Recursive prediction of chaotic time series (Q1322783): Difference between revisions

This is an article on time series prediction for chaotic systems. Chaotic systems are seen as systems with impossible long term predictions due to the exponential divergence of trajectories. Examples of error prediction are based on the Henon map system : \(x_{n+1} = 1 + y_ n - ax^ 2_ n\), \(y_{n+1} = bx_ n\). The proposed prediction method is based on radial bases centers and a recursive approach. This means approximation with linear combinations of, possibly nonlinear, functions of observations. For the prediction both simple recursive least squares methods and recursive modified Gram-Schmidt methods are analyzed. The author confirms the numerical stability of the orthogonal method when the number of observation grows. The advantages of using such recursive methods are: they need less computer memory even for great number of observation, predictions are made even from the beginning of observations, especially useful for real time systems, and a good estimation for nonstationary systems, where changes of dynamics are captured. The style is very clear and even a person unfamiliar with time series prediction can understand the theory behind the proposed method of prediction. Thus the author explains the Takens embedding theorem, then he reviews the problems like: selection of the dimension of the observation space, the time interval between observations and the selection of the base functions. One can also find a few historical information on the recursive least squares algorithms. There is also a very good reference list.