Detecting and Addressing Model Structural Error in Forecasting for Model Predictive Control

Abstract

Research in dynamical systems forecasting has focused on how an understanding of the nonlinear behavior of models, chaos, and individual challenges predictability. A systemic approach is presented for improving model predictive strategies in uncertainties in dynamical system forecasting for decision-making. In this thesis, as a first step, different dynamical system modeling approaches, time-series embedding, and data-driven and machine learning techniques in literature are presented and how structural model error can be leveraged in improving prediction in chaotic systems like Lorenz and Kinetic Swinging Sticks. Using great models from existing literature, a structural model error approach in dynamical systems and chaos framework is explored in defining error bounds for evaluating forecasting, an addition to mean square error