Abstract
The nonlinear and non-convex properties of alternative current (AC) power flow (ACPF), the integration of energy storage devices with inter-temporal dynamics, and the uncertainty from renewable energy and uncontrollable power loads, bring tremendous computational challenges to the optimization problems of power system operation (PSO). With the availability of PSO data and the success of machine learning methods and big data techniques, data-driven approaches play a significant role in power system analysis, such as in state estimation, estimating distribution factors, the Jacobian matrix, and the admittance matrix. Therefore, this dissertation provides some discussions related to using machine learning approaches to develop data-driven approximations of ACPF and verify the efficacy of these data-driven approximations applied in optimal power flow (OPF) problems. Meanwhile, this dissertation also discusses the development of data-driven optimization approaches to deal with the complex optimization problems of PSO, such as multi-period OPF with energy storage devices under the uncertainty of renewable energy and power loads (REPL). More specifically, chapter 1 provides a detailed introduction on the problem statement studied, the approximation of ACPF, and the optimization of PSO under uncertainty. In chapter 2, the data-driven linear approximation (DDLA) of ACPF, and data-driven convex quadratic approximation (DDCQA) of ACPF are proposed, respectively, based on the polynomial regression and ensemble learning techniques, i.e., gradient boosting and bagging; then, apply those data-driven approximations to solve the OPF problems. Chapter 3 introduces the least absolute shrinkage and selection operator (LASSO) to learn the DDCQA with better computational efficiency, and proposes the framework of strategic sampling based on the physics-assisted sampling, metric learning and reinforcement learning to formulate a data-driven optimization method for chance-constrained multi-period OPF with energy storage devices under uncertain REPL. Chapter 4 exhibits the power of Bayesian hierarchical modeling (BHM) and determinantal point process (DPP) to further improve the accuracy of the learned DDCQA and the computational efficiency of existing data-driven optimization methods, considering the data correlations, i.e., uses BHM to generalize the learning process of DDCQA as a multi-level modeling problem and develops a DPP-based strategic sampling that can measure the relative weight of each sample and output a more efficient sample selection result than the existing strategic sampling. Chapter 5 explores adaptive LASSO and elastic net, another two alternative sparse learning algorithms applied to learn DDCQA, and compared with LASSO and BHM, as well as test the learning-based DDCQA in large-scale IEEE test systems including IEEE-500, -1354, and -2000 bus systems. Eventually, chapter 6 summarizes the conclusions and discusses the potential future research work.
Notes
If this is your thesis or dissertation, and want to learn how to access it or for more information about readership statistics, contact us at STARS@ucf.edu
Graduation Date
2023
Semester
Spring
Advisor
Li, Qifeng
Degree
Doctor of Philosophy (Ph.D.)
College
College of Engineering and Computer Science
Department
Electrical and Computer Engineering
Degree Program
Electrical Engineering
Format
application/pdf
Identifier
CFE0009537; DP0027544
URL
https://purls.library.ucf.edu/go/DP0027544
Language
English
Release Date
May 2028
Length of Campus-only Access
5 years
Access Status
Doctoral Dissertation (Campus-only Access)
STARS Citation
Hu, Ren, "Data-Driven Optimal Power System Operation Under Uncertainty" (2023). Electronic Theses and Dissertations, 2020-2023. 1578.
https://stars.library.ucf.edu/etd2020/1578
Restricted to the UCF community until May 2028; it will then be open access.