Evaluation of Robustness and Performance of Environmental Influences ...

In this work we examine the impact of dynamical environments on evolutionary algorithms and ant colony optimization. In particular we evaluate these algorithms for a shortest path problem in a continuously changing network.The problem of establishing the shortest or cheapest path in a randomly changing network acts as a representative for the class of situations to be solved. Also nodes are added and removed to this network being reminiscent of today's wireless networks consisting of a variety of mobile devices having changing signal quality between the participants and peers entering and leaving the networks. Within this problem definition lies the possibility of radical changes since former best paths may not exist any longer and may have collapsed after a series of such changes. Resilience to and recovering from such incidents correspond to the robustness quality whereas finding the best solution in the least number of steps indicate good performance values.We first simulate a radical Darwinistic approach; the genes of individuals of a population are recombined and mutated randomly, evaluated by a fitness function and, dependant on the fitness value, selected for the next generation in order to gradually create a better performing population over several generations.The results are compared to a set of different Ant Colony Systems best suited for this class of problems. Choosing different evaporation rates, greediness and visibility values constitute to a variety of ACS instances yielding candidates for the optimal problem solution with respect to the quality criteria objective.The goal of this work is to obtain a better understanding of these techniques when designing and building complex systems. The results derived from this work may give new insights when creating self organizational systems based on artificial neuron networks and evolutionary algorithms in mind.