Optimization for Nonlinear Time Series and Forecast for Sleep

摘要

It is important processes that phase-space diagram and computation of geometrical eigenvalues are reconstituted in nonlinear dynamical analysis. It's difficult to analyze nonlinear system such as EEG real-time because the algorithms of phase-space diagram reconstitution and geometrical eigenvalue computation are complex on both time and space. The algorithms were optimized to reduce their complexity, after that the algorithms were parallelized, at last the integrated algorithm's running time is 1/30 of the running time before optimization and parallelization. It was found that the value of correlation dimension can reflect sleep stages after analyzing the sleep EEG, final sleep stages were also forecasted simply.