Do I need to normalize input parameters to Particle Swarm Optimization?

Hi there,

If the optimizing parameters are in different magnitudes of orders, do I need to scale the upper and lower bounds of the parameters to a scale lilke [1,0]? or does PSO actually do it internally?

Many thanks

Answer 1

[Alan Weiss]

You don't usually need to scale your parameters, but many problems are solved more reliably if you do. See Center and Scale Your Problem. I would first try to solve the problem without centering and scaling, and then center and scale if the solver doesn't converge well.

Alan Weiss

MATLAB mathematical toolbox documentation

[Buddhi Wimarshana]

Thanks Alan, I am not sure if PSO does it internally, I found this particular phrase,

For u1 and u2 uniformly (0,1) distributed random vectors of length nvars, update the velocity

v = W*v + y1*u1.*(p-x) + y2*u2.*(g-x).

mentioned in here PSO Algorithm.

Any thoughts on this? didn't quite get what u1 and u2 are on the article...