Design of Artificial Neural Networks Based on Genetic Algorithms to Forecast Time Series

In this work an initial approach to design Artificial Neural Networks (ANN) using Genetic Algorithms (GA) is tackle. A key issue for these kind of approaches is what information contains, or is included, in the chromosome that represent an ANN, and there are two principal ideas about these question: first, information about parameters of the topology, architecture, learning parameters, etc. of ANN, i.e. Direct Encoding; second, initial information related to a constructive method that give rise to an ANN topology or architecture, i.e Indirect Encoding. The results for an Direct Encoding (in order to compare with Indirect Encoding developed in future works), for design ANN to NN3 Forecasting Time Series Competition.