Genetic Algorithms for Tracking Changing Environments.

This paper contrasts three different strategies: the Standard GA using a constant level of mutation, a mechanism called Random Immigrants, that replaces part of the population each generation with randomly generated values, and an adaptive mechanism called Triggered Hypermutation that increases the mutation rate whenever there is a degradation in the performance of the time-averaged best performance.

Abstract

In this paper, we explore the use of alternative mutation strategies as a means of increasing diversity so that the GA can track the optimum of a changing environment. This paper contrasts three different strategies: the Standard GA using a constant level of mutation, a mechanism called Random Immigrants, that replaces part of the population each generation with randomly generated values, and an adaptive mechanism called Triggered Hypermutation, that increases the mutation rate whenever there is a degradation in the performance of the time-averaged best performance. The study examines each of these strategies in the context of several kinds of environmental change, including linear translation of the optimum, random movement of the optimum, and oscillation between two significantly different landscapes. These first results should lead to the development of a single mechanism that can work well in both stationary and nonstationary environments. 1