Design online strategies for adaptive tuning of existing optimal control solutions to assure (quantifiably) sufficient margins of resilience against cyber-physical adversarial events.

PI: Soumya Kundu, soumya.kundu@pnnl.gov

This project is developing online strategies for adaptive tuning or reformulating of existing optimal control solutions in response to cyber-physical adversarial events. The goal is to assure (quantifiably) sufficient margins of resilience to cyber-physical disruption while minimally compromising other objectives, such as efficiency.

In FY 2021, the project developed and tested algorithms to verify the resilient performance of a closed-loop system and adapt the controller for assured resilience. Given system objectives and constraints, the resilience verification algorithm quantifies the largest set of adversarial conditions that the system can tolerate without violating critical resilience specifications. This allows us to pre-compute the capability of the combined control and physical system ahead of time. The project then developed a robust optimization algorithm to assure sufficient reserves in power generation and non-critical load curtailment can be available to mitigate the impact of any adversarial event within the pre-computed set. The specific resilience measures used for the control design include surrogate measures for serviceability of critical loads and voltage safety (per ANSI C84.1 Standard). For resilience adaptation, the project developed control feedback that can tap into available reserves to maintain resilience under adversarial events. Unlike the state-of-the-art control tuning methods, the developed approach systematically determines the controller gains from the optimality conditions, thereby assuring that the adapted local controls are consistent with the system-level resilience control objectives. The ALERT control modules have been successfully implemented and demonstrated to mitigate a combined cyber and physical threat involving a loss of generator power and a replay attack that masks significant load change. 

In FY 2022, the project will focus on developing the full functionalities of the ALERT technology and further prove its efficacy for a broad set of adverse events. This will include (1) developing risk-based optimization algorithms that can effectively handle uncertain or probabilistic inputs and (2) integrating ALERT with a layered control architecture.

• Goal I: Complete design of optimal adaptive control gains with probabilistic resilience guarantees, especially in the context of high-impact-low-probability events.
• Goal II: Complete a working prototype of ALERT with demonstrable resilience performance tested on the RD2C Thrust 1 use case and identified threat scenarios.

View all RD2C projects