| The utility of Machine Learning (ML) approaches to solve non-linear systems is gaining
attention within the research community. As computational capacity improves, proliferation
of ML applications increases as well. The vast number of the systems people interact with
are dynamic in nature. ML approaches often solve a complex dynamic system with limited
or no domain knowledge but have shown success by applying deep learning techniques that
capture the system’s characteristics as a black box model. Numerous popular artificial
neural networks deliver high performance with or without transfer learning but without
much explanation as to why deep learning efforts s how f requent s uccesses i n a particular
application. The study investigates several ML techniques for a wide spectrum of applications
to address effective workflow and adaptation of ML approaches along with control strategies
applied to complex problems. In addition to the investigations considered in this research
document, effort is made to include domain knowledge as to preserve some degree of insight to
the internal system characteristics. The study described here is segmented into six separate
research works. The dissertation addresses identifying merits and challenges associated with
ML and data-driven control strategies applied to dynamic systems.
Keywords: Machine Learning, Deep Learning, Dynamic Systems, Control Design, Rein-
forcement Learning |