Design and Voltage Supply of High-Speed Induction - Aaltodoc

138
C. GENETIC ALGORITHM
The idea in genetic optimization is to imitate the evolution of nature. A whole population of designs
is studied instead of a single one. Each design has a design vector consisting of the values of the
design variables. This vector x can be viewed as a chromosome where a single design variable
corresponds to one gene
x = [x1,...,xn], (C1)
where xi is a value of i:th design variable and n is the number of variables.
Genetic operators imitate the phenomena in nature which change the genes of the designs. These
operators can be divided into mutation and crossover operators. Mutation operators change the
value of one or more genes in a chromosome. Crossover operators imitate breeding where two
designs give birth to new designs. In this case, the genes of the new designs are a mix or a
combination of the parents’ genes. New design vectors replace some of the old ones and thus we
have a new generation of the population. Selection routines that select the parents for breeding and
the individuals to perish imitate the selective laws of evolution in nature. Strong individuals are
more likely to produce offspring and weak individuals are more likely to die.
The GA program used in this study has the following algorithm:
1. Initialize a population of designs
2. Evaluate each design in a population
3. Select designs for breeding and designs to die
4. Apply genetic operators to create new designs
5. Replace the dying designs with the new designs
6. Evaluate the new designs in a population
7. To continue the optimization, return to step 3. Otherwise return the solution.
The program can handle the following types of linear constraints.
Domain constraints l < x < u, (C2)
where l = [l1,...,ln] and u = [u1,...,un].