
was measured as an average over the entire population rather than a
population best. Finally, on f5 the algorithm was able to attain a best
value of 0.943665 on 20 out of 20 attempts, in 34 dimensions; we pre-
sume that to be the global optimum. The system converged rapidly on
this fitness peak every time.
The five functions were implemented in a single program, where the
only code changed from one test to another was the evaluation function.
All other aspects of the program, including parameter values, ran identi-
cally on the various functions. Thus it appeared from this preliminary re-
search that the binary particle swarm was flexible and robust.
No ...