Apga: an adaptive Parallel Genetic Algorithm

ABSTRACT We review the concept of deceptive functions, suggest that practical problems often have deceptive subproblems, and use this to motivate the need for robust genetic algorithms that can solve deceptive problems as well as straightforward ones. We develop apGA, an adaptive parallel Genetic Algorithm, that combines aggressive search with perpetual novelty, yet is able to preserve enough genetic structure to optimally solve variably scaled, non-uniform block deceptive and hierarchical deceptive problems. apGA combines elitism, adaptive mutation, adaptive exponential scaling, and temporal memory. We present empirical results for six classes of problems, including the DeJong test suite. Although we have not investigated hybrids, we note that apGA could be incorporated into other recent GA variants such as GENITOR, CHC, and the recombination stage of mGA.