27.06.2013 Views

Evolution and Optimum Seeking

Evolution and Optimum Seeking

Evolution and Optimum Seeking

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

156 <strong>Evolution</strong> Strategies for Numerical Optimization<br />

Table 5.1: Two point crossover within a GA <strong>and</strong> its e ect on the<br />

phenotypes<br />

Bit strings Phenotype<br />

Parent 1 0111 1100 7 12<br />

Parent 2 1000 1011 8 11<br />

Two point crossover<br />

O spring 1 0000 1000 0 8<br />

O spring 2 1111 1111 15 15<br />

completely contradicts the proverbial saying that the apple never falls far from the tree.<br />

Table 5.1 shows a simple situation with two parents producing two o spring by means of<br />

two point crossover, on a bit string of length 8, <strong>and</strong> encoding two phenotypic variables<br />

in the range [0 15]in the st<strong>and</strong>ard Boolean form. Neither discrete nor intermediary<br />

recombination within ESs can be that disruptive intermediary recombination always<br />

delivers phenotypic values for the o spring between those of their parents. The assumption<br />

that mutations are not necessary for the GA process may even stem from that disruptive<br />

character of recombination that permits crossover points not only at the boundaries of<br />

meaningful parental information but also within the genes themselves.<br />

ESs obey the general rule, that mutations are undirected, by means of using normally<br />

distributed changes with zero mean{even in the case of correlated mutations. That this<br />

is not so for GAs can easily be seen from Figure 5.16. Without selection, the GA process<br />

thus provides biased genetic drift, depending on the actual situation.<br />

Table 5.2 presents the probability transition matrix for one phenotypic integer variable<br />

xi in the range [0 3]encoded by means of two bitsonly.Let<br />

p = pm<br />

1<br />

2<br />

single bit inversion probability <strong>and</strong><br />

q =1; pm probability of not inverting the bit<br />

From Table 5.2 it is obvious that among all possible transitions (except for those with-<br />

Table 5.2: Transition probabilities for mutations within a GA<br />

xi new<br />

Genotype 00 01 10 11<br />

Phenotype 0 1 2 3<br />

00 0 q 2 pq pq p 2<br />

xi old 01 1 pq q 2 p 2 pq<br />

10 2 pq p 2 q 2 pq<br />

11 3 p 2 pq pq q 2

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!