mutation operation, the slot of each non elite RPN chromosome

d by a new RPN chromosome mutated from a randomly selected

N chromosome. For a dual-chromosome crossover operation, the

wo non-elite RPN chromosomes are replaced by two new RPN

omes generated through the crossover operation based on two

y selected elite RPN chromosomes. These new RPN

omes are then inserted into two slots of the non-elite RPN

ome set. For the single-chromosome crossover or the self-

r operation, only one elite RPN chromosome is selected. This new

ome is inserted into the non-elite RPN chromosome set.

monstrate how GP works, a simple function approximation (or

regression) problem was used. The function was designed as

here x, y and z were three variables, t was the function output and

regression error,

ݐൌሺݖെݔሻ∗ሺݖ൅ݕሻ൅ߝ

ntroduction of an error term was based on the assumption that

data may be corrupted with a noise. The error term in this

was following a normal distribution, i.e., ߝ~࣡ሺ0, 0.1ሻ. For this

300 data points were generated. Figure 8.21 shows the

on of the target values, i.e., the t values. The pool size was

as 100. The elite set size was 10 and the non-elite set size was

value of ߙ was 0.9. The maximum learning cycle was 100. The

mination rule was designed as the stabilisation of the fitness

ment. If the fitness measurement is maintained and unchanged

n the equation below) in continuous ten cycles, the learning was

d, where ߜൌ1݁െ6,

1

ܰ^{෍ሺݐ௜െݐ̂௜ሻଶ}

ே

௜ୀଵ

൑ߜ

(8.10)