I'm still running the simulations. I see the points that to improve the estimation results one can consider to change the bounds, add in hybrid function and the last one, changing the initial values.

But why the GA has no requirement on initial values to put in?

[Shashank Prasanna]

GA is stochastic, it makes random choices for start point as well as for successive generations. You will have to read the theory for the working of GA. The documentation page does a fairly good overview:

http://www.mathworks.com/help/gads/what-is-the-genetic-algorithm.html

[Khoo]

Yea, I know the random initial value just like SA does. But then how can I find the mean square error then? The mean square error is defined by

mean(error^2)

where the error =estimated-initial point

Thanks!

[Khoo]

Okay. I guess i get it. By setting the 'InitialPopulation' in gaoptimset to my preference initial point.

Thank you very much!