Abstract:
This thesis proposes a method to overcome the parameter setting problem of genetic algorithms. This method is denoted as 'Adaptive Parameter Control Algorithm' (APCA). The concept fo APCA is based on two levels of genetic algorithms. The task level genetic algorithm (lower level genetic algorithm) solves the original problem, while the meta-level genetic algorithm (upper level genetic algorithm) optimizes the parameters of the task level. Both levels operate concurrently. Each individual in the population of the meta-level genetic algorithm is a parameter set for the task level genetic algorithm. The evaluation of each individual inthe meta-level population is carried out by assigning it as the parameter set of the task level genetic algorithm, the performance of the task level genetic algorithm is then used as the fitness. The task level genetic algorithm with multiple subpopulations is used to parallelize the evaluation of the meta-level population.This fits well with a coarse-grained model parallel genetic algorithm. The empirical results indicate that APCA is not only faster than other algorithms, but APCA also more reliably finds optimal solutions.
