Neural computing in pharmaceutical products and process development - Computer-Aided Applications in Pharmaceutical Technology - page 140

get mired at a local minimum/maximum (Chi et al., 2009). GAs can
search noisy, discontinuous, and non-convex solution spaces with great
profi ciency (Venkatasubramanian et al., 1994).
The GA searches through a population of potential solutions to the
problem, in order to fi nd the optimal one. Usually this population is
randomly created. If the optimal solution is not found, then the next
generation of potential solutions is created. Creation of a new population
of solutions is based upon biological principles; solutions from the
original population that were the closest to the optimal one are selected
in the new generation and are mutually combined, imitating in this
way mating and reproduction. Furthermore, random changes (mutations)
are also included in the process. Analogously to evolution, a population
of solutions gradually evolves and only the best (fi ttest) solutions are
held onto until the optimum solution is found, usually after several
generations. Combination of solutions (reproduction) increases the
area where potential solutions are sought (e.g. limits of variables can
become wider), whereas random changes (mutations) bring small,
but sometimes necessary, changes to potential solutions. Mutation
operators are stochastic transformations of an individual (Eiben and
Schoenauer, 2002).
Fitness functions (fi tness criteria) are used to estimate potential
solutions. Based on their fi tness scores, solutions can be either
eliminated or chosen for breeding of the next generation. The cycle
is continued until an optimal solution (based on particular criterion) is
found or when the previously set limit of number of generations is
reached.
GAs can operate either in generation mode or in the steady-state mode
(Goldberg, 1989). This means that either a pool of solutions is selected
and replaced by their children solutions in each cycle (generation mode),
or children solutions only replace the least fi t members of the parent's
generation solutions (steady-state mode).
In accordance to inspiration in biology, each individual solution in a
generation is called a chromosome (or string) and features (parameters,
properties, attributes) of the solution are denominated genes. Different
values of genes are its alleles. Chromosomes are commonly a collection
of integers or binary numbers, but can also be virtually any other type of
information (Parrill, 1996).
Selection functions that are most commonly applied in GAs are roulette
wheel selection, tournament selection, and elitism (Parrill, 1996). As its
name suggests, the roulette wheel is a random selection of population
members. Tournament selection compares, randomly or pair-wise, the

￿

￿ ￿