Abstract:
Nature inspired and bio-inspired algorithms have been recently used for solving low
and high dimensional search and optimization problems. In this context, Bacterial
Foraging Optimization Algorithm (BFOA) and Differential Evolution (DE) have been
widely employed as global optimization techniques inspired from social foraging behavior
of Escheria coli bacteria and evolutionary ideas such as mutation, crossover, and selection,
respectively.
BFOA employs chemotaxis (tumble and run steps of a bacterium in its lifetime)
activity for local search whereas the global search is performed by elimination-dispersal
operator. Elimination-dispersal operator kills or disperses some bacteria and replaces
others randomly in the search space. This operator mimics bacterium’s death or dispersal
in case of high temperature or sudden water flow in the environment. DE employs the mutation and crossover operators to make a local and a global search
that explore the search space. Exploration and exploitation balance of DE is performed
by two different parameters: mutation scaling factor and crossover rate. These two
parameters along with the number of population have an enormous impact on optimization
performance.
In this thesis, two novel hybrid techniques called Chemotaxis Differential Evolution
Optimization Algorithm (CDEOA) for low dimensions and micro CDEOA (μCDEOA)
for high dimensional problems are proposed. In these techniques, we incorporate the
principles of DE into BFOA with two conditions. What makes our techniques different
from its counterparts is that it is based on two optimization strategies: exploration of a
bacterium in case of its failure to explore its vicinity for food source and exploitation of
a bacterium in case of its achievement to exploit more food source. By means of these
evolutionary ideas, we manage to establish an efficient balance between exploration of
new areas in the search space and exploitation of search space gradients. Statistics of
the computer simulations indicate that μCDEOA outperforms, or is comparable to, its
competitors in terms of its convergence rates and quality of final solution for complex high
dimensional problems.