Fuzzy approach to time series prediction and its applications

Exponential smoothing (ES) as a technique for smoothing and forecasting of time series has been extensively used since its introduction. Its main feature is simplicity and hence ease of implementation. We present a new, fuzzy version of the smoothing and time series prediction (TSP) operator. It is a generalization of the ES procedure, namely, its nonlinear version. The operator can be used both for slowly varying trends and for fast and jump-like changes. At the same time, it keeps the simplicity of the corresponding processing. Comparison by simulation of several versions of ES (classical, adaptive, nonlinear and fuzzy) show the advantages and efficiency of our fuzzy generalization of the nonlinear ES. Possible application areas of the proposed fuzzy approach to time series prediction include DSP and pattern recognition, industrial engineering (robotics) and telecommunications (control and management).