Modelado Estructural No Lineal de Series Temporales

Structural models are tools conceptually useful in time series modeling, because they allow the individual interpretation behavior of each structural component in the series; however, the difficult in the representation of nonlinear relationships of these models, have despised their use in real series. The Artificial Neural Networks (RNA) models are a promising alternative by the nonlinear time series modeling; nevertheless, the inabilities for to explain the parameters calculate of these models do that RNA have seen as black box. In this paper, we propose a hybrid approach that uses an explicit modeling of the structural characteristics of time series by structural models, joint with modeling of nonlinear relationships provided by neural networks. Experimental results with real series suggest the combination of this models can be an effective way to nonlinear time series modeling that the individual models.