Off-campus WSU users: To download campus access dissertations, please use the following link to log into our proxy server with your WSU access ID and password, then click the "Off-campus Download" button below.
Non-WSU users: Please talk to your librarian about requesting this dissertation through interlibrary loan.
Access Type
WSU Access
Date of Award
January 2024
Degree Type
Dissertation
Degree Name
Ph.D.
Department
Industrial and Manufacturing Engineering
First Advisor
Murat Yildirim
Abstract
Energy systems are experiencing a significant transformation marked by emerging requirements associated with the integration of renewables, electric vehicles, and distributed generation into the modern power grid. These changing dynamics and requirements mandate a new set of (i) predictive methods that use streaming industrial data to improve situational awareness and (ii) prescriptive models to offer improved control and optimization of operational decisions. An underlying enabler for these new set of methods is the availability of data, processing capabilities, and the innovative use of machine learning (ML) methods to address domain specific deployment challenges. This dissertation develops new approaches that revisit and transform two fundamental problems in energy systems: 1) federated prognostics: leveraging ML for privacy-preserving predictive models for predicting remaining life of lithium ion batteries, 2) ML-enhanced operations optimization: incorporating ML techniques for prescriptive models to accelerate solution methods for a central operational problem in power systems called stochastic unit-commitment (SUC).
The first problem concentrates on predicting the remaining useful lifetime of lithium-ion batteries, utilizing a federated learning (FL) approach. The proposed approach offers a paradigm shift from the existing methods that build on centralized data collection and processing from various clients, which suffer from communication and processing bottlenecks due to the substantial volume of information. Additionally, these methods raise significant privacy concerns and potential liabilities related to data breaches. To address these challenges and to enable scalable deployment of battery management systems, we propose a novel approach: a federated battery prognosis model. This model offers prognostics methods that decentralize the processing of battery standard data, such as current-voltage-time-usage information, with a focus on prioritizing privacy. Rather than exchanging raw data, our framework only shares model parameters with the central server, reducing the load on communication channels and safeguarding data confidentiality.
In the second research problem, we address computational scalability of SUC, a computationally challenging problem with complex constraints such as transmission line capacities and ramping limits. The SUC problem determines the most efficient combination of generating units for commitment and identifies optimal generation levels, by considering a large number of constraints and a significant uncertainty from demand fluctuations. Our research aims to develop new methods that use ML to enhance existing solution methodologies to reduce computation times. Specifically, our study introduces a hybrid approach that combines reinforcement learning and optimization for improving Benders decomposition implementation in SUC problems.
Recommended Citation
Altinpulluk, Nur Banu, "New Generation Of Machine Learning Models To Improve Prediction And Optimization In Energy Systems" (2024). Wayne State University Dissertations. 4030.
https://digitalcommons.wayne.edu/oa_dissertations/4030
