Performance Comparison of Local Zero-Order Predictive Methods for Chaotic Time Series

Performances of local zero-order prediction methods for chaotic time series were compared in aspects of prediction accuracy, anti-noise and the ability of multi-step prediction through computer simulation. Simulation results show that distance-weighted predictive method is the best when there are no noises or only weak noises; exponential weighted predictive method is better than the others when there are large noises; exponential weighted predictive method and averaging method are basically the same in respect to multi-step prediction, but distance-weighted predictive method is the best in short-term prediction; multi-step prediction errors of standard discrete chaotic time series for the three methods do not increase with prediction step after they arrive at certain values, but for the chaotic time series sampled by continuous system, its multi-step prediction errors have periodicity.