Protecting the power grid: strategies against distributed controller compromise

The electric power grid is a complex, interconnected cyber-physical system comprised of collaborating elements for monitoring and control. Distributed controllers play a prominent role in deploying this cohesive execution and are ubiquitous in the grid. As global information is shared and acted upon, faster response to system changes is achieved. However, failure or malfunction of a few or even one distributed controller in the entire system can cause cascading, detrimental effects. In the worst case, widespread blackouts can result, as exemplified by several historic cases. Furthermore, if controllers are maliciously compromised by an adversary, they can be manipulated to drive the power system to an unsafe state. Due to the shift from proprietary control protocols to popular, accessible network protocols and other modernization factors, the power system is extremely vulnerable to cyber attacks. Cyber attacks against the grid have increased significantly in recent years and can cause severe, physical consequences. Attack vectors for distributed controllers range from execution of malicious commands that can cause sensitive equipment damage to forced system topology changes creating instability. These vulnerabilities and risks need to be fully understood, and greater technical capabilities are necessary to create resilient and dynamic defenses. Proactive strategies must be developed to protect the power grid from distributed controller compromise or failure. This research investigates the role distributed controllers play in the grid and how their loss or compromise impacts the system. Specifically, an analytic method based on controllability analysis is derived using clustering and factorization techniques on controller sensitivities. In this manner, insight into the control support groups and sets of critical, essential, and redundant controllers for distributed controllers in the power system is achieved. Subsequently, we introduce proactive strategies that utilize these roles and grouping results for responding to controller compromise using the remaining set. These actions can be taken immediately to reduce system stress and mitigate compromise consequences as the compromise itself is investigated and eliminated by appropriate security mechanisms. These strategies are demonstrated with several compromise scenarios, and an overall framework is presented. Additionally, the controller role and group insights are applied to aid in developing an analytic corrective control selection for fast and automated remedial action scheme (RAS) design. Techniques to aid the verification of control commands and the detection of abnormal control action behavior are also presented. In particular, an augmented DC power flow algorithm using real-time measurements is developed that obtains both faster speed and higher accuracy than existing linear methods. For detecting abnormal behavior, a generator control action classification framework is presented that leverages known power system behaviors to enhance the use of data mining tools. Finally, the importance of incorporating power system knowledge into machine learning applications is emphasized with a study that improves power system neural network construction using modal analysis. This dissertation details these methodologies and their roles in realizing a more cohesive and resilient power system in the increasingly cyber-physical world