Pattern Search dependance on initial population

I am working on an optimization problem using Pattern Search. It appears the optimized solution is heavily dependent on the initial population, and I suspect the optimized solution is not the global minimum. What action could I take to ensure Pattern Search finds the global minimum as opposed to a local minimum near the initial population? Thank you for your help.

Answer 1

[Alan Weiss]

Firstly, patternsearch is not population based. Instead, it takes an initial point, usually named x0.

Secondly, like every other solver that is based on initial points, there is no guarantee that the solver reaches a global minimum.

If you have finite bounds lb and ub (lower bound and upper bound vectors), you can repeatedly try the random initial points

 x0 = lb + rand(size(lb)).*(ub-lb);

You might also want to examine the suggestions in Refining the start points, realizing that they are not written specifically for patternsearch.

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation