The prediction of a single observable time series has been achieved with reasonable accuracy and duration for the nonlinear systems developed by Rossler and Lorenz. Based on Takens' Delay-vector Space, an artificial system has been generated using a polynomial least squares technique that includes all possible fifth order combinations of the vectors in the delay space. Furthermore, an optimum shift value has been shown to exist, such that any deviation decreases the accuracy and stability of the prediction. Additionally, an augmented form of the autocorrelation function, similar to the delay vector expansion, has been investigated. The first inflection of this correlation, typically in the dimension of the system, tends to coincide with the optimum shift value required for the best prediction. This method has also been utilized in conjunction with the Grassberger-Procaccia Distance correlation function to accurately determine the fractal dimension of the systems being investigated.
Library of Congress Subject Headings
System identification; Nonlinear theories; System analysis
Department, Program, or Center
Mechanical Engineering (KGCOE)
Ziegler, Edward H., "Nonlinear system identification" (1994). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus