Date of Award

5-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical and Computer Engineering

Committee Chair/Advisor

Dr. Christopher S. Edrington

Committee Member

Dr. Gokhan Ozkan

Committee Member

Dr. Ramtin Hadidi

Committee Member

Dr. Fatemeh Afghah

Committee Member

Dr. Behnaz Papari

Abstract

This dissertation presents a model-free, adaptive delay prediction and compensation framework for geographically distributed real-time power system co-simulation environments. Communication delays—both constant and real-time-varying—significantly degrade the accuracy, fidelity, and stability of co-simulated systems, particularly in dynamic and transient analyses of partitioned power systems. To address this, a predictor-based framework is developed that compensates for delays without requiring system models, computationally intensive signal transformations, or manual intervention.

The proposed solution leverages a Damping Impedance Method as the interface algorithm, combined with a sliding-mode control-inspired predictor system. Both single-parameter and multi-parameter predictor configurations are implemented, with the multi-parameter design providing an additional degree of freedom for delay compensation and extending the open-loop stability region. An adaptive, delay-aware tuning mechanism further enhances implementation by making predictor parameters dynamic functions of observed real-time delays. Validation was conducted through offline simulations and hardware-in-the-loop (HIL) experiments, incorporating typical three-phase AC systems, partitioned IEEE test systems (power system transmission-distribution co-simulation), and a real-time partitioned military-based 600 V DC microgrid.

Performance metrics such as state-tracking error, coupling error, subsystem frequency discrepancies, and a proposed residual energy metric were utilized to quantitatively assess fidelity and energy consistency across subsystems. Results show a significant improvement in operational stability and coupling fidelity with the implemented prediction systems over traditional approaches based on coupling signal transformation. Unlike conventional methods that rely on RMS, dynamic phasors, wave variable transformation, or synchronous reference frame transformations, the proposed model-free framework operates directly in the time domain, reducing complexity while improving robustness under time-varying delays.

Additionally, a Deep Deterministic Policy Gradient (DDPG) reinforcement learning agent is integrated into the co-simulation platform for real-time energy management across partitioned military-based 600 V DC microgrid with vehicle-to-grid (V2G) and vehicle-to-vehicle (V2V) capabilities. By formulating the problem as a Markov Decision Process, the distributed agent dynamically optimizes power generation, storage, and distribution, minimizing operational costs and adapting to load variations. Experimental results demonstrate that the DDPG agent effectively manages energy flows, maintains critical load supply, and enhances overall system resilience.

The developed co-simulation platform with adaptive delay prediction and compensation thus provides a scalable and practical solution for improving accuracy, stability, and energy management in distributed real-time power system simulations. Future research will focus on expanding the framework’s design flexibility and scaling to larger, multi-partitioned systems with simultaneous real-time simulator exchanges.

Author ORCID Identifier

https://orcid.org/0000-0002-0054-5778

Available for download on Sunday, May 31, 2026

Share

COinS