Experimental evaluation of some prediction methods on time series

• 1、East China Normal University
• 2、China Ship Scientific Research Center

Abstract：This paper presents out experimental evaluation of several methods of predicting time series on the platform of Matlab. Those methods include Wiener-filter method, Kalman-filter method, power spectrum method, autoregressive model, the methods based on autoregressive moving average model, and neural networks. Note that short-term forecasting of ship motion remains a challenge issue in the field of ship science. Results relating to that issue greatly affect on the techniques of taking-off of carrier planes and the reliability during taking-off of carrier planes. As the main external load of a ship structure is ocean wave, the prediction of ocean-wave signal is crucial to the forecasting of ship motion. Proper methods of predicting ocean-wave signals together with ship structural mechanics can provide the base of the forecast of ship motion. There are several methods of prediction of time series. Hence, the selection of prediction methods becomes a research issue in the field. Based on this, this research work uses a real ocean-wave signal for the evaluation of those methods of prediction on the platform of Matlab, providing the experimental evidence for the selection of prediction methods for ocean-wave signals.

Keywords： prediction time series filter power spectrum autoregressive neural network