Forecasting for nonlinear and nonstationary systems using intrinsic functional decomposition models

Abstract

The purpose of this study is to develop nonlinear and nonstationary time series forecasting methods to address modeling and prediction of real-world, complex systems. Particular emphasis has been placed on nonlinear and nonstationary time series forecasting in systems and processes that are of interest to IE researchers. Two new advanced prediction methods are developed using nonlinear decomposition techniques and a battery of advanced statistical methods. The research methodologies include empirical mode decomposition (EMD)-based prediction, structural relationship identification (SRI) methodology, and intrinsic time-scale decomposition (ITD)-based prediction. The advantages of using these prediction methods are local characteristic time scales and the use of an adaptive basis that does not require a parametric functional form (during the decomposition process). The utilization of SRI methodology in ITD-based prediction also provides a relationship identification advantage that can be used to capture the interrelationships of variables in the system for prediction application. The empirical results of using these new prediction methods have shown a significant improvement in the accuracy for customer willingness-to-pay and automobile demand prediction applications.