High roughness time series forecasting based on energy associated of series

Abstract

In this study, an algorithm to adjust parameters of high roughness time series based on energy associated of series using a feed-forward NN-based model is presented. The criterion for adjustment consists of building time series values from forecasted time series area and taking into account the roughness of series. These values are approximated by the NN to make a primitive calculated as an area by the predictor filter used as a new entrance. A comparison between this work and another that involves a similar approach to test time series prediction, indicates an improvement for certain sort of series. The NN filter output is intended to approximate the current value available from the series which has the same Hurst Parameter as the real time series. The proposed approach is tested over five time series obtained from samples of Mackey-Glass delay differential equations (MG). Therefore, these results show a model performance for time series forecasting and encourage to be applied for meteorological variables measurements such as soil moisture series, daily rainfall and monthly cumulative rainfall time series forecasting.