A Novel Online Pseudospectral Method for Approximation of Nonlinear Systems Dynamics

원문 초록

This paper presents a novel online system identification approach utilizing the Chebyshev pseudospectral (PS) method to approximate the dynamics of a continuous-time nonlinear system. Unlike conventional periodic sampling, the proposed identification scheme employs aperiodic state sampling, leveraging the Chebyshev nodes to achieve the desired approximation accuracy. Unlike traditional off-line PS approaches, the scheme utilizes a moving time-window strategy to compute the sampling instants (Chebyshev nodes) forward in time for state measurements. Within each time window, the number of sampling instants is adaptively determined to meet the specified approximation accuracy. The Chebyshev basis is also shifted for each time window to accommodate arbitrary approximation intervals. The least-squares approach is employed to estimate the coefficients of the shifted Chebyshev basis functions using the measured state and its derivatives at the end of each window, resulting in a piecewise approximation of the drift dynamics of the system. In the second step, the identified drift dynamics are utilized to design an adaptive state estimator to reconstruct the continuous system states from the aperiodic state measurements. To address the smoothness of the piecewise approximated system dynamics, the Chebyshev coefficients are recomputed at the window transition instants, between two consecutive time windows, by enforcing continuity of the approximated function and its derivatives. In addition, analytical results are provided to determine the number of sampling instants within each time window, which guarantees the desired approximation accuracy. The boundedness of function approximation and state estimation errors is also proved analytically. Finally, numerical simulation results are included to validate the proposed scheme.