A Hybrid Time-Series Forecasting Model Using Extreme Learning Machines

Abstract

This study proposes a hybrid model which combines the linear autoregression (AR) with the nonlinear neural network (NN) based on the extreme learning machine (ELM) in an integral structure in order to improve the accuracy of time-series prediction. Unlike the developed hybrid forecasting models introduced in the literature, which usually treat the original forecasting models as a separate linear or nonlinear unit, the proposed hybrid model is an integrated model which can adapt well to both linear and non-linear situations often in periodical time series with a complicated structure. The hybrid algorithm is tested against different kinds of time series data and the results indicate that the hybrid algorithm outperforms the AR and the ELM-based neural network.