Self-adaptivity for constraint satisfaction: learning penalty functions

摘要: Treating constrained problems with EAs is a very challenging problem. Whether one considers optimization or constraint satisfaction problems, the presence of fitness function (penalty function) reflecting violation essential. The definition such penalty has great impact on GA performance, and it therefore important to choose properly. We show that ad hoc setting penalties for violations can be circumvented by using self-adaptivity. illustrate matter discrete CSP, Zebra problem, learned are large extent independent applied genetic operators as well initial weights.