CHEMOTAXIS DIFFERENTIAL EVOLUTION OPTIMIZATION TECHNIQUES FOR GLOBAL OPTIMIZATION

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.