Abstract
This paper presents a method of time series pattern recognition using recurrent neural networks (RNN) based on the genetic algorithms (GAs). The RNN could have dynamical characteristics because the RNN has feedback connections with time delay. The connection weights of the RNN transform into the gene which described by 16 bits binary code. In order to acquire the dynamics of the time series patterns, GAs operators which are a selection, a crossover and a mutation are applied to the gene of the RNN. In the numerical simulations, the training convergence of the RNN for some time series patterns are investigated. The simulation results show that the RNN evolved by the GAs has good performances of the training for time series patterns which generated from sine functions and for some pulse patterns. Finally, it is shown that the new method could determine some suitable network structures for time series training.
