Date of Award
Summer 8-22-2026
Document Type
Dissertation
Degree Name
Ph.D. in Electrical Engineering
Organizational Unit
Daniel Felix Ritchie School of Engineering and Computer Science, Electrical and Computer Engineering
First Advisor
Rui Fan
Second Advisor
Mohammad Mahoor
Third Advisor
Mohammad Abdul Matin
Fourth Advisor
Yousu Chen
Copyright Statement / License for Reuse
All Rights Reserved.
Keywords
Energy storage systems (ESS), Modern power grids, Oscillations, Deep reinforcement learning (DRL), Electrical engineering, Energy storage
Abstract
This dissertation introduces a novel strategy for the adaptive control of energy storage systems (ESS) aimed at enhancing the stability of modern power grids. With the rise in electricity consumption, modern power systems are often operated near their operational thresholds, increasing the risk of inter-area oscillations and large frequency deviation that threaten system stability. To address these challenges, this research proposes an innovative ESS control strategy designed to improve the damping of such oscillations. A deep reinforcement learning (DRL) approach is employed to address the limitations associated with static control parameters and insufficient damping effectiveness. A guided surrogategradient- based evolutionary strategy (GSES), representing state-of-the-art techniques, is utilized to train the DRL agents in an efficient, robust, and reproducible manner, further accelerated through parallel computing methods. The GSES-DRL based control strategy is applied to both inter-area oscillation damping and frequency response.
For inter-area oscillation damping control, the DRL agents generate power modulation signals, which serves as an operating reference for ESSs. The effectiveness of the proposed methodology has been validated on both medium (IEEE 39-bus) and large (MinniWECC) systems, demonstrating successful mitigation of various inter-area oscillations within 20 seconds, preventing system collapse and substantially improving grid stability. For frequency response, the strategy is extended to manage continuous extraction or absorption of power as required. Additionally, this work advances a coordinated approach for managing multiple ESS to provide enhanced frequency support during contingencies. To scale this strategy for coordinating multiple ESS, the framework transitions from a centralized controller to a decentralized Multi-Agent DRL (MADRL) architecture.
DRL agent execution relies heavily on wide-area phasor measurement unit (PMU), it introduces communication vulnerabilities. To address this, the dissertation integrates a mathematically deterministic online data recovery pipeline. Utilizing adaptive median absolute deviation (MAD) and singular value thresholding (SVT), this module intercepts and reconstructs PMU data affected by data loss, noise, delay, and disorder. Comprehensive case study via real-time software-in-the-loop co-simulation in an electromagnetic transient environment (PSCAD/EMTDC) on a modified multi-area low inertia IEEE 39-bus system confirmed that the GSES-DRL based control strategy consistently outperforms existing frequency response solutions.
Copyright Date
8-2026
Publication Statement
Copyright is held by the author. User is responsible for all copyright compliance.
Rights Holder
Abu Shouaib Hasan
Provenance
Received from ProQuest
File Format
application/pdf
Language
English (eng)
Extent
149 pgs
File Size
16.9 MB
Recommended Citation
Hasan, Abu Shouaib, "Deep Reinforcement Learning Based Adaptive Control of Energy Storage for Stability Enhancement in Modern Power Grids" (2026). Electronic Theses and Dissertations. 2768.
https://digitalcommons.du.edu/etd/2768