 |
    |
An Examination of Hypermutation and Random Immigrant Variants of mrCGA for Dynamic Environments
Gregory R. Kramer and John C. Gallagher
Department of Computer Science and Engineering
Wright State University,
Dayton, OH, 45435-0001
{gkramer,johng}@cs.wright.edu
Abstract. The mrCGA is a GA that represents its population as a vector of probabilities, where each vector component contains the probability that the corresponding bit in an individual's bitstring is a one [2]. This approach offers significant advantages during hardware implementation for problems where power and space are severely constrained. However, the mrCGA does not currently address the problem of continuous optimization in a dynamic environment. While, many dynamic optimization techniques for population-based GAs exist in the literature, we are unaware of any attempt to examine the effects of these techniques on probability-based GAs. In this paper we examine the effects of two such techniques, hypermutation and random immigrants, which can be easily added to the existing mrCGA without significantly increasing the complexity of its hardware implementation. The hypermutation and random immigrant variants will be compared to the performance of the original mrCGA on a dynamic version of the single-leg locomotion benchmark.
LNCS 2723, p. 454 f.
lncs@springer.de
© Springer-Verlag Berlin Heidelberg 2003