Indirect prediction system for variables that have gaps in their time series

Abstract

Abstract Gaps in time series as well as the absence of such series make the implementation of prediction system difficult. This paper proposes a new methodology to fill gaps in time series that do not present fixed sampling rate. This paper also proposes the development of two forecast models for time series. The first model is based on autoregressive multilayer neural network that uses only the desired time series, while the second one is developed with multilayer neural network that uses pattern recognition in order to perform indirect predictions of a certain variable. Therefore, the second model does not need the variable time series to make predictions, but any time series that has correlation with the desired variable. The methodology is tested in limnological variables collected in the Paraguay River since 1987, and the results observed in each process are presented in order to validate the methodology of gap filling and forecast used.