A New EA for Dynamic Problems

Abstract

Chapter 3 identified possible EA enhancements from nature and engineering that could improve the performance of EAs in dynamic environments. Key to these techniques was the management of diversity. Diversity management was identified as a method for ensuring continued exploration of the search space to identify possible changes. Chapter 4 provided the basis for an improved method for measuring the diversity that is more appropriate to dynamic fitness landscapes: the dispersion index. As discussed in Chap. 1, diversity maintenance has undergone initial investigations in the context of dynamic fitness landscapes; however, the quantification of the amount of diversity needed remains largely unexplored. In this chapter we provide the design details for an enhanced EA that permits us to exploit our new understanding of diversity and dispersion, and quantify the need for dispersion in dynamic fitness landscapes. This new EA is very flexible in controlling the amount of population dispersion at any time in an EA run, and addresses the problem of detecting changes in the fitness landscape. These features permit us to conduct experiments to quantitatively identify diversity needs for different types of dynamic problems.