The full text of this item is not available at this time because the student has placed this item under an embargo for a period of time. The Libraries are not authorized to provide a copy of this work during the embargo period, even for Texas A&M users with NetID.
Addressing Uncertainty in Cyber-Physical Power Systems - Modeling to Integration in a Cyber-Physical Energy Management System
Abstract
Energy infrastructures are mission critical cyber-physical systems that are targets of persistent cyber attacks. While introducing new computing technologies and networks in power systems adds new capabilities for monitoring and control, dealing with the vast quantity of diverse devices with unknown trustworthiness and origin that can connect to the network and whose impact on operational reliability is also unknown is a frightening prospect. The threat landscape today is ex-tensive and constantly changing. Hence, to design resilient power systems it is inherent to develop cyber-physical models that can provide a platform to study the impact of cyber threats in a large scale interconnected grid as well as to propose a defense mechanism to operate them resiliently.
This work proposes a synthetic communication network model for synthetic electric grid with a novel contribution of designing optimized firewall model that follows NERC-CIP-005 standards. The proposed cyber model for the electric grid is validated through creation of a cyber physical testbed RESLab which integrates multiple simulators, emulators and hardware devices to implement threat models targeting critical operations. A multi-sensor multi-domain fusion methodology is proposed to integrate sensor data from physical, cyber and security emulators. Given the un-certainty and untrustworthiness of these sensors under the compromised state, the real-time data generated in the testbed is treated with a theory of uncertainty called Dempster Shafer Theory of Evidence, to improve the inferencing of intrusion for better detection. This approach improved the performance in comparison to the conventional supervised and semi-supervised learning techniques. But a major limitation of this approach is that it does not easily support utilization of the existing domain knowledge. Hence, a Bayesian Approach is undertaken for inferencing and learning the structure of a novel cyber attack model, called Bayesian Attack Graph. The outcome of this approach is considered for risk assessment in Cyber Physical Dynamic Situational Awareness, necessary for state-estimation as well as partially observable control problem.
Controlling the grid operations under uncertainty is another big challenge which is currently addressed with various data-driven approaches from the machine learning fraternity. In the work, we have developed environment for making uncertain MDP models, called Partially Observable MDPs, for the power system cases and solve the control problem using Bayesian Reinforcement Learning that follows the principles of Bayesian Inferencing. A challenge for the control problem is selection of an appropriate metric to optimize, as resilience is time, situation, and state dependent. Hence, an adaptive resilience metric quantification mechanism using Inverse Reinforcement Learning is proposed that not only learns a resilience metric but improves the performance of learning an optimal policy for resilient control.
The proposed algorithms for communication network modeling, inferencing under uncertainty, and integration with the testbed are validated by developing software applications and incorporating them with the CYPRES Energy Management System.
Subject
Cyber-Physical ModelingCyber-Physical Testbed
Reinforcement Learning
Data Fusion
Bayesian Inference and Learning
Dempster Shafer Theory
Volt-Var Control
Citation
Sahu, Abhijeet (2022). Addressing Uncertainty in Cyber-Physical Power Systems - Modeling to Integration in a Cyber-Physical Energy Management System. Doctoral dissertation, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /198621.