Identification environment and robust forecasting for nonlinear time series

Abstract

Abslract. In this paper, the methods of time series for nonlinearity are briefly surveyed, with particular attention paid to a new test design based on a neural network specification. The proposed integrated expert system contains two main components: an identification environment and a robust forecasting design. The identification environment can be viewed as a integrated dynamic design in which cognitive capabilities arise as a direct consequence of their self-organizational properties. The integrated framework used for discussing the similarities and differences in the nonlinear time series behavior is presented. Moreover, its performance in prediction proves to be superior than the former work. For the investigation of robust forecasting, we perform a simulation study to demonstrate the applicability and the forecasting performance.

Forecasting Fuzzy Time Series on a Heuristic High-Order Model

“…Drawing on Song and Chissom's first-order time invariant and time-variant models, Sullivian and Woodall proposed the time-invariant Markov model with linguistic labels for probability distribution (Sullivian and Woodall 1994). Then, Wu proposed the fuzzy method to predict future values and yielded better results than traditional forecasting methods when applied to forecast teacher numbers, government expenditure, and other variables (Wu 1994). Chen proposed a new fuzzy-time series model that yielded good forecasts (Chen 1996).…”