Forecasting Time Series with a New Architecture for Polynomial Artificial Neural Network

Polynomial artificial neural networks have shown to be a powerful Network for forecasting non linear time series. With this type of networks it is possible to have information about the nature of the time series analyzed. However, the problem with this type of network is the computation time required and sometimes the huge number of terms of the polynomial obtained. In this paper, a novel optimization algorithm that improves the number of terms of the polynomial is presented. The architecture adaptation uses genetic algorithm to find the optimal architecture for every example. Some examples of non linear time series are shown.