Optimisation of neural networks in computer games with evolutionary algorithms

U ovom radu dan je uvid u primjenu neuroevolucije u računalnim igrama. Detaljno je opisan algoritam NEAT (NeuroEvolution of Augmenting Topologies) i principi na kojima se algoritam temelji. Kroz primjer izrade inteligentnih igrača i igranja jednostavnih računalnih igara prikazani su način korištenja i karakteristike evolucijskog algoritma NEAT pri kreiranju i treniranju neuronske mreže s ciljem optimalnog igranja računalne igre. Napravljen je pregled i analiza parametara genetskog algoritma te usporedba rezultata dobivenih izmjenama parametara genetskog algoritma.This paper provides an insight into the application of neuroevolution in computer games. The NEAT (NeuroEvolution of Augmenting Topologies) algorithm and the principles on which the algorithm is based are described in detail. Usage and characteristics of the evolutionary algorithm NEAT in creating and training a neural network with the aim of optimal computer game play are presented by evolving intelligent agents and playing simple computer games. Additionally, an overview and analysis of the parameters of the genetic algorithm and a comparison of the results obtained by changes in the parameters of the genetic algorithm were made