Time series forecasting through Polynomial Artificial Neural Networks and Genetic Programming

The Polynomial Artificial Neural Network (PANN) has shown to be a powerful Network for time series forecasting. Moreover, the PANN has the advantage that it encodes the information about the nature of the time series in its architecture. However, the problem with this type of network is that the terms needed to be analyzed grow exponentially depending on the degree selected for the polynomial approximation. In this paper, a novel optimization algorithm that determines the architecture of the PANN through Genetic Programming is presented. Some examples of non linear time series are included and the results are compared with those obtained by PANN with Genetic Algorithm.