Modelling chaotic time series using hybrid approaches

Abstract

Chaotic time series have as main feature, the possibility of short time prediction of their future evolution. The time series of the natural systems are seldom generated by purely linear or nonlinear systems. They are usually the result of the evolution in time of nonlinear systems. Therefore, the time series will encapsulate both linear and nonlinear components. This hybrid character of such time series makes the prediction process very complex and not accurate using a single technique. Combining different techniques and models increases the chances of representing and modeling complex relationships between data and to improve the quality of time series prediction. The paper presents the results obtained in the case of the electricity spot price time series modeled using ARIMA technique for the linear part and a neural network for the nonlinear component.