Print

Brahma, Sarnaduti

Electrical and Biomedical Engineering

2021

Ph. D.

Quasilinear Control (QLC) is a theory with a set of tools used for the analysis and design of controllers for nonlinear feedback systems driven by stochastic inputs. It is based on the concept of Stochastic Linearization (SL), which is a method of linearizing a nonlinear function that, unlike traditional Jacobian linearization, uses statistical properties of the input to the nonlinearity to linearize it. Until now in the literature of QLC, SL was applied only to feedback systems with single-variable nonlinearities that appear only in actuators and/or sensors. In this dissertation, my recent contributions to the literature of QLC are summarized. First, the QLC theory is extended to feedback systems with isolated multivariate nonlinearities that can appear anywhere in the loop and applied to optimal controller design problems, including systems with state-multiplicative noise. Second, the numerical properties of SL, particularly, the accuracy, robustness, and computation of SL, are investigated. Upper bounds are provided for the open-loop relative accuracy and, consequently, the closed-loop accuracy of SL. A comparison of the computational costs of several common numerical algorithms in solving the SL equations is provided, and a coordinate transformation proposed to improve most of their success rates. A numerical investigation is carried out to determine the relative sensitivities of SL coefficients to system parameters. Finally, QLC is applied to the optimal primary frequency control of power systems with generator saturation, and control of virtual batteries in distribution feeders. The expected impacts of this work are far-reaching. On the technical front, this work provides: i) a new set of theoretical and algorithmic tools that can improve and simplify control of complex systems affected by noise, ii) information to control engineers on accuracy guarantees, choice of solvers, and relative sensitivities of SL coefficients to system parameters to guide the analysis and design of nonlinear stochastic systems in the context of QLC, iii) a new computationally efficient method of addressing saturation in generators or virtual batteries in modern electric power systems, resulting in efficient utilization of resources in providing grid services. On the societal front, this work: i) enables technologies that rely on computationally-efficient algorithms for automation of complex systems, e.g., control of soil temperature for agriculture, which depends on multiple factors like soil moisture and net radiation, ii) allows effective coordination of controllable smart devices in people's homes, so as not to hamper their quality of service, and iii) provides a stepping stone towards key societal challenges like combating climate change by facilitating reliable operation of the grid with significant renewable penetration.