Application of Robust Observers to Time-Series Forecasting

Abstract

Robust observers are frequently designed to estimate the states of a system that prohibit direct measurements. In this thesis we develop an efficient algorithm for designing an observer which provides robust estimation of higher-order derivatives of a continuous time-series. These estimations are subsequently used in design of a novel procedure which offers a fairly accurate forecast of time-series on relatively short time-horizons. The dependence of forecast error upon duration of forecast-horizon is studied on several representative examples.