Fuzzy prediction of time series

Abstract

An approach to time series extrapolation based on fuzzy control is described. The standard exponential averaging scheme is inflexible in that it gives a fixed weight to past history, thus ignoring transient phases in system dynamics. A modification to the scheme where the control parameter of the averaging scheme is dynamically adjusted by a simple fuzzy logic controller is presented. The design of the controller is described. The scheme was evaluated by simulation on test workloads and by application to the real-world problem of flow control in communication networks. The sensitivity of the system to its descriptive parameters is outlined.<>