Document Type
Dissertation - Open Access
Award Date
2026
Degree Name
Doctor of Philosophy (PhD)
Department / School
Electrical Engineering and Computer Science
First Advisor
Junjian Qi
Abstract
Cascading failures in electric power systems, the uncontrolled, successive loss of system elements, can result in large-scale blackouts with significant economic and social impacts. To prevent cascading blackouts, analytical methods and mitigation strategies are important for understanding the failure propagation mechanism and reducing blackout risks. Previous research proposed cascading failure simulation models to generate cascade samples and address the scarcity of outage data. Real utility outage data has also been used in interaction model-based approaches to reveal how components interact during outage propagation. Also, by using reinforcement learning, online intervention methods have been introduced for blackout prevention. However, several challenges remain. On the supply side, increasing wind power penetration reduces system inertia and requires analysis of its impact on failure propagation. On the demand side, the fast expansion of large-scale data centers causes further cascading risks. Beyond simulation, existing interaction-based methods do not fully use spatial information in real outage data, and fail to accurately estimate failure consequences on networks with directed cycles. Moreover, existing methods cannot provide real-time intervention once a cascade starts. This dissertation addresses these challenges with four objectives: 1) We develop a cascading failure simulation that captures the impacts of high wind power penetration and large-scale data center loads on the IEEE RTS-96 system; 2) We use the interaction network to study spatial propagation from 14 years of real utility outage data and develop a mitigation strategy to suppress the spatial spread of cascading failures; 3) We estimate the expected consequences of line outages and load shedding accurately and efficiently on the adjusted directed cyclic interaction network; 4) We propose an interaction-model-based hierarchical reinforcement learning (IMHRL) framework that uses interactions to provide real-time intervention and reduce the risk of severe cascades. Results show that higher wind power penetration and data center loads both increase cascading failure risks. The proposed spatial mitigation outperforms methods that only consider the number of outages. The consequence estimation on the cyclic network is more accurate and about ten times faster than the tree-based approach. The IMHRL framework offers real-time intervention with significantly reduced cascade sizes and computational cost.
Publisher
South Dakota State University
Recommended Citation
Huang, Shuchen, "Interaction Models for Cascading Failure Analysis and Intervention" (2026). Electronic Theses and Dissertations. 2042.
https://openprairie.sdstate.edu/etd2/2042
Rights Statement
