The comparison was conducted using test functions collected by
Dr. Eric Sandgren in his dissertation (Sandgren, Eric. “The Utility
of Nonlinear Programming Algorithms”, a thesis submitted for the
Degree of Doctor of Philosophy, Purdue Univercity, 1977). By superimposing
the multiplicative numerical noise on the main function we simulated
stochastic behavior. The numerical noise was distributed according
to normal distribution:
Ysto=Yini*(1+N(0,s)).
We compared four different
optimization methods:
IOSO (version IOSO NS 1.0);
MFD - modified method of feasible directions;
SQP - sequental quadratic programming;
SLP - sequental linear programming.
|