The Effects of Randomness
Maurice Clerc · 2015
This chapter discusses three types of initialization, generally considered to be the most useful. The first is uniform initialization, which is the simplest and most widespread model. The second is regular initialization methods, which are intended to improve performance by providing better coverage of the search space at the start of the iteration period. The third type of initialization technique is the No Man's Land approach, generally intended as a means of adding individuals to an existing population during the search process. The bulk of research on the subject has focused on the most efficient means of doing this in order to find an optimal position. Partially random movement defines a subset of the search space, accompanied by a probability distribution, known as the distribution of next possible positions (DNPP). Certain algorithms use a collection of strategies, moving from one to the other on the basis of intermediate results.