Abstract:

In order to improve chaotic time series forecasted precision and convergence rate of kernels about Volterra series model. A new method is presented to estimate the Volterra series’ kernels based on Kalman filtering algorithm. Meanwhile, based on the reconstruction of the chaotic dynamic space,  the chaos characteristics of ship motion time series are identified using the correlation dimension method and Lyapunov exponent method. The experiments of ship rolling motions time series multi-step forecasts are done based on the obtained Volterra series model. The simulation results indicate this method with Kalman filtering algorithm is feasible, the convergence rate and the prediction precision are better than the recursive least-square (RLS) algorithm and normalization least-mean-square (NLMS) algorithm.The theory basis is provided for the real-time online forecast.