Cellular Embryogenic Representations in Evolutionary Design (Master thesis)

Evolutionary Algorithms are robust tools for the optimization of complex systems. They have given rise to the modern area of automated engineering-Evolutionary Design. Inspired by biological systems, several approaches to evolving complex phenotypes have been proposed. This work focuses on a developmental approach with indirect genetic encodings in the representation of cellular automata capable of optimizing multiple complex target flags. More specifically, simultaneous optimization of multiple color and structure patterns was tested using cellular representations. Cellular grid topology, diffusion-reaction model, feedforward neural network, and evolutionary algorithm were used to optimize the cellular representation and decode its genotype into the phenotype, thus into a color picture similar to the optimized target flag. A target specific impulse was encoded into border cells ’ values in order to optimize multiple color patterns. A variable cellular grid has been used for the structure optimization purpose. The results are promising because they showed the capability of the cellular representation to perform multiobjective optimization of complex patterns. Keywords