Least Squares Optimization Basics

Hello All,

I'm very new to optimization. I have a problem and not sure the best way to go about it. Lets say from data from the field for a model at a given time, I am able to have one data point. For example,

(t1,y1) (t2,y2) (t3,y3)

Now lets say I have an analytical code that tries to model this behavior such that the output is,

(t1,z1) (t2,z2) (t3,z3)

The analytical code uses 10 parameters that I want to guess such that the error between the "y" and "z" is at a minimum by using a least squares minimum.

I'm looking at fmincon and lsqlin but am confused to say the very least. If there is any good material that anyone can link or tell me the direction to head in, that would be great. Thank you for your time!

Answer 1

[Matt J]

I'd recommend lsqcurvefit,

x= lsqcurvefit(fhandle,x0,[t1,t2,t3],[y1,y2,y3])

where x0 is a vector of your 10 initial parameter guesses. You need to provide the function handle fhandle to the function z=F(x,t) that generates your z-samples.

[Alan Weiss]

To expand on this recommendation, look at Nonlinear Data-Fitting and maybe at Nonlinear Curve Fitting with lsqcurvefit.

Alan Weiss

MATLAB mathematical toolbox documentation