A Recipe for the Simple Genetic Algorithm

The following prescription allows a quick implementation of a simple genetic algorithm to a given optimization problem. 

Define the problem. An optimization problem can be characterized by two sets of parameters, those that one wants to optimize according to a target solution, and those that one has to fiddle around with in order to approach the desired solution. Once defined in such terms, one can link these parameter sets to an algorithm that alters the free parameters while controlling the quality of the solution by the feedback parameters. 

Represent the free parameters of the problem in terms of search variables Si,i = 1. N where the dimensionality N of the search space is determined 

Define the genetic operators. They generate a new set of search variables from a given set. The two most frequently used operators are mutation and recombination. The mutation operator flips a variable Si to its inverse value if it is binary (or adds a random number to it drawn from a fixed distribution, if Si is real valued). This is most easily done with a fixed probability pm for each variable. The recombination operator swaps variables between two members of a population, the simplest procedure being uniform crossover where each variable 5 is swapped with a fixed probability Pc- 

Define a fitness function that measures the performance of a member (i.e., set of variables) in terms of a single real number. This is now the measure that defines the goal of evolution. In theoretical problems of optimization it is quickly defined. However, in some practical problems there might be some room for interpretation. Consider, for example, optimizing the timetable of an international airplane carrier A number of conflicting concepts of what is optimal have to be discussed and cast into one unified picture. 

---