„Badanie wpływu parametrów mutacji na proces ewolucji”

Abstract

The work is devoted to the study of the processes of population evolution occurring with the use of a neuroevolutionary algorithm. Populations are individuals that have a genetic code, which is a representation of a neural network. The network is used to solve the problem of landing a lander in a designated place. The information on the basis of which the decision is made to start a given engine with the appropriate power, is provided from the sensors of the vehicle. The aim of the evolution is to develop a neural network, which guarantees the required effectiveness in successful landings. In this paper you will find descriptions of tests performed on three sets of variables corresponding to the probability of occurrence of a structural mutation in a neural network of individuals from a population. From among the four variants of mutation parameter values used by me, the best results were achieved by set four. Also, regardless of the mentioned parameters, I made conclusions common to all evolutions, related to the use of the NEAT algorithm. This solution in most cases generates small neural networks without redundant nodes or connections, this gives an advantage over manual determination of the network structure, however, due to the randomness of mutations occurring and the nature of the evolution process, the appearance of redundant nodes is possible.